{ "schema": "brianletort-forecast-ledger-v2", "asOf": "2026-09-06", "honestClaim": "I am testing whether a transparent, public-data forecasting process can add signal on a bounded set of AI-infrastructure questions. Every forecast is timestamped and scored. The track record is still early, and I am not claiming a demonstrated forecasting edge.", "stats": { "totalPredictions": 104, "resolved": 57, "pending": 47, "overdue": 0, "hits": 23, "partials": 15, "misses": 19, "hitRate": 0.5350877192982456, "brierScore": 0.1996403508771931, "bands": [ { "label": "Bold (<55%)", "min": 1, "max": 54, "resolved": 1, "hits": 1, "partials": 0, "misses": 0, "hitRate": 1, "meanConfidence": 0.43 }, { "label": "Core (55-80%)", "min": 55, "max": 80, "resolved": 55, "hits": 21, "partials": 15, "misses": 19, "hitRate": 0.5181818181818182, "meanConfidence": 0.6625454545454545 }, { "label": "High-conviction (>80%)", "min": 81, "max": 99, "resolved": 1, "hits": 1, "partials": 0, "misses": 0, "hitRate": 1, "meanConfidence": 0.84 } ] }, "scoreboard": { "brierHalf": { "point": 0.1996403508771931, "lo": 0.1637356140350877, "hi": 0.23682, "level": 0.9, "iterations": 2000 }, "brierMissPolicy": { "point": 0.31507894736842107, "lo": 0.27531315789473687, "hi": 0.35327640350877193, "level": 0.9, "iterations": 2000 }, "logLoss": { "half": 0.7284212487375203, "miss": 0.8360492081469956 }, "baselines": { "coinFlip": { "half": 0.18421052631578946, "miss": 0.25 }, "baseRate": { "half": 0.18957270408163251, "miss": 0.2493622448979591 }, "statedMean": { "half": 0.19897940904893802, "miss": 0.3072896583564169 } }, "skill": { "vsCoinFlip": { "half": -0.08376190476190537, "miss": -0.2603157894736843 }, "vsBaseRate": { "half": -0.05310704852965187, "miss": -0.26353910351325927 }, "vsStatedMean": { "half": -0.003321659419003087, "miss": -0.02534836041559707 } }, "reliability": { "bins": [ { "lo": 0.5, "hi": 0.6, "n": 6, "meanConfidence": 0.565, "hitRate": 0.25, "gap": 0.31499999999999995 }, { "lo": 0.6, "hi": 0.7, "n": 29, "meanConfidence": 0.6341379310344828, "hitRate": 0.5689655172413793, "gap": 0.06517241379310346 }, { "lo": 0.7, "hi": 0.8, "n": 17, "meanConfidence": 0.7211764705882352, "hitRate": 0.5588235294117647, "gap": 0.16235294117647048 }, { "lo": 0.8, "hi": 0.9, "n": 4, "meanConfidence": 0.81, "hitRate": 0.5, "gap": 0.31000000000000005 }, { "lo": 0.9, "hi": 1, "n": 0, "meanConfidence": null, "hitRate": null, "gap": null } ], "unbinned": 1, "expectedCalibrationError": 0.1389285714285714 }, "overconfidenceGap": { "point": 0.1264912280701751, "lo": 0.031394736842105204, "hi": 0.2196666666666666, "level": 0.9, "iterations": 2000 }, "byLens": [ { "label": "software", "n": 20, "hits": 8, "partials": 2, "misses": 10, "meanConfidence": 0.638, "hitRateHalf": 0.45, "brierHalf": 0.23979999999999996, "brierMissPolicy": 0.2828 }, { "label": "hardware", "n": 12, "hits": 5, "partials": 5, "misses": 2, "meanConfidence": 0.6666666666666669, "hitRateHalf": 0.625, "brierHalf": 0.12368333333333333, "brierMissPolicy": 0.29868333333333336 }, { "label": "networking", "n": 10, "hits": 1, "partials": 4, "misses": 5, "meanConfidence": 0.659, "hitRateHalf": 0.3, "brierHalf": 0.25585, "brierMissPolicy": 0.41585 }, { "label": "capital", "n": 11, "hits": 6, "partials": 4, "misses": 1, "meanConfidence": 0.6854545454545454, "hitRateHalf": 0.7272727272727273, "brierHalf": 0.1428, "brierMissPolicy": 0.3264363636363637 }, { "label": "power", "n": 4, "hits": 3, "partials": 0, "misses": 1, "meanConfidence": 0.7050000000000001, "hitRateHalf": 0.75, "brierHalf": 0.24250000000000002, "brierMissPolicy": 0.24250000000000002 } ], "byHorizon": [ { "label": "<=30d", "n": 14, "hits": 4, "partials": 6, "misses": 4, "meanConfidence": 0.6699999999999999, "hitRateHalf": 0.5, "brierHalf": 0.16937142857142856, "brierMissPolicy": 0.36794285714285724, "minDays": 0, "maxDays": 30 }, { "label": "31-90d", "n": 29, "hits": 15, "partials": 6, "misses": 8, "meanConfidence": 0.6579310344827586, "hitRateHalf": 0.6206896551724138, "brierHalf": 0.18433793103448276, "brierMissPolicy": 0.2670965517241379, "minDays": 31, "maxDays": 90 }, { "label": ">90d", "n": 14, "hits": 4, "partials": 3, "misses": 7, "meanConfidence": 0.6607142857142858, "hitRateHalf": 0.39285714285714285, "brierHalf": 0.2616071428571428, "brierMissPolicy": 0.36160714285714285, "minDays": 91, "maxDays": null } ], "errorClasses": { "total": 34, "classes": { "conjunctive": { "count": 9, "ids": [ "p29-vera-rubin-cadence", "p23-anthropic-round-close-final", "p14-anthropic-round-close", "p28-gemini-3-5-pro-june", "p59-tsmc-q2-capex-raise", "p53-skhy-debut-validates-memory", "p57-gemini-3-5-pro-ga-jul31", "p62-deepseek-v4-ga-jul31", "p22-nvidia-q2-fy27" ] }, "named-primary-disclosure-by-date": { "count": 2, "ids": [ "p12-voice-frontier-procurement", "p24-agent-platform-lock-in" ] }, "timing-slip": { "count": 13, "ids": [ "p1-2gw-customer-funded", "p32-gemini-3-5-pro-ga", "p37-gemini-3-5-pro-below-fable5", "p26-gemini-3-5-pro", "p17-samsung-walkout-shipments", "p38-fable5-access-restored", "p21-ebo-msa-spec", "p46-ferc-rto-compliance", "p35-cpo-design-win", "p43-open-weight-top5", "p44-hbm4-allocation-2027", "p51-cpo-partner-rack", "p3-capex-revise" ] }, "threshold-miss": { "count": 5, "ids": [ "p6-custom-silicon-35", "p7-bedrock-gpt55-anchor", "p18-claude-code-churn", "p10-fabric-second-confirm", "p25-fabric-vendor-q2" ] }, "other": { "count": 5, "ids": [ "p63-hbm-soldout-2027", "p4-interconnect-outpaces", "p64-colo-interconnect-outpaces", "p13-mrc-second-fabric", "p30-optics-design-wins" ] } } } }, "liveQuestions": [ { "id": "lq-p2", "weeklyPredictionId": "p2-f500-on-prem", "text": "At least one Fortune 500 enterprise discloses an on-prem AI workload greater than 50MW equivalent by September 30, 2026.", "lens": "software", "deadline": "2026-09-30", "resolutionSource": "Named-customer press release, 8-K, or equivalent primary disclosure.", "resolutionRule": "Hit if a Fortune 500 name and >50MW (or kW equivalent stated as such) appear in a primary disclosure dated on or before 2026-09-30.", "ambiguityPolicy": "resolves_against_forecaster", "stated": 0.6, "current": 0.2351677792746896, "direction": "down", "history": [ { "asOf": "2026-04-25", "p": 0.6, "source": "stated", "runId": null }, { "asOf": "2026-09-06", "p": 0.2351677792746896, "source": "engine", "runId": "2026-09-06T22-42-59Z-51a6cc" } ], "engine": { "id": "lq-p2", "weeklyPredictionId": "p2-f500-on-prem", "stated": 0.6, "referenceClass": { "p": 0.5, "className": "qtype:disclosure", "n": 29, "note": "most specific ledger class with n >= 8; Resolved predictions of type disclosure: disclosure = requires a named primary document or named company; threshold = numeric bar without a named-disclosure requirement; timing = event-by-date without a numeric bar; other = none of the above." }, "calibrated": { "p": 0.5344981305723325, "method": "hierarchical-beta-binomial-by-lens", "ci80": [ 0.4935410457069923, 0.5753334026686714 ], "variants": { "platt-on-stated-logit": 0.4994 } }, "model": { "p": 0.253435559756668, "method": "logistic-features-l2", "ci80": [ 0.12735552105567594, 0.5257197400777847 ], "features": { "conjuncts": 1, "hasNumericThreshold": true, "horizonDays": 158, "isMarketPriceQuestion": false, "isTimingQuestion": true, "requiresNamedPrimaryDisclosure": true, "statedConfidence": 0.6, "thresholdMagnitudeLog": 2.699, "weekIndex": 0 }, "gbm": 0.2194 }, "council": { "p": 0.11820982720092842, "aggregation": "log-odds-mean", "extremization": 1, "votes": [ { "persona": "grid-and-power-economist", "model": "openai/gpt-5.6-sol", "p": 0.09, "rationale": "The underlying buildout is plausible, but the resolution burden is unusually strict: one primary disclosure must name a Fortune 500 customer, characterize the workload as on-prem, and explicitly state more than 50 MW or its kW equivalent. Enterprises and vendors more often disclose GPU counts, investment, or total facility capacity, none of which safely establishes AI-workload power under the ambiguity rule. Fifty MW is also far above most enterprise-owned AI deployments. With only 24 days remaining and no qualifying disclosure indicated in the digest, I estimate well below the forecaster’s 0.6.", "keyEvidence": [], "whatWouldChangeMyMind": "A scheduled September launch or filed announcement preview explicitly linking a named Fortune 500 company’s on-prem AI system to more than 50 MW." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.23, "rationale": "The threshold is unusually large for enterprise on-premises deployment: 50MW implies a hyperscale-class AI installation, while Fortune 500 customers usually disclose accelerator counts, investment dollars, or facility capacity—not an AI workload’s electrical equivalent. Even a large data-center announcement may fail because capacity is not clearly an on-prem AI workload. The strict requirement for both the named customer and an explicit figure in a primary disclosure further lowers resolution odds. Only 24 days remain, and the supplied series provides no qualifying disclosure evidence. The software and disclosure base rates are materially less informative than these event-specific constraints.", "keyEvidence": [], "whatWouldChangeMyMind": "A Fortune 500 filing or company release explicitly identifying its own on-prem AI deployment and stating power consumption or equivalent capacity above 50MW." }, { "persona": "capital-markets-credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.08, "rationale": "The 0.60 forecast is too high. A >50MW on-prem AI workload is exceptionally large, and enterprises usually disclose compute scale through accelerator counts or spending—not power-equivalent workload. Named customers and site-level power figures are also frequently withheld. The digest shows extraordinary hyperscaler capex and RPO growth but no qualifying enterprise disclosure. With only 24 days remaining, normal 10-Q calendars offer little help; an unusually specific press release or 8-K would be required. The generic disclosure and software reference rates substantially overstate this narrowly worded event, while ambiguity resolves against qualification.", "keyEvidence": [ "https://www.sec.gov/edgar/search/", "https://fortune.com/ranking/fortune500/" ], "whatWouldChangeMyMind": "A scheduled enterprise AI-factory launch, investor event, or filed agreement explicitly linking a Fortune 500 customer to more than 50MW of on-prem AI capacity." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.11, "rationale": "I start from the software-lens base rate of 0.452; the disclosure-class rate is 0.500. I cut sharply because only 24 days remain and resolution requires several details in one timely primary disclosure: a Fortune 500 name, an explicitly on-prem AI workload, and capacity above 50MW. The digest’s large capex and RPO figures support AI infrastructure growth but establish none of those details. Its supplied filings also contain no qualifying disclosure. The forecaster’s 0.60 confidence band realized only 0.556, indicating overconfidence. Most of the original forecast window has elapsed, while hard-date disclosure timing and ambiguity resolving against the forecast add substantial failure risk.", "keyEvidence": [], "whatWouldChangeMyMind": "An official issuer filing or event agenda previewing a September disclosure that names a Fortune 500 enterprise and explicitly quantifies an on-prem AI workload above 50MW." }, { "persona": "geopolitics-regulation-analyst", "model": "openai/gpt-5.6-sol", "p": 0.12, "rationale": "Only 24 days remain, and the digest shows strong AI infrastructure spending but no qualifying named-customer disclosure. A >50MW on-prem workload is unusually large, while enterprises commonly omit site power, describe capacity indirectly, or announce plans rather than an actual workload. Capex, RPO, permitting, and model-release evidence cannot satisfy the strict primary-disclosure rule. Quarter-end announcements provide some residual chance, but ambiguity resolves against a hit. The April 0.6 forecast is therefore too high after most of the forecast window elapsed without cited resolution evidence.", "keyEvidence": [], "whatWouldChangeMyMind": "A scheduled enterprise AI-factory announcement, vendor briefing, or filing preview explicitly linking a Fortune 500 customer, on-prem deployment, and more than 50MW before September 30." } ], "dissent": null, "spread": 0.15 }, "final": { "p": 0.2351677792746896, "method": "stacked-fixed-weights (resolved n=57 < 80)", "weights": { "calibrated": 0.3, "council": 0.5, "model": 0.2 }, "ci80": [ 0.16605701690886854, 0.3793123335028279 ] }, "direction": "down", "drivers": [ "Reference class qtype:disclosure runs 0.50 (n=29) vs stated 0.60.", "Hierarchical calibration moves stated 0.60 to 0.53 (lens software, kappa_stated=24.0, kappa_lens=200.0).", "Feature model at 0.25: conjuncts=1, named disclosure=True, horizon=158d.", "Council of 5 at 0.12." ], "warnings": [ "all 5 votes came from one vendor (openai); extremization disabled, persona diversity only" ], "evidence": [ { "url": "https://www.sec.gov/edgar/search/", "title": "SEC.gov | EDGAR Full Text Search", "publishedAt": null, "source": "direct", "excerpt": "Cited by capital-markets-credit-analyst: The 0.60 forecast is too high. A >50MW on-prem AI workload is exceptionally large, and enterprises usually disclose compute scale through accelerator counts or spending—not power-equivalent workload. Named customers and site-level power figures are also frequ" }, { "url": "https://fortune.com/ranking/fortune500/", "title": "Fortune 500 – The largest companies in the U.S. by revenue | Fortune", "publishedAt": null, "source": "direct", "excerpt": "Cited by capital-markets-credit-analyst: The 0.60 forecast is too high. A >50MW on-prem AI workload is exceptionally large, and enterprises usually disclose compute scale through accelerator counts or spending—not power-equivalent workload. Named customers and site-level power figures are also frequ" } ], "bubbleBasket": null, "daysRemaining": 24, "deadline": "2026-09-30", "lens": "software" } }, { "id": "lq-p5", "weeklyPredictionId": "p5-neocloud-anchor-loss", "text": "At least one neocloud loses an anchor tenant or sees backlog growth turn negative by September 30, 2026.", "lens": "capital", "deadline": "2026-09-30", "resolutionSource": "SEC filing or earnings transcript of a publicly traded neocloud.", "resolutionRule": "Hit if a named publicly traded neocloud discloses tenant loss or sequential backlog decline in a filing dated on or before 2026-09-30.", "ambiguityPolicy": "resolves_against_forecaster", "bubbleBasket": "capex_durability", "stated": 0.35, "current": 0.14579444008096526, "direction": "down", "history": [ { "asOf": "2026-04-25", "p": 0.35, "source": "stated", "runId": null }, { "asOf": "2026-09-06", "p": 0.14579444008096526, "source": "engine", "runId": "2026-09-06T22-42-59Z-51a6cc" } ], "engine": { "id": "lq-p5", "weeklyPredictionId": "p5-neocloud-anchor-loss", "stated": 0.35, "referenceClass": { "p": 0.5, "className": "qtype:disclosure", "n": 29, "note": "most specific ledger class with n >= 8; Resolved predictions of type disclosure: disclosure = requires a named primary document or named company; threshold = numeric bar without a named-disclosure requirement; timing = event-by-date without a numeric bar; other = none of the above." }, "calibrated": { "p": 0.5251810377006346, "method": "hierarchical-beta-binomial-by-lens", "ci80": [ 0.4834172337279728, 0.5668525014203926 ], "variants": { "platt-on-stated-logit": 0.3581 } }, "model": { "p": 0.6766602510534848, "method": "logistic-features-l2", "ci80": [ 0.4279362449850583, 0.8758216030960221 ], "features": { "conjuncts": 1, "hasNumericThreshold": true, "horizonDays": 158, "isMarketPriceQuestion": false, "isTimingQuestion": true, "requiresNamedPrimaryDisclosure": true, "statedConfidence": 0.35, "thresholdMagnitudeLog": 0, "weekIndex": 0 }, "gbm": 0.7455 }, "council": { "p": 0.02, "aggregation": "extremized-log-odds-mean", "extremization": 1.5, "votes": [ { "persona": "grid-and-power-economist", "model": "openai/gpt-5.6-sol", "p": 0.06, "rationale": "Only 24 days remain, and the main public neoclouds’ regular June-quarter disclosure cycle has likely passed. CoreWeave’s latest digest shows $103.7B of RPO at June 30, up 244% year over year, making an already-existing sequential contraction less likely. A qualifying result now probably requires an unscheduled material-termination filing or an unusual interim backlog disclosure. Anchor-customer concentration creates some tail risk, but strong equity markets and tight broad high-yield spreads do not indicate acute near-term funding stress. The strict filing/transcript and deadline requirements push this well below the April forecast of 0.35.", "keyEvidence": [ "https://www.sec.gov/edgar/search/" ], "whatWouldChangeMyMind": "A scheduled September earnings release, disclosed contract dispute, amended material customer agreement, or filing showing RPO/backlog below the immediately preceding reported period." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.09, "rationale": "Only 24 days remain, and the major public neoclouds’ regular quarterly reporting windows have largely passed; their next earnings disclosures are generally after September 30. CoreWeave’s June RPO was $103.7 billion with 244% year-over-year growth, offering no visible stress signal, while Nebius’s available RPO figure is stale rather than evidence of sequential contraction. Credit conditions are mixed but not distressed enough to make an unscheduled customer-loss filing likely. Resolution requires an explicit filing or earnings-transcript disclosure, so rumors, customer concentration, slower growth, or post-deadline reporting do not qualify. The conditional probability is therefore far below the April forecast.", "keyEvidence": [ "https://www.sec.gov/edgar/search/" ], "whatWouldChangeMyMind": "A scheduled September earnings report from an eligible issuer, a material-contract termination 8-K, or evidence that an already-filed report explicitly showed sequential backlog decline." }, { "persona": "capital-markets-credit-analyst", "model": "gemini/gemini-3.5-flash", "p": 0.04, "rationale": "With only 24 days until the September 30, 2026 deadline, the window for a qualifying disclosure is extremely narrow. Q2 filings (through June 30) have already been submitted, as seen by CRWV's RPO of $103.7B. The next quarterly earnings cycle (Q3 ending September 30) will not report until October or November. Any qualifying disclosure before the deadline would require an unscheduled, material ad-hoc filing (such as an 8-K) announcing a major anchor tenant termination or a sequential backlog decline. Given the current strong secular demand for GPU capacity and the rarity of mid-quarter backlog revisions, such an filing is highly unlikely.", "keyEvidence": [], "whatWouldChangeMyMind": "An unexpected SEC filing (e.g., Form 8-K) by a publicly traded neocloud like CRWV or NBIS before September 30, 2026, disclosing a tenant loss or sequential backlog drop." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.08, "rationale": "I use the disclosure-question base rate of 0.50, then reduce sharply for the remaining 24-day window and hard filing deadline. The supplied digest shows no qualifying disclosure; CoreWeave’s June RPO was $103.7B and up 244% year over year, evidence against an imminent backlog reversal, though sequential decline is the required test. Broad high-yield spreads and VIX also show little systemic stress. Few qualifying issuers are likely to report again before September 30, leaving mainly unscheduled disclosures. Ambiguous definitions of “neocloud,” “anchor tenant,” or backlog resolve against a hit. The forecaster’s lone sub-0.50 calibration case is uninformative.", "keyEvidence": [], "whatWouldChangeMyMind": "A scheduled September earnings release, or a pre-deadline 8-K, 10-Q, or transcript explicitly reporting a named anchor-tenant departure or backlog below the prior quarter." }, { "persona": "geopolitics-regulation-analyst", "model": "gemini/gemini-3.5-flash", "p": 0.08, "rationale": "As of September 2026, major neoclouds like CoreWeave (CRWV) and Nebius (NBIS) exhibit immense backlog figures, with CRWV reporting RPO of $103.7B as of June 30, 2026 (up 244% YoY). There is no indication of sequential backlog decline or tenant loss in their Q2 filings. Because Q3 reporting periods end on September 30, any subsequent 10-Q filings will fall after the deadline. For this to resolve positive, a neocloud would have to file an unexpected material 8-K or registration amendment disclosing tenant loss or backlog decline within the next 24 days. Given strong demand and multi-year contract structures, this is highly improbable.", "keyEvidence": [], "whatWouldChangeMyMind": "An unexpected 8-K filing by CRWV or NBIS before September 30, 2026, disclosing the termination of a major customer contract or restating RPO downward." } ], "dissent": null, "spread": 0.05 }, "final": { "p": 0.14579444008096526, "method": "stacked-fixed-weights (resolved n=57 < 80)", "weights": { "calibrated": 0.3, "council": 0.5, "model": 0.2 }, "ci80": [ 0.10245334439561764, 0.3350521158767658 ] }, "direction": "down", "drivers": [ "Reference class qtype:disclosure runs 0.50 (n=29) vs stated 0.35.", "Hierarchical calibration moves stated 0.35 to 0.53 (lens capital, kappa_stated=24.0, kappa_lens=200.0).", "Feature model at 0.68: conjuncts=1, named disclosure=True, horizon=158d.", "Council of 5 at 0.02." ], "warnings": [], "evidence": [ { "url": "https://www.sec.gov/edgar/search/", "title": "SEC.gov | EDGAR Full Text Search", "publishedAt": null, "source": "direct", "excerpt": "Cited by grid-and-power-economist: Only 24 days remain, and the main public neoclouds’ regular June-quarter disclosure cycle has likely passed. CoreWeave’s latest digest shows $103.7B of RPO at June 30, up 244% year over year, making an already-existing sequential contraction less likely. A qualifyi" } ], "bubbleBasket": "capex_durability", "daysRemaining": 24, "deadline": "2026-09-30", "lens": "capital" } }, { "id": "lq-p9", "weeklyPredictionId": "p9-hbm4-second-source", "text": "Samsung HBM4 reaches greater than 25% share of NVIDIA Vera Rubin BOM by Q3 2026.", "lens": "hardware", "deadline": "2026-09-30", "resolutionSource": "Samsung or NVIDIA primary disclosure, or a named industry mix print that cites those filings.", "resolutionRule": "Hit only if a primary or named-mix source states Samsung HBM4 share of Vera Rubin BOM >25% for Q3 2026. Analyst inference without a cited mix is a miss.", "ambiguityPolicy": "resolves_against_forecaster", "stated": 0.6, "current": 0.129946189313545, "direction": "down", "history": [ { "asOf": "2026-05-01", "p": 0.6, "source": "stated", "runId": null }, { "asOf": "2026-09-06", "p": 0.129946189313545, "source": "engine", "runId": "2026-09-06T22-42-59Z-51a6cc" } ], "engine": { "id": "lq-p9", "weeklyPredictionId": "p9-hbm4-second-source", "stated": 0.6, "referenceClass": { "p": 0.5, "className": "qtype:disclosure", "n": 29, "note": "most specific ledger class with n >= 8; Resolved predictions of type disclosure: disclosure = requires a named primary document or named company; threshold = numeric bar without a named-disclosure requirement; timing = event-by-date without a numeric bar; other = none of the above." }, "calibrated": { "p": 0.5462607790663099, "method": "hierarchical-beta-binomial-by-lens", "ci80": [ 0.5046735756627343, 0.5876790652078271 ], "variants": { "platt-on-stated-logit": 0.4994 } }, "model": { "p": 0.486007866192302, "method": "logistic-features-l2", "ci80": [ 0.31592828653224747, 0.7626639306842565 ], "features": { "conjuncts": 1, "hasNumericThreshold": true, "horizonDays": 152, "isMarketPriceQuestion": false, "isTimingQuestion": true, "requiresNamedPrimaryDisclosure": true, "statedConfidence": 0.6, "thresholdMagnitudeLog": 1.398, "weekIndex": 1 }, "gbm": 0.2466 }, "council": { "p": 0.02, "aggregation": "extremized-log-odds-mean", "extremization": 1.5, "votes": [ { "persona": "grid-and-power-economist", "model": "openai/gpt-5.6-sol", "p": 0.07, "rationale": "The disclosure hurdle dominates the hardware question. NVIDIA generally does not publish supplier-level HBM allocation, and Samsung is unlikely to quantify its share of a customer’s BOM. The supplied digest includes NVIDIA’s quarter ending July 26 but no qualifying mix disclosure. Samsung’s Q3 reporting would ordinarily arrive after the September 30 deadline. A trade-source estimate might appear within 24 days, but the rule requires a named Q3 2026 mix above 25% tied to primary filings; generic qualification, supply, or expected-share reports fail. Ambiguity also resolves against a hit. The forecaster’s 0.6 substantially understates this documentation risk.", "keyEvidence": [], "whatWouldChangeMyMind": "A Samsung or NVIDIA release explicitly quantifying Samsung’s Q3 Rubin HBM4 allocation, or a reputable named supplier-mix report giving more than 25% and citing primary disclosures." }, { "persona": "semiconductor-supply-chain-analyst", "model": "gemini/gemini-3.5-flash", "p": 0.02, "rationale": "NVIDIA and Samsung do not disclose vendor-specific BOM shares in primary filings due to commercial confidentiality, making a citing industry mix print virtually impossible. Additionally, NVIDIA's Rubin (Vera Rubin) platform is not yet in high-volume production as of Q3 2026, with HBM4 volume shipments slated for late 2026 or 2027. Given that today is September 6, 2026, and the deadline is September 30, 2026, the strict requirement for primary disclosure of a >25% BOM share cannot be met under current industry timelines and reporting standards.", "keyEvidence": [], "whatWouldChangeMyMind": "An official SEC filing or joint press release from NVIDIA or Samsung explicitly stating a >25% Samsung HBM4 BOM share on Rubin by Q3 2026." }, { "persona": "capital-markets-credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.07, "rationale": "The economic claim may be plausible, but the resolution hurdle is much narrower. NVIDIA and Samsung rarely disclose supplier-specific BOM shares, especially an exact quarterly threshold, because sourcing allocations are commercially sensitive. A qualification announcement, supply agreement, or analyst estimate would not suffice. With only 24 days remaining, a qualifying named-mix publication must both state Samsung exceeds 25% for Q3 2026 and cite the required primary filings. The forecaster’s 0.6 materially underweights this disclosure risk and the resolves-against policy.", "keyEvidence": [], "whatWouldChangeMyMind": "A scheduled Samsung or NVIDIA investor disclosure addressing Rubin HBM4 sourcing, or a reputable named supply-chain mix report explicitly citing primary filings and assigning Samsung above 25% for Q3." }, { "persona": "superforecaster-statistician", "model": "gemini/gemini-3.5-flash", "p": 0.02, "rationale": "Using a base rate of 0.50 for disclosure-type questions, we adjust this significantly downward. The deadline is September 30, 2026, but official Q3 2026 reports from Samsung and NVIDIA will not be released until October and November 2026, respectively. Furthermore, semiconductor companies do not disclose proprietary Bill of Materials (BOM) vendor shares in primary SEC filings or earnings releases. Finally, NVIDIA's Rubin platform is scheduled for late 2026 or 2027, making Q3 2026 volume shipments of Rubin with HBM4 highly unlikely. Given these severe timing, confidentiality, and product roadmap hurdles, a qualifying disclosure by the deadline is virtually impossible.", "keyEvidence": [ "https://investor.nvidia.com", "https://www.samsung.com" ], "whatWouldChangeMyMind": "A primary disclosure or official filing by NVIDIA or Samsung before September 30, 2026, explicitly stating Samsung HBM4 comprised over 25% of NVIDIA's Q3 2026 Vera Rubin BOM." }, { "persona": "geopolitics-regulation-analyst", "model": "openai/gpt-5.6-sol", "p": 0.14, "rationale": "The underlying supply outcome is plausible, but the resolution hurdle is much stricter: an exact Samsung share above 25%, tied specifically to Vera Rubin and Q3 2026, must appear by September 30. NVIDIA and Samsung generally avoid disclosing supplier allocation percentages, while filings rarely contain product-level memory mix. Only 24 days remain, and the supplied public digest contains no qualifying mix disclosure. A trade report based on analyst inference would fail; “BOM share” also risks ambiguity versus unit or supply share, which resolves against the forecast. These documentary constraints outweigh the hardware reference-class base rate.", "keyEvidence": [ "https://investor.nvidia.com/" ], "whatWouldChangeMyMind": "A Samsung or NVIDIA release naming the Q3 Vera Rubin HBM4 allocation, or a clearly attributable industry mix table stating Samsung exceeds 25% under the resolution’s required sourcing." } ], "dissent": null, "spread": 0.12 }, "final": { "p": 0.129946189313545, "method": "stacked-fixed-weights (resolved n=57 < 80)", "weights": { "calibrated": 0.3, "council": 0.5, "model": 0.2 }, "ci80": [ 0.09611152331409827, 0.3617300205879319 ] }, "direction": "down", "drivers": [ "Reference class qtype:disclosure runs 0.50 (n=29) vs stated 0.60.", "Hierarchical calibration moves stated 0.60 to 0.55 (lens hardware, kappa_stated=24.0, kappa_lens=200.0).", "Feature model at 0.49: conjuncts=1, named disclosure=True, horizon=152d.", "Council of 5 at 0.02." ], "warnings": [ "1 cited URL(s) failed verification and were dropped." ], "evidence": [ { "url": "https://investor.nvidia.com", "title": "NVIDIA Corporation - Home", "publishedAt": null, "source": "direct", "excerpt": "Cited by superforecaster-statistician: Using a base rate of 0.50 for disclosure-type questions, we adjust this significantly downward. The deadline is September 30, 2026, but official Q3 2026 reports from Samsung and NVIDIA will not be released until October and November 2026, respectively. Furtherm" }, { "url": "https://www.samsung.com", "title": "Samsung US | Mobile | TV | Home Electronics | Home Appliances | Samsung US", "publishedAt": null, "source": "direct", "excerpt": "Cited by superforecaster-statistician: Using a base rate of 0.50 for disclosure-type questions, we adjust this significantly downward. The deadline is September 30, 2026, but official Q3 2026 reports from Samsung and NVIDIA will not be released until October and November 2026, respectively. Furtherm" }, { "url": "https://investor.nvidia.com/", "title": "NVIDIA Corporation - Home", "publishedAt": null, "source": "direct", "excerpt": "Cited by geopolitics-regulation-analyst: The underlying supply outcome is plausible, but the resolution hurdle is much stricter: an exact Samsung share above 25%, tied specifically to Vera Rubin and Q3 2026, must appear by September 30. NVIDIA and Samsung generally avoid disclosing supplier allocati" } ], "bubbleBasket": null, "daysRemaining": 24, "deadline": "2026-09-30", "lens": "hardware" } }, { "id": "lq-p11", "weeklyPredictionId": "p11-gated-cyber-revenue", "text": "By Q3 2026, at least two frontier labs publicly disclose a separately-priced gated cyber offering.", "lens": "software", "deadline": "2026-09-30", "resolutionSource": "Lab pricing page, blog, or SEC filing.", "resolutionRule": "Hit if two distinct frontier labs show a separately SKU'd gated-cyber product by 2026-09-30.", "ambiguityPolicy": "resolves_against_forecaster", "stated": 0.6, "current": 0.21188108224536822, "direction": "down", "history": [ { "asOf": "2026-05-01", "p": 0.6, "source": "stated", "runId": null }, { "asOf": "2026-09-06", "p": 0.21188108224536822, "source": "engine", "runId": "2026-09-06T22-42-59Z-51a6cc" } ], "engine": { "id": "lq-p11", "weeklyPredictionId": "p11-gated-cyber-revenue", "stated": 0.6, "referenceClass": { "p": 0.5, "className": "qtype:disclosure", "n": 29, "note": "most specific ledger class with n >= 8; Resolved predictions of type disclosure: disclosure = requires a named primary document or named company; threshold = numeric bar without a named-disclosure requirement; timing = event-by-date without a numeric bar; other = none of the above." }, "calibrated": { "p": 0.5344981305723325, "method": "hierarchical-beta-binomial-by-lens", "ci80": [ 0.4935410457069923, 0.5753334026686714 ], "variants": { "platt-on-stated-logit": 0.4994 } }, "model": { "p": 0.43299065468866715, "method": "logistic-features-l2", "ci80": [ 0.20989214980543802, 0.7339497625031709 ], "features": { "conjuncts": 1, "hasNumericThreshold": true, "horizonDays": 152, "isMarketPriceQuestion": true, "isTimingQuestion": true, "requiresNamedPrimaryDisclosure": true, "statedConfidence": 0.6, "thresholdMagnitudeLog": 0, "weekIndex": 1 }, "gbm": 0.6206 }, "council": { "p": 0.06898983468257704, "aggregation": "extremized-log-odds-mean", "extremization": 1.5, "votes": [ { "persona": "grid-and-power-economist", "model": "gemini/gemini-3.5-flash", "p": 0.05, "rationale": "As of September 2026, there is no public evidence of two distinct frontier labs (such as OpenAI, Anthropic, Google, or Meta) commercially offering separately-priced, gated cybersecurity-specific model SKUs. While safety frameworks (e.g., Anthropic's RSP) detail theoretical gating protocols for high-risk cyber capabilities, actual deployments remain internal or within standard enterprise API structures. Existing products like Google's Security AI Companion are standard enterprise defensive tools, not gated dual-use cyber offerings. With only 24 days remaining in the tracking period and the ambiguity policy resolving against the forecaster, a positive resolution is highly improbable.", "keyEvidence": [], "whatWouldChangeMyMind": "An official announcement or pricing page update from OpenAI or Anthropic before September 30, 2026, introducing a restricted-access, separately-billed cyber-capability SKU." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.36, "rationale": "The strict requirement is harder than general cyber-product availability: two frontier labs must each document a distinct paid SKU that is also gated. Labs often keep advanced cyber agents in previews or bundle security features into enterprise/API plans, neither of which clearly qualifies. With only 24 days left, two fresh qualifying disclosures are unlikely; the main path is that qualifying products have already appeared but are absent from the supplied digest. Ambiguity over whether cloud-security suites count as frontier-lab products further lowers resolution odds. This is below the original 0.6 and the software/disclosure base rates because of the two-lab and separate-pricing requirements.", "keyEvidence": [], "whatWouldChangeMyMind": "Official pricing or product pages from one qualifying lab would raise the probability substantially; pages from two labs explicitly showing distinct paid SKUs and access gating would nearly settle it." }, { "persona": "capital-markets-credit-analyst", "model": "gemini/gemini-3.5-flash", "p": 0.05, "rationale": "With only 24 days remaining until the September 30, 2026 deadline, there is no public evidence of two distinct frontier labs (such as OpenAI and Anthropic) offering separately priced, gated cybersecurity SKUs on their pricing pages or blogs. While cloud partners offer security-focused integrations (like Microsoft Copilot for Security), these do not qualify as separately SKU'd gated-cyber products from two distinct labs. Given the tight remaining window and the strict ambiguity policy, the probability of this resolving positively is very low.", "keyEvidence": [ "https://www.anthropic.com" ], "whatWouldChangeMyMind": "Public announcements or pricing page updates from OpenAI or Anthropic introducing distinct, gated cybersecurity model SKUs with separate pricing before the deadline." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.22, "rationale": "I use the lens:software base rate of 0.452, consistent with the >90-day rate of 0.400. I adjust downward because resolution requires two distinct labs, separate SKUs, gating, pricing, and qualifying public disclosures—not merely cyber features or restricted model access. With only 24 days left, the supplied digest provides no verifiable qualifying disclosure from even one lab, though it is not a comprehensive product-page search. Ambiguity resolves against the forecast. The forecaster’s 0.60–0.69 band has realized 0.556 versus mean stated 0.663, indicating overconfidence and reinforcing the reduction.", "keyEvidence": [], "whatWouldChangeMyMind": "Official pricing or product pages showing one qualifying SKU, plus a dated near-term launch commitment from a second frontier lab." }, { "persona": "geopolitics-regulation-analyst", "model": "openai/gpt-5.6-sol", "p": 0.28, "rationale": "The exact criterion is demanding: two distinct frontier labs must publish cyber-specific, gated products with separate SKUs or pricing, not merely safety controls, security features, pilots, or enterprise access. The supplied digest shows continued model activity but no qualifying commercial disclosures. With only 24 days left, two independent documented launches are unlikely unless at least one already exists outside the digest. The software base rate is 0.452, while the forecaster’s 0.60–0.69 calls have realized below stated confidence. Strict ambiguity treatment further lowers the estimate.", "keyEvidence": [], "whatWouldChangeMyMind": "A qualifying pricing page from one lab would raise the estimate materially; announcements from two labs identifying gated access and distinct pricing would make resolution highly likely." } ], "dissent": "semiconductor-supply-chain-analyst at 0.36 vs grid-and-power-economist at 0.05: The strict requirement is harder than general cyber-product availability: two frontier labs must each document a distinct paid SKU that is also gated. / As of September 2026, there is no public evidence of two distinct frontier labs (such as OpenAI, Anthropic, Google, or Meta) commercially offering separately...", "spread": 0.31 }, "final": { "p": 0.21188108224536822, "method": "stacked-fixed-weights (resolved n=57 < 80)", "weights": { "calibrated": 0.3, "council": 0.5, "model": 0.2 }, "ci80": [ 0.14866809595424627, 0.5015906469228318 ] }, "direction": "down", "drivers": [ "Reference class qtype:disclosure runs 0.50 (n=29) vs stated 0.60.", "Hierarchical calibration moves stated 0.60 to 0.53 (lens software, kappa_stated=24.0, kappa_lens=200.0).", "Feature model at 0.43: conjuncts=1, named disclosure=True, horizon=152d.", "Council of 5 at 0.07 with dissent." ], "warnings": [ "1 cited URL(s) failed verification and were dropped." ], "evidence": [ { "url": "https://www.anthropic.com", "title": "Home \\ Anthropic", "publishedAt": null, "source": "direct", "excerpt": "Cited by capital-markets-credit-analyst: With only 24 days remaining until the September 30, 2026 deadline, there is no public evidence of two distinct frontier labs (such as OpenAI and Anthropic) offering separately priced, gated cybersecurity SKUs on their pricing pages or blogs. While cloud partn" } ], "bubbleBasket": null, "daysRemaining": 24, "deadline": "2026-09-30", "lens": "software" } }, { "id": "lq-p15", "weeklyPredictionId": "p15-amd-second-frontier-train", "text": "A second public open-weights frontier-class model trained end-to-end on AMD silicon by September 30, 2026.", "lens": "hardware", "deadline": "2026-09-30", "resolutionSource": "Model card, lab blog, or AMD primary disclosure naming the training cluster.", "resolutionRule": "Hit if a second named open-weights model (not the first already-known AMD-trained model) states end-to-end AMD training.", "ambiguityPolicy": "resolves_against_forecaster", "stated": 0.55, "current": 0.36194808031625814, "direction": "down", "history": [ { "asOf": "2026-05-08", "p": 0.55, "source": "stated", "runId": null }, { "asOf": "2026-09-06", "p": 0.36194808031625814, "source": "engine", "runId": "2026-09-06T22-42-59Z-51a6cc" } ], "engine": { "id": "lq-p15", "weeklyPredictionId": "p15-amd-second-frontier-train", "stated": 0.55, "referenceClass": { "p": 0.5, "className": "qtype:disclosure", "n": 29, "note": "most specific ledger class with n >= 8; Resolved predictions of type disclosure: disclosure = requires a named primary document or named company; threshold = numeric bar without a named-disclosure requirement; timing = event-by-date without a numeric bar; other = none of the above." }, "calibrated": { "p": 0.541176033303598, "method": "hierarchical-beta-binomial-by-lens", "ci80": [ 0.49956071785364115, 0.582640997905298 ], "variants": { "platt-on-stated-logit": 0.4704 } }, "model": { "p": 0.5714438101455206, "method": "logistic-features-l2", "ci80": [ 0.3653679675686774, 0.809134825031616 ], "features": { "conjuncts": 1, "hasNumericThreshold": false, "horizonDays": 145, "isMarketPriceQuestion": false, "isTimingQuestion": true, "requiresNamedPrimaryDisclosure": true, "statedConfidence": 0.55, "thresholdMagnitudeLog": 0, "weekIndex": 2 }, "gbm": 0.6558 }, "council": { "p": 0.20620044071373655, "aggregation": "log-odds-mean", "extremization": 1, "votes": [ { "persona": "grid-and-power-economist", "model": "openai/gpt-5.6-sol", "p": 0.22, "rationale": "Only 24 days remain, and the digest provides no model card, lab post, or AMD disclosure identifying a second qualifying model. AMD has strong incentives to publicize frontier training wins, and one known success establishes technical feasibility, but the rule requires a named open-weights frontier-class model plus explicit end-to-end AMD training. That exact disclosure is substantially harder than merely releasing a model using AMD for inference or part of training. The forecaster’s comparable 0.50–0.59 calls have realized poorly, and disclosure/horizon base rates do not overcome the short remaining window and ambiguity-against policy.", "keyEvidence": [ "https://huggingface.co/models" ], "whatWouldChangeMyMind": "A scheduled September model launch, an AMD event agenda naming a training partner, or a pre-release model card explicitly describing an all-AMD training cluster." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.22, "rationale": "Only 24 days remain, and the supplied digest contains no indication of an imminent qualifying model disclosure. The rule is stricter than merely using AMD accelerators: a second named, frontier-class, open-weights model must explicitly document end-to-end AMD training in a primary source. Labs often disclose serving hardware while omitting training provenance, and “frontier-class” ambiguity resolves against a hit. This is materially below the May forecast of 0.55; the forecaster’s 0.50–0.59 band has also realized only 0.367 so far.", "keyEvidence": [], "whatWouldChangeMyMind": "A model teaser, repository, or AMD announcement naming an MI300/MI350 training cluster and an open-weights release scheduled before September 30 would raise the estimate substantially." }, { "persona": "capital-markets-credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.23, "rationale": "Only 24 days remain, and the supplied public-series digest contains no evidence of an imminent qualifying release. Resolution requires a second named open-weights, frontier-class model plus explicit end-to-end AMD training and identification of the training cluster. Model launches can occur with little notice, but this unusually specific training disclosure is not standard, and any ambiguity about mixed silicon, fine-tuning only, openness, or frontier status causes a miss. The forecaster’s 0.50–0.59 band has also realized materially below its stated confidence, supporting a downward revision from 0.55.", "keyEvidence": [], "whatWouldChangeMyMind": "A dated launch announcement or model card preview explicitly describing full pretraining on a named AMD-only cluster, with weights scheduled for release by September 30." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.15, "rationale": "I use the >90-day reference-class base rate of 0.40, tempered by the 0.50 disclosure base rate. With only 24 days left, no specific candidate model or qualifying primary disclosure is provided. The rule additionally requires a second named model, open weights, frontier-class status, end-to-end AMD training, and sufficiently explicit cluster attribution; ambiguity counts as failure. Disclosure timing is narrower than underlying technical completion. The forecaster’s 0.50–0.59 calibration band realized only 0.367, indicating material overconfidence relative to the original 0.55.", "keyEvidence": [], "whatWouldChangeMyMind": "A credible lab announcement scheduling an open-weights frontier model release before September 30 and explicitly identifying end-to-end training on a named AMD cluster." }, { "persona": "geopolitics-regulation-analyst", "model": "openai/gpt-5.6-sol", "p": 0.22, "rationale": "Only 24 days remain, and the supplied digest contains no qualifying model-card, lab, or AMD disclosure. The first known AMD-trained model establishes technical feasibility, but resolution requires another named frontier-class open-weights model plus an explicit end-to-end AMD-training statement and identified cluster. Such hardware provenance is often omitted or described ambiguously, which fails under the stated policy. Broad hardware and disclosure base rates are materially weakened by the short remaining window and strict documentation requirements.", "keyEvidence": [], "whatWouldChangeMyMind": "A lab teaser naming an AMD training cluster and an imminent open-weights release, or an AMD event agenda promising a second end-to-end trained model." } ], "dissent": null, "spread": 0.08 }, "final": { "p": 0.36194808031625814, "method": "stacked-fixed-weights (resolved n=57 < 80)", "weights": { "calibrated": 0.3, "council": 0.5, "model": 0.2 }, "ci80": [ 0.27323801962536437, 0.4464094842491682 ] }, "direction": "down", "drivers": [ "Reference class qtype:disclosure runs 0.50 (n=29) vs stated 0.55.", "Hierarchical calibration moves stated 0.55 to 0.54 (lens hardware, kappa_stated=24.0, kappa_lens=200.0).", "Feature model at 0.57: conjuncts=1, named disclosure=True, horizon=145d.", "Council of 5 at 0.21." ], "warnings": [ "all 5 votes came from one vendor (openai); extremization disabled, persona diversity only", "1 cited URL(s) failed verification and were dropped." ], "evidence": [ { "url": "https://huggingface.co/models", "title": "Models – Hugging Face", "publishedAt": null, "source": "direct", "excerpt": "Cited by grid-and-power-economist: Only 24 days remain, and the digest provides no model card, lab post, or AMD disclosure identifying a second qualifying model. AMD has strong incentives to publicize frontier training wins, and one known success establishes technical feasibility, but the rule requi" } ], "bubbleBasket": null, "daysRemaining": 24, "deadline": "2026-09-30", "lens": "hardware" } }, { "id": "lq-p19", "weeklyPredictionId": "p19-specialist-multimodal-procurement", "text": "At least one Fortune 500 enterprise discloses a production specialist-multimodal deployment >$25M annualized by September 30, 2026.", "lens": "software", "deadline": "2026-09-30", "resolutionSource": "Named-customer disclosure.", "resolutionRule": "Hit requires Fortune 500 name, production (not pilot) language, and a dollar or capacity figure meeting the trigger.", "ambiguityPolicy": "resolves_against_forecaster", "stated": 0.6, "current": 0.25450513198775904, "direction": "down", "history": [ { "asOf": "2026-05-17", "p": 0.6, "source": "stated", "runId": null }, { "asOf": "2026-09-06", "p": 0.25450513198775904, "source": "engine", "runId": "2026-09-06T22-42-59Z-51a6cc" } ], "engine": { "id": "lq-p19", "weeklyPredictionId": "p19-specialist-multimodal-procurement", "stated": 0.6, "referenceClass": { "p": 0.5, "className": "qtype:disclosure", "n": 29, "note": "most specific ledger class with n >= 8; Resolved predictions of type disclosure: disclosure = requires a named primary document or named company; threshold = numeric bar without a named-disclosure requirement; timing = event-by-date without a numeric bar; other = none of the above." }, "calibrated": { "p": 0.5344981305723325, "method": "hierarchical-beta-binomial-by-lens", "ci80": [ 0.4935410457069923, 0.5753334026686714 ], "variants": { "platt-on-stated-logit": 0.4994 } }, "model": { "p": 0.17725830349823898, "method": "logistic-features-l2", "ci80": [ 0.08186247463543082, 0.3788842973456124 ], "features": { "conjuncts": 1, "hasNumericThreshold": true, "horizonDays": 136, "isMarketPriceQuestion": false, "isTimingQuestion": true, "requiresNamedPrimaryDisclosure": true, "statedConfidence": 0.6, "thresholdMagnitudeLog": 7.398, "weekIndex": 3 }, "gbm": 0.2194 }, "council": { "p": 0.1654300580396918, "aggregation": "log-odds-mean", "extremization": 1, "votes": [ { "persona": "grid-and-power-economist", "model": "openai/gpt-5.6-sol", "p": 0.18, "rationale": "Only 24 days remain, and the supplied digest contains no qualifying named-customer disclosure. The rule is unusually strict: a Fortune 500 customer, explicit production status, specialist multimodality, and a disclosed dollar or capacity figure proving more than $25M annualized. Enterprise AI deployments are increasingly common, but customers and vendors rarely publish contract economics, while announcements often describe pilots or omit measurable scale. September also offers fewer routine earnings disclosures than quarter-end reporting months. The software and disclosure base rates are materially higher, but they do not reflect this late-stage conditional window or the conjunction of documentary requirements. Ambiguity resolving against the forecaster further lowers the chance of a marginal announcement qualifying.", "keyEvidence": [], "whatWouldChangeMyMind": "A scheduled September vendor event with a pre-announced Fortune 500 customer case study, or a filed customer contract quantifying production multimodal usage above the threshold." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.16, "rationale": "Only 24 days remain, and the digest contains no qualifying named-customer disclosure. Rapid model releases and hyperscaler commitments support growing multimodal adoption, but they do not satisfy the strict resolution test. Enterprise announcements commonly describe pilots, production workflows, users, or broad vendor commitments while withholding annualized deployment value. Requiring a Fortune 500 name, explicit production status, specialist-multimodal scope, and a figure above $25M creates substantial conjunction and ambiguity risk. The software-ledger base rate of 0.452 is therefore less relevant after most of the forecast window has elapsed without a documented hit. This is materially below the May forecast of 0.6.", "keyEvidence": [], "whatWouldChangeMyMind": "A scheduled September vendor event or filing preview identifying a Fortune 500 customer, production multimodal workload, and contract value or clearly qualifying capacity would raise the estimate sharply." }, { "persona": "capital-markets-credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.18, "rationale": "The underlying adoption trend is strong, but the disclosure hurdle is unusually strict: a named Fortune 500 customer, explicit production status, specialist multimodality, and a dollar or capacity figure clearly exceeding $25M annualized. Enterprises and vendors rarely publish deployment economics at that granularity. Only 24 days remain, and most calendar-Q3 10-Qs will arrive after September 30. Rapid hyperscaler capex and RPO growth support large deployments generally but do not establish this specific customer-level trigger. Ambiguity resolving against the forecast further lowers the odds versus the software-disclosure base rate.", "keyEvidence": [], "whatWouldChangeMyMind": "A scheduled September customer event, earnings release, or official case study preview indicating a named Fortune 500 production multimodal deployment with quantified spend or dedicated capacity." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.14, "rationale": "Base rate: 0.452 for software questions, with disclosure questions at 0.500 and original >90-day questions at 0.400. I reduce sharply because only 24 days remain and the digest provides no qualifying named-customer evidence. Resolution requires one disclosure to identify a Fortune 500 customer, state production rather than pilot status, establish specialist multimodality, and quantify over $25M annualized or qualifying capacity. Such customer announcements commonly omit spending and use ambiguous partnership language, which resolves as failure. Broad “at least one” scope and strong AI investment help, but model cadence and vendor capex do not verify customer-level economics. The forecaster’s 0.60–0.69 band realized 0.556, supporting a modest overconfidence correction.", "keyEvidence": [], "whatWouldChangeMyMind": "A primary filing or company release before September 30 naming the Fortune 500 customer and explicitly documenting production status, specialist multimodality, and annualized value or qualifying capacity above the threshold." }, { "persona": "geopolitics-regulation-analyst", "model": "openai/gpt-5.6-sol", "p": 0.17, "rationale": "The broad software and disclosure base rates near 0.45–0.50 materially overstate this unusually strict event. Resolution simultaneously requires a Fortune 500 name, explicit production status, specialist multimodality, and a disclosed dollar or capacity figure establishing more than $25M annualized. Enterprise AI growth and rapid model releases support the underlying deployment trend, but companies rarely publish customer-level economics. Only 24 days remain, with limited major reporting or conference catalysts before quarter-end. The supplied digest contains no qualifying named-customer evidence. Ambiguity resolving against the forecast further lowers the chance below the forecaster’s 0.6.", "keyEvidence": [], "whatWouldChangeMyMind": "A scheduled September customer event, filing, or vendor announcement previewing a named Fortune 500 production deployment with contract value, annual spend, or clearly convertible capacity above the threshold." } ], "dissent": null, "spread": 0.04 }, "final": { "p": 0.25450513198775904, "method": "stacked-fixed-weights (resolved n=57 < 80)", "weights": { "calibrated": 0.3, "council": 0.5, "model": 0.2 }, "ci80": [ 0.19799747830568384, 0.3173585718417584 ] }, "direction": "down", "drivers": [ "Reference class qtype:disclosure runs 0.50 (n=29) vs stated 0.60.", "Hierarchical calibration moves stated 0.60 to 0.53 (lens software, kappa_stated=24.0, kappa_lens=200.0).", "Feature model at 0.18: conjuncts=1, named disclosure=True, horizon=136d.", "Council of 5 at 0.17." ], "warnings": [ "all 5 votes came from one vendor (openai); extremization disabled, persona diversity only" ], "evidence": [], "bubbleBasket": null, "daysRemaining": 24, "deadline": "2026-09-30", "lens": "software" } }, { "id": "lq-p33", "weeklyPredictionId": "p33-vera-rubin-first-shipments", "text": "At least one major OEM announces customer shipment or formal order availability of Vera Rubin systems by September 30, 2026.", "lens": "hardware", "deadline": "2026-09-30", "resolutionSource": "OEM press release or NVIDIA partner disclosure.", "resolutionRule": "Hit on shipment or order-open language from a major OEM. Architecture announcements without orderability miss.", "ambiguityPolicy": "resolves_against_forecaster", "stated": 0.7, "current": 0.5245266053042982, "direction": "down", "history": [ { "asOf": "2026-06-06", "p": 0.7, "source": "stated", "runId": null }, { "asOf": "2026-09-06", "p": 0.5245266053042982, "source": "engine", "runId": "2026-09-06T22-42-59Z-51a6cc" } ], "engine": { "id": "lq-p33", "weeklyPredictionId": "p33-vera-rubin-first-shipments", "stated": 0.7, "referenceClass": { "p": 0.5, "className": "qtype:disclosure", "n": 29, "note": "most specific ledger class with n >= 8; Resolved predictions of type disclosure: disclosure = requires a named primary document or named company; threshold = numeric bar without a named-disclosure requirement; timing = event-by-date without a numeric bar; other = none of the above." }, "calibrated": { "p": 0.5564302705917336, "method": "hierarchical-beta-binomial-by-lens", "ci80": [ 0.5149124721063718, 0.5977420191096907 ], "variants": { "platt-on-stated-logit": 0.5617 } }, "model": { "p": 0.6027467607937605, "method": "logistic-features-l2", "ci80": [ 0.4484301750244226, 0.8223259810175377 ], "features": { "conjuncts": 1, "hasNumericThreshold": true, "horizonDays": 116, "isMarketPriceQuestion": false, "isTimingQuestion": true, "requiresNamedPrimaryDisclosure": true, "statedConfidence": 0.7, "thresholdMagnitudeLog": 0, "weekIndex": 6 }, "gbm": 0.7293 }, "council": { "p": 0.473422340285086, "aggregation": "log-odds-mean", "extremization": 1, "votes": [ { "persona": "grid-and-power-economist", "model": "openai/gpt-5.6-sol", "p": 0.61, "rationale": "Rubin was positioned for second-half 2026 deployment, and several major server OEMs have strong incentives to open orders before volume shipment. The disjunction across Dell, HPE, Lenovo, and Supermicro makes one qualifying disclosure plausible, while “formal order availability” is easier to satisfy than actual customer shipment. However, only 24 days remain, hardware schedules commonly slip, and partner announcements may describe architecture or future systems without explicitly opening orders. The strict documentary and ambiguity rules therefore keep the estimate below the forecaster’s June probability.", "keyEvidence": [], "whatWouldChangeMyMind": "A dated OEM product page or NVIDIA partner disclosure explicitly saying Rubin systems are orderable would make this nearly certain; revised NVIDIA guidance pushing partner availability beyond September would sharply lower it." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.36, "rationale": "Rubin’s broad second-half-2026 roadmap leaves some chance of a quarter-end launch, especially across multiple OEMs. However, only 24 days remain, and the supplied evidence contains no product-specific indication that a qualifying announcement has occurred. HBM4 qualification, CoWoS-L availability, and NVL144 rack integration make early shipments vulnerable to slippage. OEMs also commonly announce architectures or future availability before systems are actually orderable; those formulations miss under the strict rule. Strong semiconductor and server backlogs support demand but do not establish Rubin readiness.", "keyEvidence": [], "whatWouldChangeMyMind": "An OEM launch event, configurator listing paired with a press release, or NVIDIA partner disclosure explicitly saying Rubin systems are shipping or available to order before September 30." }, { "persona": "capital-markets-credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.48, "rationale": "Rubin’s stated 2026 ramp makes a September OEM milestone plausible, and multiple major server vendors create several chances for qualifying language. However, the rule requires shipment or orders formally open—not roadmap, sampling, qualification, or future-availability language. OEM launches often cluster around industry events, while new rack-scale platforms can slip into the fourth quarter as networking, cooling, and system validation finish. With only 24 days remaining and no qualifying disclosure in the supplied evidence, I put this below the forecaster’s 0.7 and near even odds.", "keyEvidence": [ "https://nvidianews.nvidia.com/" ], "whatWouldChangeMyMind": "An OEM product page or partner release stating “shipping now,” “available to order,” or a specific already-open ordering date would raise this sharply; qualification-only updates or Q4 timing would lower it." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.51, "rationale": "I use a 0.53 base rate, centered on the overall ledger and supported by disclosure (0.50), hardware (0.615), and >90-day horizon (0.40) classes. Multiple eligible OEMs and strong Dell/SMCI backlog indicators raise the chance that one opens orders. However, only 24 days remain, the digest provides no Vera Rubin-specific orderability evidence, and architecture or roadmap announcements do not qualify. The hard disclosure deadline and ambiguity policy materially increase timing risk. The forecaster’s >=0.70 calls realized only 0.545 overall, supporting a sizable reduction from 0.70 despite better calibration in the narrower 0.7–0.8 bin.", "keyEvidence": [], "whatWouldChangeMyMind": "A dated OEM or NVIDIA partner release explicitly saying Vera Rubin systems are shipping or available to order would settle yes; an official delay beyond September would sharply lower the probability." }, { "persona": "geopolitics-regulation-analyst", "model": "openai/gpt-5.6-sol", "p": 0.41, "rationale": "Rubin’s stated 2026 ramp and participation by several major server OEMs make eventual orderability likely, but the strict September 30 cutoff leaves only 24 days. The easier qualifying path is formal order-open language rather than actual shipment. Still, major OEM launch calendars tend to cluster around spring events or October/November OCP and SC announcements, limiting remaining catalysts. Strong NVIDIA, Dell, and SMCI demand indicators support commercialization but do not establish Rubin orderability. With ambiguity resolved against the forecast, roadmap, sampling, qualification, or future-availability language would all miss.", "keyEvidence": [ "https://nvidianews.nvidia.com/", "https://www.sec.gov/edgar/search/" ], "whatWouldChangeMyMind": "A dated OEM product page, channel notice, or NVIDIA partner announcement explicitly saying Rubin systems are shipping or available to order before September 30." } ], "dissent": null, "spread": 0.25 }, "final": { "p": 0.5245266053042982, "method": "stacked-fixed-weights (resolved n=57 < 80)", "weights": { "calibrated": 0.3, "council": 0.5, "model": 0.2 }, "ci80": [ 0.42282554745785045, 0.6567654840340856 ] }, "direction": "down", "drivers": [ "Reference class qtype:disclosure runs 0.50 (n=29) vs stated 0.70.", "Hierarchical calibration moves stated 0.70 to 0.56 (lens hardware, kappa_stated=24.0, kappa_lens=200.0).", "Feature model at 0.60: conjuncts=1, named disclosure=True, horizon=116d.", "Council of 5 at 0.47." ], "warnings": [ "all 5 votes came from one vendor (openai); extremization disabled, persona diversity only" ], "evidence": [ { "url": "https://nvidianews.nvidia.com/", "title": "Home | NVIDIA Newsroom", "publishedAt": null, "source": "direct", "excerpt": "Cited by capital-markets-credit-analyst: Rubin’s stated 2026 ramp makes a September OEM milestone plausible, and multiple major server vendors create several chances for qualifying language. However, the rule requires shipment or orders formally open—not roadmap, sampling, qualification, or future-a" }, { "url": "https://www.sec.gov/edgar/search/", "title": "SEC.gov | EDGAR Full Text Search", "publishedAt": null, "source": "direct", "excerpt": "Cited by geopolitics-regulation-analyst: Rubin’s stated 2026 ramp and participation by several major server OEMs make eventual orderability likely, but the strict September 30 cutoff leaves only 24 days. The easier qualifying path is formal order-open language rather than actual shipment. Still, maj" } ], "bubbleBasket": null, "daysRemaining": 24, "deadline": "2026-09-30", "lens": "hardware" } }, { "id": "lq-p36", "weeklyPredictionId": "p36-power-first-followthrough", "text": "A hyperscaler announces another >500MW power-first AI campus or behind-the-meter generation block by September 30, 2026.", "lens": "power", "deadline": "2026-09-30", "resolutionSource": "Hyperscaler IR or utility filing.", "resolutionRule": "Hit if a hyperscaler primary source states >500MW and power-first or behind-the-meter generation. Land-bank announcements without power miss.", "ambiguityPolicy": "resolves_against_forecaster", "bubbleBasket": "energization", "stated": 0.6, "current": 0.34749810125412955, "direction": "down", "history": [ { "asOf": "2026-06-06", "p": 0.6, "source": "stated", "runId": null }, { "asOf": "2026-09-06", "p": 0.34749810125412955, "source": "engine", "runId": "2026-09-06T22-42-59Z-51a6cc" } ], "engine": { "id": "lq-p36", "weeklyPredictionId": "p36-power-first-followthrough", "stated": 0.6, "referenceClass": { "p": 0.5, "className": "qtype:disclosure", "n": 29, "note": "most specific ledger class with n >= 8; Resolved predictions of type disclosure: disclosure = requires a named primary document or named company; threshold = numeric bar without a named-disclosure requirement; timing = event-by-date without a numeric bar; other = none of the above." }, "calibrated": { "p": 0.5456909818405663, "method": "hierarchical-beta-binomial-by-lens", "ci80": [ 0.5033765968102275, 0.5878326393900954 ], "variants": { "platt-on-stated-logit": 0.4994 } }, "model": { "p": 0.5467815977023598, "method": "logistic-features-l2", "ci80": [ 0.24703411773744527, 0.8945852318734624 ], "features": { "conjuncts": 1, "hasNumericThreshold": true, "horizonDays": 116, "isMarketPriceQuestion": false, "isTimingQuestion": true, "requiresNamedPrimaryDisclosure": true, "statedConfidence": 0.6, "thresholdMagnitudeLog": 2.699, "weekIndex": 6 }, "gbm": 0.2466 }, "council": { "p": 0.19074988651814342, "aggregation": "log-odds-mean", "extremization": 1, "votes": [ { "persona": "grid-and-power-economist", "model": "openai/gpt-5.6-sol", "p": 0.18, "rationale": "Only 24 days remain, and the qualifying bar is unusually strict: a primary hyperscaler source or utility filing must explicitly connect more than 500 MW to power-first development or behind-the-meter generation. Large campus, land, or load announcements without that linkage fail. Quarter-end can concentrate disclosures, and strong generator/equipment backlogs support continued deal activity, but September is generally outside hyperscaler earnings season. Long interconnection timelines also favor exploratory announcements while making specific, documentable power commitments scarce. The small power-lens reference class is outweighed by the short remaining window and ambiguity resolving against the forecast.", "keyEvidence": [], "whatWouldChangeMyMind": "A scheduled hyperscaler infrastructure event, or a pending utility docket identifying a hyperscaler and an explicit >500 MW dedicated or behind-the-meter supply block, would raise the probability materially." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.21, "rationale": "Only 24 days remain, sharply reducing the opportunity relative to the original forecast window. Strong utility backlogs, generation capex, and equipment orders support continued hyperscaler power procurement, but they do not imply an imminent qualifying disclosure. The rule is narrow: a hyperscaler primary source or utility filing must explicitly state more than 500MW and characterize it as power-first or behind-the-meter generation. Quarter-end can prompt announcements or filings, but ambiguous campus capacity, future land, or aggregated power plans would miss. The forecaster’s 0.60 confidence band has also historically overpredicted outcomes.", "keyEvidence": [ "https://www.sec.gov/edgar/search/" ], "whatWouldChangeMyMind": "A scheduled hyperscaler infrastructure announcement or an identifiable utility docket describing a dedicated greater-than-500MW generation arrangement before month-end would raise the probability materially." }, { "persona": "capital-markets-credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.21, "rationale": "Only 24 days remain, and no qualifying announcement appears in the supplied digest. Converting the original 0.6 over roughly 116 days to a constant-hazard probability for the final 24 days gives about 0.17. Strong power-sector RPO growth and active hyperscaler infrastructure spending raise the chance somewhat, but the rule requires an explicit >500MW figure plus power-first or behind-the-meter language in a hyperscaler primary source or utility filing. General campus, land, energy-procurement, or future-generation announcements would miss under the ambiguity policy.", "keyEvidence": [], "whatWouldChangeMyMind": "A scheduled utility proceeding, regulatory agenda, or hyperscaler event before September 30 that previews a quantified generation block or power-secured campus exceeding 500MW." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.15, "rationale": "Base rate: 0.50 for disclosure questions (n=29); the 0.70 power-lens rate has only four observations. With just 24 days remaining, time-scaling the full-window base rate implies roughly 0.13. Strong supplier RPO and capex growth supports continued AI-power investment, modestly raising announcement risk. However, resolution requires a primary hyperscaler or utility source, an explicit capacity above 500MW, and power-first or behind-the-meter generation—not merely land or load plans. Those conjunctive disclosure requirements and ambiguity-against policy offset the positive sector signal. The forecaster’s 0.60–0.69 bin realized 0.556, confirming mild overconfidence.", "keyEvidence": [], "whatWouldChangeMyMind": "A scheduled hyperscaler infrastructure event or active utility docket explicitly previewing a named project above 500MW with behind-the-meter generation or power-first development would raise the estimate materially." }, { "persona": "geopolitics-regulation-analyst", "model": "openai/gpt-5.6-sol", "p": 0.21, "rationale": "Only 24 days remain from an originally roughly four-month window, with no qualifying disclosure supplied. A constant-hazard update from the 0.4 long-horizon base rate implies about 0.17 for the remaining period. Quarter-end announcement clustering and strong generation-equipment backlogs modestly raise that estimate. However, resolution requires a hyperscaler primary source or utility filing that explicitly exceeds 500MW and characterizes the project as power-first or behind-the-meter; generic campus, land, capacity, or power-procurement announcements fail. The lens:power sample is favorable but too small (n=4) to dominate.", "keyEvidence": [], "whatWouldChangeMyMind": "A scheduled hyperscaler infrastructure event, imminent utility commission filing, or credible preview explicitly tying one hyperscaler to more than 500MW of dedicated onsite generation before September 30." } ], "dissent": null, "spread": 0.06 }, "final": { "p": 0.34749810125412955, "method": "stacked-fixed-weights (resolved n=57 < 80)", "weights": { "calibrated": 0.3, "council": 0.5, "model": 0.2 }, "ci80": [ 0.25234447181646186, 0.4679770479431792 ] }, "direction": "down", "drivers": [ "Reference class qtype:disclosure runs 0.50 (n=29) vs stated 0.60.", "Hierarchical calibration moves stated 0.60 to 0.55 (lens power, kappa_stated=24.0, kappa_lens=200.0).", "Feature model at 0.55: conjuncts=1, named disclosure=True, horizon=116d.", "Council of 5 at 0.19." ], "warnings": [ "all 5 votes came from one vendor (openai); extremization disabled, persona diversity only", "2 cited URL(s) failed verification and were dropped." ], "evidence": [ { "url": "https://www.sec.gov/edgar/search/", "title": "SEC.gov | EDGAR Full Text Search", "publishedAt": null, "source": "direct", "excerpt": "Cited by semiconductor-supply-chain-analyst: Only 24 days remain, sharply reducing the opportunity relative to the original forecast window. Strong utility backlogs, generation capex, and equipment orders support continued hyperscaler power procurement, but they do not imply an imminent qualifying d" } ], "bubbleBasket": "energization", "daysRemaining": 24, "deadline": "2026-09-30", "lens": "power" } }, { "id": "lq-p45", "weeklyPredictionId": "p45-second-1-6t-design-win", "text": "A second non-Broadcom vendor cites a 1.6T or co-packaged-optics hyperscaler design win by September 30, 2026.", "lens": "networking", "deadline": "2026-09-30", "resolutionSource": "Marvell, Credo, or peer earnings/call.", "resolutionRule": "Hit requires a named second vendor besides Broadcom and 1.6T or CPO language tied to a hyperscaler.", "ambiguityPolicy": "resolves_against_forecaster", "stated": 0.55, "current": 0.3446075428814045, "direction": "down", "history": [ { "asOf": "2026-06-20", "p": 0.55, "source": "stated", "runId": null }, { "asOf": "2026-09-06", "p": 0.3446075428814045, "source": "engine", "runId": "2026-09-06T22-42-59Z-51a6cc" } ], "engine": { "id": "lq-p45", "weeklyPredictionId": "p45-second-1-6t-design-win", "stated": 0.55, "referenceClass": { "p": 0.5, "className": "qtype:disclosure", "n": 29, "note": "most specific ledger class with n >= 8; Resolved predictions of type disclosure: disclosure = requires a named primary document or named company; threshold = numeric bar without a named-disclosure requirement; timing = event-by-date without a numeric bar; other = none of the above." }, "calibrated": { "p": 0.5265707002549108, "method": "hierarchical-beta-binomial-by-lens", "ci80": [ 0.4847212778304268, 0.5683222675187027 ], "variants": { "platt-on-stated-logit": 0.4704 } }, "model": { "p": 0.2865544808157486, "method": "logistic-features-l2", "ci80": [ 0.12427027916944482, 0.531422518927866 ], "features": { "conjuncts": 1, "hasNumericThreshold": false, "horizonDays": 102, "isMarketPriceQuestion": false, "isTimingQuestion": true, "requiresNamedPrimaryDisclosure": true, "statedConfidence": 0.55, "thresholdMagnitudeLog": 0.204, "weekIndex": 8 }, "gbm": 0.437 }, "council": { "p": 0.2719770507053037, "aggregation": "log-odds-mean", "extremization": 1, "votes": [ { "persona": "grid-and-power-economist", "model": "openai/gpt-5.6-sol", "p": 0.21, "rationale": "The hurdle is unusually specific: a second identifiable non-Broadcom supplier must explicitly connect either 1.6T or co-packaged optics to a hyperscaler design win. Strong Credo and Marvell revenue growth supports underlying demand but does not satisfy that wording. With only 24 days left, and quarter-end dates indicating the most relevant Marvell and Credo reporting windows have likely already occurred, few scheduled disclosure opportunities remain. The forecaster’s networking base rate is only 0.318, and ambiguity resolves against a hit. I therefore discount the original 0.55 substantially.", "keyEvidence": [ "https://www.sec.gov/edgar/search/" ], "whatWouldChangeMyMind": "A scheduled September peer earnings call, or a filed transcript or release explicitly naming 1.6T/CPO hyperscaler design-win status, would raise the estimate sharply." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.23, "rationale": "The strict wording requires an explicit design win, 1.6T or CPO, and a hyperscaler linkage—not merely sampling, engagement, or a generic cloud customer. Marvell and Credo often avoid identifying customer classes precisely enough, and ambiguity resolves against the forecast. Their principal late-summer reporting windows are likely already past, leaving limited opportunity for a new qualifying earnings-call disclosure before September 30. Rapid Credo growth and broad hyperscaler networking investment preserve some chance that recent or upcoming transcript language qualifies. This is materially below the forecaster’s 0.55 and closer to the weak networking reference class.", "keyEvidence": [], "whatWouldChangeMyMind": "A transcript quote from Marvell, Credo, or a peer explicitly connecting a hyperscaler design win to 1.6T connectivity or co-packaged optics." }, { "persona": "capital-markets-credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.36, "rationale": "Credo’s 331% trailing revenue growth and Marvell’s 31% growth make hyperscaler networking wins plausible, and both have credible 1.6T exposure. However, the rule requires explicit design-win language, explicit 1.6T or CPO language, and a hyperscaler linkage. Product-roadmap or sampling comments would fail. The main late-August/early-September earnings windows are likely already past, leaving limited opportunity for a qualifying earnings disclosure before September 30. The networking reference rate (0.318) and the forecaster’s weak realized rate in the 0.50–0.59 band also argue below 0.55.", "keyEvidence": [ "https://www.sec.gov/edgar/search/" ], "whatWouldChangeMyMind": "A transcript or filed earnings release from Credo, Marvell, or another vendor explicitly describing a hyperscaler 1.6T or CPO design win." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.21, "rationale": "Base rate: 0.318 for the forecaster’s networking questions. Credo’s 331% and Marvell’s 31% trailing revenue growth support strong hyperscaler-networking activity, but resolution requires unusually specific public wording: a named non-Broadcom vendor, 1.6T or CPO, and a hyperscaler tie. Only 24 days remain, and ambiguity resolves against a hit. Disclosure timing is substantially less reliable than underlying design-win timing. The forecaster’s 0.50–0.59 calls realized only 0.367 versus 0.586 stated, reinforcing a downward correction from 0.55.", "keyEvidence": [], "whatWouldChangeMyMind": "An announced September earnings call or official transcript preview from Marvell, Credo, or a peer explicitly indicating hyperscaler adoption of 1.6T connectivity or co-packaged optics." }, { "persona": "geopolitics-regulation-analyst", "model": "openai/gpt-5.6-sol", "p": 0.38, "rationale": "The strongest disclosure catalysts—late-August or early-September earnings calls from Marvell, Credo, and networking peers—likely fall inside the window, while CRDO’s 331% trailing revenue growth supports active hyperscaler deployments. However, the rule demands explicit 1.6T or CPO, design-win, and hyperscaler linkage. Vendors often discuss sampling, engagements, or unnamed cloud customers without satisfying all three. With most relevant earnings catalysts probably already completed by September 6 and ambiguity resolving against the forecast, the networking and stated-confidence base rates warrant a material discount from 0.55.", "keyEvidence": [ "https://investor.marvell.com/", "https://investors.credosemi.com/", "https://www.sec.gov/edgar/search/" ], "whatWouldChangeMyMind": "A transcript explicitly saying Marvell, Credo, or another named vendor won a 1.6T/CPO design at a hyperscaler would raise this sharply; completed calls lacking that wording would lower it." } ], "dissent": null, "spread": 0.17 }, "final": { "p": 0.3446075428814045, "method": "stacked-fixed-weights (resolved n=57 < 80)", "weights": { "calibrated": 0.3, "council": 0.5, "model": 0.2 }, "ci80": [ 0.25515164131120943, 0.4657785537570452 ] }, "direction": "down", "drivers": [ "Reference class qtype:disclosure runs 0.50 (n=29) vs stated 0.55.", "Hierarchical calibration moves stated 0.55 to 0.53 (lens networking, kappa_stated=24.0, kappa_lens=200.0).", "Feature model at 0.29: conjuncts=1, named disclosure=True, horizon=102d.", "Council of 5 at 0.27." ], "warnings": [ "all 5 votes came from one vendor (openai); extremization disabled, persona diversity only" ], "evidence": [ { "url": "https://www.sec.gov/edgar/search/", "title": "SEC.gov | EDGAR Full Text Search", "publishedAt": null, "source": "direct", "excerpt": "Cited by grid-and-power-economist: The hurdle is unusually specific: a second identifiable non-Broadcom supplier must explicitly connect either 1.6T or co-packaged optics to a hyperscaler design win. Strong Credo and Marvell revenue growth supports underlying demand but does not satisfy that wording" }, { "url": "https://investor.marvell.com/", "title": "Investor Relations | Marvell Technology, Inc. (MRVL)", "publishedAt": null, "source": "direct", "excerpt": "Cited by geopolitics-regulation-analyst: The strongest disclosure catalysts—late-August or early-September earnings calls from Marvell, Credo, and networking peers—likely fall inside the window, while CRDO’s 331% trailing revenue growth supports active hyperscaler deployments. However, the rule dema" }, { "url": "https://investors.credosemi.com/", "title": "Credo Technology Group Holding Ltd - Investor Relations", "publishedAt": null, "source": "direct", "excerpt": "Cited by geopolitics-regulation-analyst: The strongest disclosure catalysts—late-August or early-September earnings calls from Marvell, Credo, and networking peers—likely fall inside the window, while CRDO’s 331% trailing revenue growth supports active hyperscaler deployments. However, the rule dema" } ], "bubbleBasket": null, "daysRemaining": 24, "deadline": "2026-09-30", "lens": "networking" } }, { "id": "lq-p50", "weeklyPredictionId": "p50-groq-enterprise-customer", "text": "Groq announces at least one named Fortune 500 or hyperscaler inference-cloud customer by September 30, 2026.", "lens": "capital", "deadline": "2026-09-30", "resolutionSource": "Groq or customer primary disclosure.", "resolutionRule": "Hit on a named customer. Unnamed 'enterprise logos' miss.", "ambiguityPolicy": "resolves_against_forecaster", "stated": 0.57, "current": 0.3710853853606606, "direction": "down", "history": [ { "asOf": "2026-06-27", "p": 0.57, "source": "stated", "runId": null }, { "asOf": "2026-09-06", "p": 0.3710853853606606, "source": "engine", "runId": "2026-09-06T22-42-59Z-51a6cc" } ], "engine": { "id": "lq-p50", "weeklyPredictionId": "p50-groq-enterprise-customer", "stated": 0.57, "referenceClass": { "p": 0.5, "className": "qtype:disclosure", "n": 29, "note": "most specific ledger class with n >= 8; Resolved predictions of type disclosure: disclosure = requires a named primary document or named company; threshold = numeric bar without a named-disclosure requirement; timing = event-by-date without a numeric bar; other = none of the above." }, "calibrated": { "p": 0.5476491228070176, "method": "hierarchical-beta-binomial-by-lens", "ci80": [ 0.5059819240116754, 0.5891415900656223 ], "variants": { "platt-on-stated-logit": 0.4819 } }, "model": { "p": 0.6523297658212814, "method": "logistic-features-l2", "ci80": [ 0.44438417010093917, 0.8243390328027903 ], "features": { "conjuncts": 1, "hasNumericThreshold": true, "horizonDays": 95, "isMarketPriceQuestion": false, "isTimingQuestion": true, "requiresNamedPrimaryDisclosure": true, "statedConfidence": 0.57, "thresholdMagnitudeLog": 2.699, "weekIndex": 9 }, "gbm": 0.344 }, "council": { "p": 0.19441856257809256, "aggregation": "log-odds-mean", "extremization": 1, "votes": [ { "persona": "grid-and-power-economist", "model": "openai/gpt-5.6-sol", "p": 0.17, "rationale": "Only 24 days remain, and the supplied digest contains no qualifying primary disclosure. The rule is narrow: Groq must name a Fortune 500 company or hyperscaler specifically as an inference-cloud customer; partnerships, model providers, unnamed enterprise logos, and customer rumors fail. Named-customer permissions commonly delay or prevent such announcements. The forecaster’s comparable 0.50–0.59 calls realized only 0.367, while its disclosure class realized 0.5. Strong AI capital spending supports underlying demand but is weak evidence for a customer-approved announcement before quarter-end. Conditioning on most of the original forecast window passing without resolution warrants a substantial reduction from 0.57.", "keyEvidence": [ "https://groq.com/" ], "whatWouldChangeMyMind": "A dated Groq or customer announcement naming a qualifying company and explicitly describing paid or contracted inference-cloud use would move this near certainty." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.24, "rationale": "Only 24 days remain, and no qualifying primary disclosure is provided. Groq could use a quarter-end announcement to validate adoption, while strong AI infrastructure spending makes a real deployment plausible. However, the rule requires both a named Fortune 500 or hyperscaler and an inference-cloud customer relationship; partnerships, infrastructure suppliers, unnamed logos, trials, and GroqCloud availability would miss. Customer approval and confidentiality often delay naming beyond deployment. Conditioning the original thesis on roughly three-quarters of the forecast window passing without a hit materially lowers the probability.", "keyEvidence": [], "whatWouldChangeMyMind": "A scheduled Groq launch event, customer earnings commentary naming Groq, or a jointly promoted production inference deployment before quarter-end would raise it substantially." }, { "persona": "capital-markets-credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.21, "rationale": "Only 24 days remain, and the supplied digest contains no Groq-specific disclosure catalyst or qualifying named customer. Strong hyperscaler capex and benign broad credit conditions support inference demand but do not imply permission to name an enterprise customer. Customer announcements often require coordinated approvals, while private vendors can continue citing adoption or logos without disclosing names. The literal rule also excludes unnamed enterprises and non-Fortune-500 partners. Given the shortened window and the forecaster’s weak realized rate in the 0.50–0.59 band, the original 0.57 now appears too high.", "keyEvidence": [ "https://groq.com/" ], "whatWouldChangeMyMind": "A scheduled Groq/customer event, launch materials identifying a qualifying company, or a credible report that joint announcement approvals are complete." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.16, "rationale": "I use the disclosure-question base rate of 0.50 (29 cases), tempered by the forecaster’s 0.50–0.59 realized rate of 0.367 and matched-ledger rate of 0.286. Only 24 days remain, and no qualifying primary disclosure is supplied. The rule requires a named Fortune 500 or hyperscaler customer plus a Groq-or-customer primary announcement; generic enterprise traction, partner claims, or later confirmation fail. Named-customer disclosures face confidentiality and approval delays. The capital-lens rate of 0.708 is small-sample and less diagnostic than disclosure calibration. Quarter-end provides a modest catalyst, but not enough to offset the short window and strict documentation requirement.", "keyEvidence": [], "whatWouldChangeMyMind": "A scheduled September Groq launch with a customer speaker, or an official customer teaser explicitly tying Groq to production inference-cloud use, would raise the probability substantially." }, { "persona": "geopolitics-regulation-analyst", "model": "openai/gpt-5.6-sol", "p": 0.2, "rationale": "Only 24 days remain, and the rule requires a primary disclosure naming a qualifying Fortune 500 company or hyperscaler specifically as an inference-cloud customer. Groq can announce partnerships, integrations, or unnamed enterprise adoption without satisfying that test. Hyperscalers also have limited incentive to publicize reliance on an external inference provider. Quarter-end announcement clustering and strong AI infrastructure spending provide some upside, but supportive capital conditions do not overcome the narrow documentation requirement. The forecaster’s 0.50–0.59 band has materially underperformed, further favoring a downward adjustment from 0.57.", "keyEvidence": [ "https://groq.com/" ], "whatWouldChangeMyMind": "A scheduled September Groq launch featuring enterprise case studies, or a customer teaser explicitly identifying production GroqCloud inference use, would raise the probability substantially." } ], "dissent": null, "spread": 0.08 }, "final": { "p": 0.3710853853606606, "method": "stacked-fixed-weights (resolved n=57 < 80)", "weights": { "calibrated": 0.3, "council": 0.5, "model": 0.2 }, "ci80": [ 0.29595957129462147, 0.46033125361677507 ] }, "direction": "down", "drivers": [ "Reference class qtype:disclosure runs 0.50 (n=29) vs stated 0.57.", "Hierarchical calibration moves stated 0.57 to 0.55 (lens capital, kappa_stated=24.0, kappa_lens=200.0).", "Feature model at 0.65: conjuncts=1, named disclosure=True, horizon=95d.", "Council of 5 at 0.19." ], "warnings": [ "all 5 votes came from one vendor (openai); extremization disabled, persona diversity only" ], "evidence": [ { "url": "https://groq.com/", "title": "Groq is the premier neocloud for fast inference", "publishedAt": null, "source": "direct", "excerpt": "Cited by grid-and-power-economist: Only 24 days remain, and the supplied digest contains no qualifying primary disclosure. The rule is narrow: Groq must name a Fortune 500 company or hyperscaler specifically as an inference-cloud customer; partnerships, model providers, unnamed enterprise logos, and" } ], "bubbleBasket": null, "daysRemaining": 24, "deadline": "2026-09-30", "lens": "capital" } }, { "id": "lq-p60", "weeklyPredictionId": "p60-scale-across-follow-on", "text": "A second named vendor or operator announces a commercial cross-data-center scale-across product by September 30, 2026.", "lens": "networking", "deadline": "2026-09-30", "resolutionSource": "Vendor or operator primary disclosure.", "resolutionRule": "Hit if a second named party besides the first already-counted one announces commercial (not lab) scale-across.", "ambiguityPolicy": "resolves_against_forecaster", "stated": 0.61, "current": 0.35171141121570965, "direction": "down", "history": [ { "asOf": "2026-07-11", "p": 0.61, "source": "stated", "runId": null }, { "asOf": "2026-09-06", "p": 0.35171141121570965, "source": "engine", "runId": "2026-09-06T22-42-59Z-51a6cc" } ], "engine": { "id": "lq-p60", "weeklyPredictionId": "p60-scale-across-follow-on", "stated": 0.61, "referenceClass": { "p": 0.5, "className": "qtype:disclosure", "n": 29, "note": "most specific ledger class with n >= 8; Resolved predictions of type disclosure: disclosure = requires a named primary document or named company; threshold = numeric bar without a named-disclosure requirement; timing = event-by-date without a numeric bar; other = none of the above." }, "calibrated": { "p": 0.532724546408757, "method": "hierarchical-beta-binomial-by-lens", "ci80": [ 0.4908945518101038, 0.5744340223913876 ], "variants": { "platt-on-stated-logit": 0.5053 } }, "model": { "p": 0.316168894701467, "method": "logistic-features-l2", "ci80": [ 0.153246017053889, 0.511304645015549 ], "features": { "conjuncts": 1, "hasNumericThreshold": false, "horizonDays": 81, "isMarketPriceQuestion": false, "isTimingQuestion": true, "requiresNamedPrimaryDisclosure": true, "statedConfidence": 0.61, "thresholdMagnitudeLog": 0, "weekIndex": 11 }, "gbm": 0.501 }, "council": { "p": 0.2702947792206023, "aggregation": "log-odds-mean", "extremization": 1, "votes": [ { "persona": "grid-and-power-economist", "model": "openai/gpt-5.6-sol", "p": 0.23, "rationale": "Only 24 days remain, and most of the original forecast window has elapsed without a qualifying second announcement. The networking reference-class rate is low at 0.318. Strong vendor revenue and investment indicate demand, but they do not establish a commercial cross-data-center product. The rule also requires a named second party, a primary disclosure, and commercial—not laboratory—status; vague interoperability, trials, or component launches would fail under the ambiguity policy. A late-September launch remains plausible, but the forecaster’s July probability should decay materially with elapsed time.", "keyEvidence": [], "whatWouldChangeMyMind": "A scheduled September launch event, partner briefing, or primary release explicitly describing an orderable cross-data-center scale-across product from a second named vendor or operator." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.24, "rationale": "Only 24 days remain, and the supplied digest shows strong networking demand but no qualifying second primary disclosure. Most likely launch venues from the longer July-to-September window have already passed, so the original 0.61 should decay materially conditional on no announcement. Arista, Cisco, Broadcom, Ciena, or an operator could respond to the first vendor, but commercial cross-data-center scale-across is a narrower threshold than a roadmap, trial, interoperability demo, or component announcement. The networking ledger’s 0.318 base rate and ambiguity-against treatment further support a below-even forecast.", "keyEvidence": [], "whatWouldChangeMyMind": "A scheduled September product event or credible vendor teaser explicitly promising an orderable multi-data-center AI networking product would raise this sharply." }, { "persona": "capital-markets-credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.34, "rationale": "Only 24 days remain, and the rule requires a second named party, a primary announcement, commercial status, and explicit cross-data-center scale-across capability. Strong networking demand and an already-counted first mover make a competitive response plausible, while a press release requires little lead time. However, the networking reference-class rate is only 0.318, and adjacent fabric, interconnect, or lab demonstrations may not satisfy the terminology under adverse ambiguity resolution. The absence of direct evidence in the supplied digest keeps my estimate well below the forecaster’s 0.61.", "keyEvidence": [], "whatWouldChangeMyMind": "A scheduled September launch or primary vendor materials explicitly describing an orderable or generally available product that scales one AI workload across separate data centers." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.26, "rationale": "Base rate: 0.318 from the networking reference class. I adjust downward because only 24 days remain and resolution requires a second named party’s primary disclosure of a commercial—not laboratory—product; broad sector growth does not establish announcement timing or commercial status. The digest contains strong networking-related revenue and backlog signals, but no qualifying product disclosure. Ambiguity resolves against a hit. The forecaster’s 0.60–0.69 band realized at 0.556, confirming modest overconfidence, though the strict deadline and documentation requirements warrant a larger reduction.", "keyEvidence": [], "whatWouldChangeMyMind": "A dated vendor event agenda, launch teaser, or credible preannouncement explicitly promising commercially available cross-data-center scale-across before September 30 would raise the estimate substantially." }, { "persona": "geopolitics-regulation-analyst", "model": "openai/gpt-5.6-sol", "p": 0.29, "rationale": "Only 24 days remain, and the supplied evidence contains no qualifying second primary disclosure. The closest empirical class—networking—has a 0.318 resolved base rate, while the forecaster’s 0.60–0.69 calls have realized below stated confidence. A September launch cycle still leaves meaningful upside, especially because one party has already validated the category. However, the strict requirement for a named party, commercial status, cross-data-center scope, and a primary announcement—combined with ambiguity resolving against the forecast—makes slippage or non-qualification likely.", "keyEvidence": [], "whatWouldChangeMyMind": "A dated September launch event or primary preview explicitly promising commercially available cross-data-center scaling would raise the estimate substantially." } ], "dissent": null, "spread": 0.11 }, "final": { "p": 0.35171141121570965, "method": "stacked-fixed-weights (resolved n=57 < 80)", "weights": { "calibrated": 0.3, "council": 0.5, "model": 0.2 }, "ci80": [ 0.2774879736861746, 0.4421077344967833 ] }, "direction": "down", "drivers": [ "Reference class qtype:disclosure runs 0.50 (n=29) vs stated 0.61.", "Hierarchical calibration moves stated 0.61 to 0.53 (lens networking, kappa_stated=24.0, kappa_lens=200.0).", "Feature model at 0.32: conjuncts=1, named disclosure=True, horizon=81d.", "Council of 5 at 0.27." ], "warnings": [ "all 5 votes came from one vendor (openai); extremization disabled, persona diversity only" ], "evidence": [], "bubbleBasket": null, "daysRemaining": 24, "deadline": "2026-09-30", "lens": "networking" } }, { "id": "lq-p71", "weeklyPredictionId": "p71-luna-task-cost-sep30", "text": "An independent evaluator publishes completed-task cost showing GPT-5.6 Luna cheaper than the stated bar by September 30, 2026.", "lens": "software", "deadline": "2026-09-30", "resolutionSource": "Artificial Analysis or another named independent evaluator.", "resolutionRule": "Hit only on a published completed-task cost comparison. Vendor blog numbers without an independent evaluator miss.", "ambiguityPolicy": "resolves_against_forecaster", "stated": 0.62, "current": 0.4386043568918027, "direction": "down", "history": [ { "asOf": "2026-08-01", "p": 0.62, "source": "stated", "runId": null }, { "asOf": "2026-09-06", "p": 0.4386043568918027, "source": "engine", "runId": "2026-09-06T22-42-59Z-51a6cc" } ], "engine": { "id": "lq-p71", "weeklyPredictionId": "p71-luna-task-cost-sep30", "stated": 0.62, "referenceClass": { "p": 0.3888888888888889, "className": "qtype:timing", "n": 8, "note": "most specific ledger class with n >= 8; Resolved predictions of type timing: disclosure = requires a named primary document or named company; threshold = numeric bar without a named-disclosure requirement; timing = event-by-date without a numeric bar; other = none of the above." }, "calibrated": { "p": 0.5364653436870865, "method": "hierarchical-beta-binomial-by-lens", "ci80": [ 0.49551631654298445, 0.5772856118655965 ], "variants": { "platt-on-stated-logit": 0.5113 } }, "model": { "p": 0.6078386296764768, "method": "logistic-features-l2", "ci80": [ 0.3912045992744522, 0.7551003315326092 ], "features": { "conjuncts": 1, "hasNumericThreshold": false, "horizonDays": 60, "isMarketPriceQuestion": true, "isTimingQuestion": true, "requiresNamedPrimaryDisclosure": false, "statedConfidence": 0.62, "thresholdMagnitudeLog": 0.778, "weekIndex": 14 }, "gbm": 0.631 }, "council": { "p": 0.3193798629757256, "aggregation": "log-odds-mean", "extremization": 1, "votes": [ { "persona": "grid-and-power-economist", "model": "openai/gpt-5.6-sol", "p": 0.32, "rationale": "Only 24 days remain, and the digest provides no evidence that GPT-5.6 Luna is publicly available or undergoing independent evaluation. Resolution requires a completed-task cost comparison, a stricter and less common artifact than token pricing or vendor benchmark claims. The software and timing reference classes are only 0.452 and 0.389, while the forecaster’s 0.60–0.69 predictions have realized below stated confidence. An evaluator could test a newly released API quickly, but release, testing, publication, and beating the specified bar must all occur by September 30; ambiguity counts as a miss.", "keyEvidence": [ "https://artificialanalysis.ai/" ], "whatWouldChangeMyMind": "Confirmed API availability plus an evaluator’s dated benchmark announcement or preliminary completed-task cost table would raise my estimate substantially." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.31, "rationale": "Only 24 days remain, and the required evidence is narrower than a model release or vendor pricing: an independent evaluator must publish a completed-task cost comparison and clear the specified bar. Evaluators can benchmark API models within days, but completed-task metrics require repeat runs and are less routinely published than token prices or quality scores. The event was forecast on August 1 and remains unmet by September 6, which materially reduces a timing forecast’s conditional probability. The forecaster’s software and timing base rates are also weak, with overprediction in the 0.60–0.69 band. Ambiguity resolving against the forecast further penalizes borderline comparisons.", "keyEvidence": [ "https://artificialanalysis.ai/" ], "whatWouldChangeMyMind": "Immediate API availability plus an Artificial Analysis announcement that its task-level benchmark is running, or a dated evaluator preview explicitly reporting completed-task cost against the stated bar." }, { "persona": "capital-markets-credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.32, "rationale": "Only 24 days remain, and the supplied digest shows no qualifying independent completed-task cost result. The resolution is stricter than ordinary model pricing or benchmark coverage: GPT-5.6 Luna must be evaluated, the metric must represent cost per completed task, and the result must beat the specified bar. Independent evaluators can add prominent models quickly, but testing and publishing a methodology-specific comparison creates timing risk. The forecaster’s software and timing reference classes are below its stated 0.62, and its 0.60–0.69 forecasts have been overconfident. Ambiguity resolving against the forecaster further reduces the chance that a near-match qualifies.", "keyEvidence": [ "https://artificialanalysis.ai/" ], "whatWouldChangeMyMind": "A public GPT-5.6 Luna release and an evaluator announcement, preview, or leaderboard schema explicitly showing completed-task cost against the stated threshold before month-end." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.24, "rationale": "I use the 0.389 timing-question base rate, cross-checked against the 0.452 software rate. This requires several linked developments within 24 days: GPT-5.6 Luna must be evaluable, an independent evaluator must run a completed-task cost comparison, the result must beat the bar, and publication must occur by the deadline. The digest supplies no model- or evaluator-specific evidence supporting those steps. The strict publication requirement and ambiguity-against policy materially reduce the estimate. The forecaster’s 0.60–0.69 band has realized only 0.556, confirming systematic overconfidence and making 0.62 too high.", "keyEvidence": [ "https://artificialanalysis.ai/" ], "whatWouldChangeMyMind": "An independent evaluator announcing an active Luna completed-task benchmark, or publishing qualifying cost results before September 30, would raise the estimate sharply." }, { "persona": "geopolitics-regulation-analyst", "model": "openai/gpt-5.6-sol", "p": 0.42, "rationale": "The 24-day window is feasible for an evaluator update, but the rule requires completed-task cost—not token pricing, vendor claims, or a general benchmark score—and the result must beat a specific bar. That creates both publication and performance hurdles. The forecaster’s timing and software reference classes are below 0.5, while its 0.60–0.69 forecasts have realized below their stated confidence. No supplied evidence confirms that GPT-5.6 Luna is already evaluator-accessible or that a qualifying cost study is scheduled. Ambiguity resolving against the forecast further reduces the chance of a marginal comparison counting.", "keyEvidence": [ "https://artificialanalysis.ai/" ], "whatWouldChangeMyMind": "Evaluator confirmation that Luna testing is underway, with a benchmark explicitly reporting cost per completed task and publication planned before September 30." } ], "dissent": null, "spread": 0.18 }, "final": { "p": 0.4386043568918027, "method": "stacked-fixed-weights (resolved n=57 < 80)", "weights": { "calibrated": 0.3, "council": 0.5, "model": 0.2 }, "ci80": [ 0.33845863477467, 0.5392455617588773 ] }, "direction": "down", "drivers": [ "Reference class qtype:timing runs 0.39 (n=8) vs stated 0.62.", "Hierarchical calibration moves stated 0.62 to 0.54 (lens software, kappa_stated=24.0, kappa_lens=200.0).", "Feature model at 0.61: conjuncts=1, named disclosure=False, horizon=60d.", "Council of 5 at 0.32." ], "warnings": [ "all 5 votes came from one vendor (openai); extremization disabled, persona diversity only" ], "evidence": [ { "url": "https://artificialanalysis.ai/", "title": "AI Model & API Providers Analysis | Artificial Analysis", "publishedAt": null, "source": "direct", "excerpt": "Cited by grid-and-power-economist: Only 24 days remain, and the digest provides no evidence that GPT-5.6 Luna is publicly available or undergoing independent evaluation. Resolution requires a completed-task cost comparison, a stricter and less common artifact than token pricing or vendor benchmark c" } ], "bubbleBasket": null, "daysRemaining": 24, "deadline": "2026-09-30", "lens": "software" } }, { "id": "lq-p83", "weeklyPredictionId": "p83-deepseek-0813-weights-sep30", "text": "DeepSeek publishes the V4-Pro-0813 build weights to Hugging Face by September 30, 2026.", "lens": "software", "deadline": "2026-09-30", "resolutionSource": "https://huggingface.co/", "resolutionRule": "Hit if the named build weights are public on Hugging Face with a commit dated on or before 2026-09-30.", "ambiguityPolicy": "resolves_against_forecaster", "stated": 0.64, "current": 0.46063666269036246, "direction": "down", "history": [ { "asOf": "2026-08-15", "p": 0.64, "source": "stated", "runId": null }, { "asOf": "2026-09-06", "p": 0.46063666269036246, "source": "engine", "runId": "2026-09-06T22-42-59Z-51a6cc" } ], "engine": { "id": "lq-p83", "weeklyPredictionId": "p83-deepseek-0813-weights-sep30", "stated": 0.64, "referenceClass": { "p": 0.3888888888888889, "className": "qtype:timing", "n": 8, "note": "most specific ledger class with n >= 8; Resolved predictions of type timing: disclosure = requires a named primary document or named company; threshold = numeric bar without a named-disclosure requirement; timing = event-by-date without a numeric bar; other = none of the above." }, "calibrated": { "p": 0.5384325568018407, "method": "hierarchical-beta-binomial-by-lens", "ci80": [ 0.4974922302125586, 0.5792371782332143 ], "variants": { "platt-on-stated-logit": 0.5234 } }, "model": { "p": 0.4288847970928873, "method": "logistic-features-l2", "ci80": [ 0.24297303267299894, 0.5900479347650347 ], "features": { "conjuncts": 1, "hasNumericThreshold": false, "horizonDays": 46, "isMarketPriceQuestion": false, "isTimingQuestion": true, "requiresNamedPrimaryDisclosure": false, "statedConfidence": 0.64, "thresholdMagnitudeLog": 0, "weekIndex": 16 }, "gbm": 0.7367 }, "council": { "p": 0.42716701866236684, "aggregation": "log-odds-mean", "extremization": 1, "votes": [ { "persona": "grid-and-power-economist", "model": "openai/gpt-5.6-sol", "p": 0.34, "rationale": "DeepSeek remains active on Hugging Face, with two August releases and an August 31 experimental V4-related upload, so another near-term release is plausible. However, the digest does not identify V4-Pro-0813, despite the build label suggesting August 13 and more than three weeks having elapsed. Exact-build weight publication is stricter than an API launch, announcement, derivative checkpoint, or similarly named V4 release. Only 24 days remain, and the timing/software reference classes are below the forecaster’s 0.64. The hard commit-date rule and adverse ambiguity policy further reduce the probability.", "keyEvidence": [ "https://huggingface.co/deepseek-ai" ], "whatWouldChangeMyMind": "A DeepSeek model card, repository placeholder, or official announcement explicitly naming V4-Pro-0813 and stating that full weights will be uploaded before September 30." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.46, "rationale": "DeepSeek remains active on Hugging Face, with two August releases and a V4-related experimental model posted August 31, so another near-term V4 upload is plausible. However, V4-Pro-0813 is still absent more than three weeks after the apparent build date, leaving only 24 days. Exact naming and public weight files are required; an announcement, API release, renamed build, or late commit would fail. The forecaster’s software and timing reference classes also resolve below 0.5, and its 0.60–0.69 forecasts have historically realized below stated confidence.", "keyEvidence": [ "https://huggingface.co/deepseek-ai" ], "whatWouldChangeMyMind": "A DeepSeek model card, repository placeholder, or official statement explicitly naming V4-Pro-0813 and promising September weights would raise the estimate substantially." }, { "persona": "capital-markets-credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.42, "rationale": "DeepSeek is an active Hugging Face publisher, with 21 repositories added from June through August and a V4-family experimental release on August 31. That makes another near-term V4 upload plausible. However, the exact V4-Pro-0813 weights were still absent by September 6, despite the build name implying an August 13 vintage. Only 24 days remain, and exact-name plus commit-date requirements eliminate credit for related V4 releases. Timing-question and software reference rates are also below the forecaster’s 0.64.", "keyEvidence": [ "https://huggingface.co/deepseek-ai" ], "whatWouldChangeMyMind": "An official DeepSeek announcement naming V4-Pro-0813 and promising open weights, or a visible gated Hugging Face repository for that exact build, would raise the estimate substantially." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.46, "rationale": "I use the timing-question base rate of 0.389, cross-checked against the software rate of 0.452. DeepSeek’s recent Hugging Face activity and an adjacent V4-family upload on August 31 raise the estimate. However, success requires the exact V4-Pro-0813 weights—not merely a related model—to receive a public commit within 24 days; ambiguity counts as failure. The forecaster’s 0.60–0.69 band has realized 0.556 versus 0.663 stated, indicating overconfidence. No digest evidence confirms this build’s release or an announced date, so 0.64 is too high.", "keyEvidence": [ "https://huggingface.co/deepseek-ai", "https://huggingface.co/" ], "whatWouldChangeMyMind": "A DeepSeek announcement naming V4-Pro-0813 and a September publication date, or a visible Hugging Face repository with weights or substantive upload commits, would raise the estimate sharply." }, { "persona": "geopolitics-regulation-analyst", "model": "openai/gpt-5.6-sol", "p": 0.46, "rationale": "The digest shows DeepSeek actively publishing on Hugging Face, including two August releases and a V4-family experimental model on August 31. That makes another near-term upload plausible. However, the exact V4-Pro-0813 weights were still absent as of September 6, despite the build name implying an August 13 vintage. Only 24 days remain, and the rule requires the precisely named build with a qualifying commit—not merely another V4 variant, model card, or API release. Timing questions in the ledger also have a low 0.389 base rate. These factors put the event below even odds and below the forecaster’s 0.64.", "keyEvidence": [ "https://huggingface.co/deepseek-ai" ], "whatWouldChangeMyMind": "A DeepSeek announcement naming V4-Pro-0813, a visible gated repository, or commits indicating an imminent weight upload would raise the estimate substantially." } ], "dissent": null, "spread": 0.12 }, "final": { "p": 0.46063666269036246, "method": "stacked-fixed-weights (resolved n=57 < 80)", "weights": { "calibrated": 0.3, "council": 0.5, "model": 0.2 }, "ci80": [ 0.3630972799498075, 0.5221240939568254 ] }, "direction": "down", "drivers": [ "Reference class qtype:timing runs 0.39 (n=8) vs stated 0.64.", "Hierarchical calibration moves stated 0.64 to 0.54 (lens software, kappa_stated=24.0, kappa_lens=200.0).", "Feature model at 0.43: conjuncts=1, named disclosure=False, horizon=46d.", "Council of 5 at 0.43." ], "warnings": [ "all 5 votes came from one vendor (openai); extremization disabled, persona diversity only" ], "evidence": [ { "url": "https://huggingface.co/deepseek-ai", "title": "deepseek-ai (DeepSeek)", "publishedAt": null, "source": "direct", "excerpt": "Cited by grid-and-power-economist: DeepSeek remains active on Hugging Face, with two August releases and an August 31 experimental V4-related upload, so another near-term release is plausible. However, the digest does not identify V4-Pro-0813, despite the build label suggesting August 13 and more th" }, { "url": "https://huggingface.co/", "title": "Hugging Face – The AI community building the future.", "publishedAt": null, "source": "direct", "excerpt": "Cited by superforecaster-statistician: I use the timing-question base rate of 0.389, cross-checked against the software rate of 0.452. DeepSeek’s recent Hugging Face activity and an adjacent V4-family upload on August 31 raise the estimate. However, success requires the exact V4-Pro-0813 weights—not" } ], "bubbleBasket": null, "daysRemaining": 24, "deadline": "2026-09-30", "lens": "software" } }, { "id": "lq-p96", "weeklyPredictionId": "p96-openrouter-volume-sep30", "text": "Highest single ISO week of OpenRouter aggregate token volume in September 2026 stays below the stated bar.", "lens": "software", "deadline": "2026-09-30", "resolutionSource": "OpenRouter public volume stats.", "resolutionRule": "Score against the weekly prediction's written trigger. If OpenRouter revises history, use the vintage available on 2026-09-30.", "ambiguityPolicy": "resolves_against_forecaster", "stated": 0.4, "current": 0.2506842521466603, "direction": "down", "history": [ { "asOf": "2026-08-29", "p": 0.4, "source": "stated", "runId": null }, { "asOf": "2026-09-06", "p": 0.2506842521466603, "source": "engine", "runId": "2026-09-06T22-42-59Z-51a6cc" } ], "engine": { "id": "lq-p96", "weeklyPredictionId": "p96-openrouter-volume-sep30", "stated": 0.4, "referenceClass": { "p": 0.65, "className": "qtype:threshold", "n": 19, "note": "most specific ledger class with n >= 8; Resolved predictions of type threshold: disclosure = requires a named primary document or named company; threshold = numeric bar without a named-disclosure requirement; timing = event-by-date without a numeric bar; other = none of the above." }, "calibrated": { "p": 0.5148259994247915, "method": "hierarchical-beta-binomial-by-lens", "ci80": [ 0.4738235887560388, 0.5557760594737233 ], "variants": { "platt-on-stated-logit": 0.3864 } }, "model": { "p": 0.34386022976838027, "method": "logistic-features-l2", "ci80": [ 0.14892459872384498, 0.5732797896482984 ], "features": { "conjuncts": 1, "hasNumericThreshold": true, "horizonDays": 32, "isMarketPriceQuestion": false, "isTimingQuestion": false, "requiresNamedPrimaryDisclosure": false, "statedConfidence": 0.4, "thresholdMagnitudeLog": 0, "weekIndex": 18 }, "gbm": 0.6333 }, "council": { "p": 0.12270395257056921, "aggregation": "log-odds-mean", "extremization": 1, "votes": [ { "persona": "grid-and-power-economist", "model": "openai/gpt-5.6-sol", "p": 0.31, "rationale": "No numeric bar, OpenRouter baseline, or September-to-date volume is supplied, so the written trigger cannot be independently tested. Structurally, taking the maximum across several ISO weeks makes staying below a bar less likely than a single-week comparison. Late-August and early-September model releases could also produce traffic spikes, though release cadence is only an indirect signal. The 2026-09-30 vintage rule is clear, but treatment of the ISO week spanning September 28–October 4 may be ambiguous; the stated policy resolves ambiguity against the forecaster. I therefore put the event modestly below the forecaster’s 0.4.", "keyEvidence": [], "whatWouldChangeMyMind": "The numeric bar plus OpenRouter’s completed August and September weekly volumes. A substantial cushion below the bar across early September would raise the estimate." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.08, "rationale": "No numeric token-volume bar appears in the supplied written trigger, and the digest provides no OpenRouter weekly volume observations. The resolution rule requires comparison with that written trigger, while ambiguity explicitly resolves against the forecaster. Therefore the affirmative “stays below” outcome is unlikely to score without a recoverable, unambiguous threshold, irrespective of underlying usage trends. The forecaster’s ledger offers little help: only one prior forecast was below 0.50 confidence.", "keyEvidence": [], "whatWouldChangeMyMind": "An explicit archived threshold plus September weekly OpenRouter volumes clearly tracking below it." }, { "persona": "capital-markets-credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.04, "rationale": "The supplied written trigger does not state a numeric bar, so the required comparison cannot be made from the question. It also leaves unclear whether an ISO week merely overlapping September qualifies and how the incomplete week beginning September 28 is treated at the September 30 vintage. Because the rule requires scoring the written trigger literally and ambiguity resolves against the forecaster, these omissions make a positive resolution unlikely regardless of the underlying volume trend.", "keyEvidence": [], "whatWouldChangeMyMind": "The original written prediction showing a numeric threshold and explicit ISO-week treatment, followed by OpenRouter’s September 30 vintage remaining below that threshold." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.06, "rationale": "I use the software-ledger base rate of 0.452; the broader threshold base rate is 0.650 but is less specific. The numerical bar is absent from the supplied written trigger, and the digest provides neither OpenRouter’s current volume nor its recent trend. Under literal scoring and an ambiguity policy that resolves against the forecaster, that omission is a major failure risk. The claim also concerns the maximum across all September ISO weeks, so one spike can defeat it. The forecaster’s low-confidence calibration bin has only one observation and provides no useful correction.", "keyEvidence": [], "whatWouldChangeMyMind": "The exact numerical bar plus official OpenRouter weekly history showing September-to-date volume safely below it, with limited remaining spike risk." }, { "persona": "geopolitics-regulation-analyst", "model": "openai/gpt-5.6-sol", "p": 0.34, "rationale": "The numeric bar and current weekly volume are absent, so there is no observable cushion to extrapolate. Literal resolution and ambiguity against the forecaster materially reduce the chance of success, especially because September 30 cuts through an ISO week and historical revisions use that day’s vintage. Several late-August and early-September model releases could also generate a weekly spike. The software reference rate of 0.452 is more relevant than the stronger generic threshold rate; after the specification and boundary-week penalties, I place this below the forecaster’s 0.4.", "keyEvidence": [ "https://openrouter.ai/rankings" ], "whatWouldChangeMyMind": "The actual numeric bar, August weekly volumes, and OpenRouter’s precise treatment of partial ISO weeks would dominate; a bar comfortably above the latest peak would raise the estimate sharply." } ], "dissent": "geopolitics-regulation-analyst at 0.34 vs capital-markets-credit-analyst at 0.04: The numeric bar and current weekly volume are absent, so there is no observable cushion to extrapolate. / The supplied written trigger does not state a numeric bar, so the required comparison cannot be made from the question.", "spread": 0.3 }, "final": { "p": 0.2506842521466603, "method": "stacked-fixed-weights (resolved n=57 < 80)", "weights": { "calibrated": 0.3, "council": 0.5, "model": 0.2 }, "ci80": [ 0.12248779479461207, 0.4488329793494572 ] }, "direction": "down", "drivers": [ "Reference class qtype:threshold runs 0.65 (n=19) vs stated 0.40.", "Hierarchical calibration moves stated 0.40 to 0.51 (lens software, kappa_stated=24.0, kappa_lens=200.0).", "Feature model at 0.34: conjuncts=1, named disclosure=False, horizon=32d.", "Council of 5 at 0.12 with dissent." ], "warnings": [ "all 5 votes came from one vendor (openai); extremization disabled, persona diversity only" ], "evidence": [ { "url": "https://openrouter.ai/rankings", "title": "LLM Rankings | OpenRouter", "publishedAt": null, "source": "direct", "excerpt": "Cited by geopolitics-regulation-analyst: The numeric bar and current weekly volume are absent, so there is no observable cushion to extrapolate. Literal resolution and ambiguity against the forecaster materially reduce the chance of success, especially because September 30 cuts through an ISO week a" } ], "bubbleBasket": null, "daysRemaining": 24, "deadline": "2026-09-30", "lens": "software" } }, { "id": "lq-p65", "weeklyPredictionId": "p65-state-moratorium-copycat", "text": "At least one additional US state announces a moratorium or discretionary-permit pause on data centers by October 31, 2026.", "lens": "power", "deadline": "2026-10-31", "resolutionSource": "State executive order, statute, or PUC order.", "resolutionRule": "Hit on a second state beyond those already counted at prediction time. Local county pauses do not count.", "ambiguityPolicy": "resolves_against_forecaster", "bubbleBasket": "power_policy", "stated": 0.57, "current": 0.4441517812502465, "direction": "down", "history": [ { "asOf": "2026-07-18", "p": 0.57, "source": "stated", "runId": null }, { "asOf": "2026-09-06", "p": 0.4441517812502465, "source": "engine", "runId": "2026-09-06T22-42-59Z-51a6cc" } ], "engine": { "id": "lq-p65", "weeklyPredictionId": "p65-state-moratorium-copycat", "stated": 0.57, "referenceClass": { "p": 0.65, "className": "qtype:threshold", "n": 19, "note": "most specific ledger class with n >= 8; Resolved predictions of type threshold: disclosure = requires a named primary document or named company; threshold = numeric bar without a named-disclosure requirement; timing = event-by-date without a numeric bar; other = none of the above." }, "calibrated": { "p": 0.5425330871037243, "method": "hierarchical-beta-binomial-by-lens", "ci80": [ 0.5002008247697138, 0.5847045598065381 ], "variants": { "platt-on-stated-logit": 0.4819 } }, "model": { "p": 0.7164321442510658, "method": "logistic-features-l2", "ci80": [ 0.4657577831307773, 0.9247266758775311 ], "features": { "conjuncts": 1, "hasNumericThreshold": true, "horizonDays": 105, "isMarketPriceQuestion": false, "isTimingQuestion": true, "requiresNamedPrimaryDisclosure": false, "statedConfidence": 0.57, "thresholdMagnitudeLog": 0, "weekIndex": 12 }, "gbm": 0.6157 }, "council": { "p": 0.28460809633010153, "aggregation": "log-odds-mean", "extremization": 1, "votes": [ { "persona": "grid-and-power-economist", "model": "openai/gpt-5.6-sol", "p": 0.22, "rationale": "Statewide data-center moratoria are uncommon; most constraints emerge through local zoning, utility tariffs, or project-specific interconnection decisions, none of which necessarily qualify. Only 55 days remain, most legislatures are out of regular session, and the rule requires a formal executive order, statute, or PUC order from an additional state. A governor or commission could still act quickly amid reliability or ratepayer concerns, but ordinary tariff investigations or proposed bills would fail the literal standard. The forecaster’s 0.50–0.59 band has also substantially underperformed its stated confidence.", "keyEvidence": [], "whatWouldChangeMyMind": "A scheduled PUC vote on an explicit pause, enacted emergency legislation, or a governor publicly committing to a statewide moratorium before October 31." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.31, "rationale": "The qualifying bar is unusually strict: a second state must issue a statewide executive order, statute, or PUC order, while local moratoria and ambiguous permitting reviews fail. Grid-load concerns create political pressure, but most state legislatures are out of regular session and formal PUC proceedings rarely produce broad pauses within 55 days. The original 0.57 also sits in a historically overconfident calibration band; roughly half its forecast window has elapsed without a cited qualifying action. Executive action remains the most plausible path, keeping the probability material but below even odds.", "keyEvidence": [], "whatWouldChangeMyMind": "A governor publicly preparing an order, a special-session bill with leadership support, or a pending PUC docket explicitly proposing a statewide data-center interconnection or permitting pause." }, { "persona": "capital-markets-credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.3, "rationale": "Power-system strain and public concern over data-center costs create a plausible catalyst, but the qualifying bar is unusually strict: a second state must issue an executive order, statute, or PUC order—not merely introduce legislation, open an inquiry, revise a tariff, or adopt a local pause. With only 55 days remaining, most state legislatures out of regular session, and ambiguity resolving against the forecast, the procedural window is narrow. Formal statewide moratoria are also much rarer than county restrictions. This is below the forecaster’s 0.57 and consistent with the weak realized performance of its 0.50–0.59 confidence band.", "keyEvidence": [], "whatWouldChangeMyMind": "A pending governor-backed bill awaiting signature, a noticed PUC vote explicitly proposing a data-center pause, or a draft executive order reported by an official state source." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.29, "rationale": "Base rate: 0.534 from the overall ledger; I discount the 0.700 power rate because n=4. The closest calibration bin (0.5–0.6) realized only 0.367 versus 0.586 stated, indicating overconfidence. A 55-day window and the requirement for a second state plus a qualifying statewide executive order, statute, or PUC order materially reduce the chance; county actions and proposals fail. Long interconnection timelines create genuine political pressure, but the digest provides no state-specific pending instrument. With many legislatures outside regular session, executive or PUC action is the likeliest route. Ambiguity resolving against the forecaster lowers the estimate further.", "keyEvidence": [], "whatWouldChangeMyMind": "A publicly docketed PUC order, introduced fast-track statute, or governor-announced executive-order process in a not-yet-counted state would raise my estimate substantially." }, { "persona": "geopolitics-regulation-analyst", "model": "openai/gpt-5.6-sol", "p": 0.31, "rationale": "Only 55 days remain, and most state legislatures are adjourned or focused on elections, leaving executive and PUC action as the realistic paths. Statewide data-center moratoria are uncommon; recent pressure more often produces studies, tariff proceedings, interconnection rules, or local pauses, none of which necessarily satisfy the strict resolution rule. Nationwide exposure and the possibility of action clustering after a grid or political flashpoint keep the probability material. Conditioning the original 0.57 forecast on no qualifying second-state action by September 6, then discounting for document-type ambiguity and the forecaster’s weak calibration in the 0.50–0.59 band, supports a substantially lower estimate.", "keyEvidence": [], "whatWouldChangeMyMind": "A pending governor order, enacted bill awaiting signature, or PUC agenda item explicitly proposing a statewide data-center permitting pause would raise this sharply; confirmation that all active proposals are local or merely studies would lower it." } ], "dissent": null, "spread": 0.09 }, "final": { "p": 0.4441517812502465, "method": "stacked-fixed-weights (resolved n=57 < 80)", "weights": { "calibrated": 0.3, "council": 0.5, "model": 0.2 }, "ci80": [ 0.3407331155463447, 0.5508856655334965 ] }, "direction": "down", "drivers": [ "Reference class qtype:threshold runs 0.65 (n=19) vs stated 0.57.", "Hierarchical calibration moves stated 0.57 to 0.54 (lens power, kappa_stated=24.0, kappa_lens=200.0).", "Feature model at 0.72: conjuncts=1, named disclosure=False, horizon=105d.", "Council of 5 at 0.28." ], "warnings": [ "all 5 votes came from one vendor (openai); extremization disabled, persona diversity only" ], "evidence": [], "bubbleBasket": "power_policy", "daysRemaining": 55, "deadline": "2026-10-31", "lens": "power" } }, { "id": "lq-p72", "weeklyPredictionId": "p72-harness-memory-reproduction-oct31", "text": "An independent party reproduces at least a 15-point ARC-AGI-3 improvement from harness memory by October 31, 2026.", "lens": "software", "deadline": "2026-10-31", "resolutionSource": "ARC Prize or named independent write-up with numbers.", "resolutionRule": "Hit requires an independent reproduction and a numeric improvement ≥15 points. Lab self-report misses.", "ambiguityPolicy": "resolves_against_forecaster", "stated": 0.44, "current": 0.3911556721212139, "direction": "down", "history": [ { "asOf": "2026-08-01", "p": 0.44, "source": "stated", "runId": null }, { "asOf": "2026-09-06", "p": 0.3911556721212139, "source": "engine", "runId": "2026-09-06T22-42-59Z-51a6cc" } ], "engine": { "id": "lq-p72", "weeklyPredictionId": "p72-harness-memory-reproduction-oct31", "stated": 0.44, "referenceClass": { "p": 0.65, "className": "qtype:threshold", "n": 19, "note": "most specific ledger class with n >= 8; Resolved predictions of type threshold: disclosure = requires a named primary document or named company; threshold = numeric bar without a named-disclosure requirement; timing = event-by-date without a numeric bar; other = none of the above." }, "calibrated": { "p": 0.5187604256542997, "method": "hierarchical-beta-binomial-by-lens", "ci80": [ 0.4777619622309755, 0.5596926460069008 ], "variants": { "platt-on-stated-logit": 0.4087 } }, "model": { "p": 0.3687104038686994, "method": "logistic-features-l2", "ci80": [ 0.216872898056272, 0.6095638471546682 ], "features": { "conjuncts": 1, "hasNumericThreshold": true, "horizonDays": 91, "isMarketPriceQuestion": false, "isTimingQuestion": true, "requiresNamedPrimaryDisclosure": false, "statedConfidence": 0.44, "thresholdMagnitudeLog": 0, "weekIndex": 14 }, "gbm": 0.6115 }, "council": { "p": 0.3285260937336452, "aggregation": "log-odds-mean", "extremization": 1, "votes": [ { "persona": "grid-and-power-economist", "model": "openai/gpt-5.6-sol", "p": 0.34, "rationale": "Software results can be replicated quickly, but this requires three things within 55 days: access to the memory method, an independent implementation, and a public numeric ARC-AGI-3 gain of at least 15 points. Benchmark-harness improvements are especially sensitive to prompts, budgets, model versions, and scoring details. A replication showing a smaller gain, lacking comparable baselines, or remaining an informal lab report would miss under the strict rule. The software reference rate supports a meaningful chance, but the large threshold, short remaining window, and documentation requirement put this below the forecaster’s 0.44.", "keyEvidence": [ "https://arcprize.org/" ], "whatWouldChangeMyMind": "A public repository with fixed ARC-AGI-3 evaluation settings and multiple unaffiliated teams actively reporting preliminary gains near 15 points would raise my estimate substantially." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.31, "rationale": "The hit requires three things within 55 days: an independent replication, the same ARC-AGI-3 comparison, and a documented gain of at least 15 points. Software replication can move quickly when code and artifacts are complete, but ARC results are unusually sensitive to model, harness, budget, and baseline choices; a 15-point delta is also large. The supplied digest contains no qualifying independent result, and the ambiguity policy penalizes incomplete or nonnumeric reports. I therefore place this below the forecaster’s earlier 0.44 despite the broader software reference rate.", "keyEvidence": [ "https://arcprize.org/" ], "whatWouldChangeMyMind": "Public runnable artifacts, a fixed baseline and evaluation protocol, or an independent preliminary replication already approaching a 15-point gain would raise the probability materially." }, { "persona": "capital-markets-credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.36, "rationale": "The remaining 55 days are enough for implementation and evaluation if harness memory is clearly specified and benchmark access is straightforward. However, resolution requires more than a promising self-report: an independent party must publish a baseline-controlled numeric gain meeting the full 15-point threshold. The supplied digest shows active model releases but no direct evidence of an ongoing reproduction or scheduled ARC Prize validation. Reporting and attribution are additional failure modes, and ambiguity resolves against the event. The software reference rate supports a material chance, but the exact threshold and independence requirements warrant a discount.", "keyEvidence": [ "https://arcprize.org/" ], "whatWouldChangeMyMind": "A public independent repository with preliminary ARC-AGI-3 runs near the threshold, or ARC Prize announcing third-party validation, would raise the estimate substantially." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.26, "rationale": "I use the software reference-class base rate of 0.452. Conditioning on no qualifying public reproduction yet and only 55 days remaining lowers the chance materially. Resolution also requires a conjunction: an independent party must reproduce the effect, reach the unusually large 15-point threshold, report numbers, and do so by the hard date; ambiguity counts as failure. General open-model activity supports some experimentation but is not evidence specific to harness memory or ARC-AGI-3. The forecaster’s low-confidence calibration bin has only one case, while broader results show some overconfidence, so 0.44 receives no upward calibration adjustment.", "keyEvidence": [ "https://arcprize.org/" ], "whatWouldChangeMyMind": "A public repository with runnable harness-memory code, ARC-AGI-3 baselines, and preliminary results from an unaffiliated evaluator would raise this sharply; failed replication attempts would lower it." }, { "persona": "geopolitics-regulation-analyst", "model": "openai/gpt-5.6-sol", "p": 0.38, "rationale": "The software reference rate is 0.452, but only 55 days remain and the supplied digest contains no resolution-grade independent result. Reproducing a harness-level memory intervention can be faster than training a model, and a large gain is plausible on an interactive benchmark where state retention matters. Still, the rule requires both independence and an explicitly reported improvement of at least 15 points; code replication without comparable scoring, a lab-associated report, or ambiguous baselines all miss. The elapsed time since the 0.44 forecast therefore warrants a modest downward update.", "keyEvidence": [ "https://arcprize.org/" ], "whatWouldChangeMyMind": "Public runnable artifacts, fixed baseline details, and an independent group announcing an underway ARC-AGI-3 replication would raise it; access restrictions or harness-version incompatibility would lower it." } ], "dissent": null, "spread": 0.12 }, "final": { "p": 0.3911556721212139, "method": "stacked-fixed-weights (resolved n=57 < 80)", "weights": { "calibrated": 0.3, "council": 0.5, "model": 0.2 }, "ci80": [ 0.30864018632129253, 0.4790862511096651 ] }, "direction": "down", "drivers": [ "Reference class qtype:threshold runs 0.65 (n=19) vs stated 0.44.", "Hierarchical calibration moves stated 0.44 to 0.52 (lens software, kappa_stated=24.0, kappa_lens=200.0).", "Feature model at 0.37: conjuncts=1, named disclosure=False, horizon=91d.", "Council of 5 at 0.33." ], "warnings": [ "all 5 votes came from one vendor (openai); extremization disabled, persona diversity only" ], "evidence": [ { "url": "https://arcprize.org/", "title": "ARC Prize", "publishedAt": null, "source": "direct", "excerpt": "Cited by grid-and-power-economist: Software results can be replicated quickly, but this requires three things within 55 days: access to the memory method, an independent implementation, and a public numeric ARC-AGI-3 gain of at least 15 points. Benchmark-harness improvements are especially sensitive" } ], "bubbleBasket": null, "daysRemaining": 55, "deadline": "2026-10-31", "lens": "software" } }, { "id": "lq-p73", "weeklyPredictionId": "p73-hbm-2027-committed-oct31", "text": "SK hynix or Samsung states in a primary release that 2027 HBM capacity is substantially committed by October 31, 2026.", "lens": "hardware", "deadline": "2026-10-31", "resolutionSource": "SK hynix or Samsung earnings/release.", "resolutionRule": "Hit on 'sold out' or 'substantially committed' language for 2027 HBM. 'Under negotiation' is a miss.", "ambiguityPolicy": "partial_credit_if_one_leg_hits", "stated": 0.81, "current": 0.6605755437592185, "direction": "down", "history": [ { "asOf": "2026-08-01", "p": 0.81, "source": "stated", "runId": null }, { "asOf": "2026-09-06", "p": 0.6605755437592185, "source": "engine", "runId": "2026-09-06T22-42-59Z-51a6cc" } ], "engine": { "id": "lq-p73", "weeklyPredictionId": "p73-hbm-2027-committed-oct31", "stated": 0.81, "referenceClass": { "p": 0.5, "className": "qtype:disclosure", "n": 29, "note": "most specific ledger class with n >= 8; Resolved predictions of type disclosure: disclosure = requires a named primary document or named company; threshold = numeric bar without a named-disclosure requirement; timing = event-by-date without a numeric bar; other = none of the above." }, "calibrated": { "p": 0.5676167112696997, "method": "hierarchical-beta-binomial-by-lens", "ci80": [ 0.5261956724716945, 0.6087908543502055 ], "variants": { "platt-on-stated-logit": 0.6434 } }, "model": { "p": 0.671539976586375, "method": "logistic-features-l2", "ci80": [ 0.5257209750599534, 0.8660985833689716 ], "features": { "conjuncts": 1, "hasNumericThreshold": false, "horizonDays": 91, "isMarketPriceQuestion": false, "isTimingQuestion": true, "requiresNamedPrimaryDisclosure": true, "statedConfidence": 0.81, "thresholdMagnitudeLog": 0, "weekIndex": 14 }, "gbm": 0.7656 }, "council": { "p": 0.7073144342112148, "aggregation": "log-odds-mean", "extremization": 1, "votes": [ { "persona": "grid-and-power-economist", "model": "openai/gpt-5.6-sol", "p": 0.74, "rationale": "HBM supply is allocated well ahead because advanced-memory fabrication and packaging capacity cannot adjust quickly. By late October, both companies should have quarterly earnings releases and enough 2027 customer visibility to discuss commitments. SK hynix has historically used strong sold-out language, making it the likelier qualifying issuer; Samsung provides a second opportunity. However, the rule is wording-sensitive: statements about negotiations, demand visibility, or expected allocation would miss. Disclosure questions in the ledger resolve near 0.5, and the forecaster’s 0.8–0.9 band has realized only 0.5, so 0.81 appears too high despite favorable industry conditions.", "keyEvidence": [ "https://news.skhynix.com/" ], "whatWouldChangeMyMind": "A scheduled October earnings release explicitly previewing 2027 allocation would raise it; evidence that 2027 contracting remains under negotiation or that releases fall after October 31 would lower it." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.69, "rationale": "AI-system demand and customer prepayments support early HBM reservations, while the deadline likely captures both vendors’ third-quarter reporting. SK hynix has historically used sold-out or fully booked language earlier and more directly than Samsung, making it the likelier route to a hit. However, 2027 supply depends heavily on HBM4/HBM4E qualification, yields, and advanced-packaging availability; vendors may therefore describe customer discussions or demand visibility without declaring capacity committed. The strict wording and primary-release requirement make 0.81 too high despite strong industry conditions.", "keyEvidence": [ "https://news.skhynix.com/" ], "whatWouldChangeMyMind": "Evidence of signed 2027 allocation agreements would raise it; qualification delays, added capacity, or guidance limited to negotiations would lower it." }, { "persona": "capital-markets-credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.72, "rationale": "SK hynix’s customary late-October earnings cadence creates a likely qualifying disclosure window before the deadline; Samsung may also publish full third-quarter results by then. Strong AI infrastructure demand, including NVDA RPO growth of 68% in the digest, supports advance HBM allocation. SK hynix has historically used explicit sold-out language, making it the more likely qualifying issuer. Still, the rule is wording-sensitive: statements about strong demand, customer discussions, or negotiations fail, and 2027 allocations could remain fluid across HBM generations. The issuers’ outcomes are also highly correlated, limiting diversification. I place this above the hardware/disclosure base rates but below the forecaster’s 0.81 given their weak realized record in that confidence band.", "keyEvidence": [ "https://news.skhynix.com/" ], "whatWouldChangeMyMind": "A scheduled October earnings release explicitly previewing 2027 customer allocations would raise it; delayed results, weaker AI-memory guidance, or language emphasizing ongoing negotiations would lower it." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.64, "rationale": "Base rate: 0.50 for disclosure questions, cross-checked against 0.615 for hardware and 0.568 for single-leg questions. I adjust upward because either of two major HBM suppliers can qualify and the deadline plausibly captures a quarterly release cycle; strong semiconductor demand indicators also make substantial 2027 commitments credible. I adjust downward for the strict primary-release and wording requirements: negotiations, reported customer interest, or secondary sourcing fail. The forecaster’s 0.81 is too high given its 0.8–0.9 calibration bin realized only 0.50, albeit on four cases.", "keyEvidence": [], "whatWouldChangeMyMind": "An official SK hynix or Samsung release explicitly saying most, substantially all, or all 2027 HBM capacity is committed would sharply raise it; releases limited to negotiations would lower it." }, { "persona": "geopolitics-regulation-analyst", "model": "openai/gpt-5.6-sol", "p": 0.74, "rationale": "SK hynix has a strong precedent of announcing that next-year HBM output is sold out or nearly allocated well before year-end, and either company’s late-October third-quarter release provides a natural disclosure point before the deadline. Strong AI-system demand and supplier lead times make substantial 2027 commitments plausible. I remain below 0.81 because the literal wording matters: management may describe customer negotiations, demand visibility, or allocation planning without saying capacity is sold out or substantially committed, and Samsung’s qualification and product-mix uncertainty further reduces the chance of qualifying language.", "keyEvidence": [ "https://news.skhynix.com/" ], "whatWouldChangeMyMind": "A scheduled October earnings date after the deadline, or primary guidance emphasizing ongoing 2027 negotiations, would lower it. A release saying 2027 HBM orders are fully booked would resolve it." } ], "dissent": null, "spread": 0.1 }, "final": { "p": 0.6605755437592185, "method": "stacked-fixed-weights (resolved n=57 < 80)", "weights": { "calibrated": 0.3, "council": 0.5, "model": 0.2 }, "ci80": [ 0.5841275217878502, 0.7367283796425979 ] }, "direction": "down", "drivers": [ "Reference class qtype:disclosure runs 0.50 (n=29) vs stated 0.81.", "Hierarchical calibration moves stated 0.81 to 0.57 (lens hardware, kappa_stated=24.0, kappa_lens=200.0).", "Feature model at 0.67: conjuncts=1, named disclosure=True, horizon=91d.", "Council of 5 at 0.71." ], "warnings": [ "all 5 votes came from one vendor (openai); extremization disabled, persona diversity only", "4 cited URL(s) failed verification and were dropped." ], "evidence": [ { "url": "https://news.skhynix.com/", "title": "SK hynix Newsroom", "publishedAt": null, "source": "direct", "excerpt": "Cited by grid-and-power-economist: HBM supply is allocated well ahead because advanced-memory fabrication and packaging capacity cannot adjust quickly. By late October, both companies should have quarterly earnings releases and enough 2027 customer visibility to discuss commitments. SK hynix has his" } ], "bubbleBasket": null, "daysRemaining": 55, "deadline": "2026-10-31", "lens": "hardware" } }, { "id": "lq-p77", "weeklyPredictionId": "p77-endpoint-index-expansion-oct31", "text": "Artificial Analysis publishes Endpoint Accuracy Index results covering at least the stated expansion by October 31, 2026.", "lens": "software", "deadline": "2026-10-31", "resolutionSource": "Artificial Analysis publication.", "resolutionRule": "Hit if the published index coverage meets the weekly trigger's stated expansion. Partial pages miss.", "ambiguityPolicy": "resolves_against_forecaster", "stated": 0.83, "current": 0.54403175087588, "direction": "down", "history": [ { "asOf": "2026-08-08", "p": 0.83, "source": "stated", "runId": null }, { "asOf": "2026-09-06", "p": 0.54403175087588, "source": "engine", "runId": "2026-09-06T22-42-59Z-51a6cc" } ], "engine": { "id": "lq-p77", "weeklyPredictionId": "p77-endpoint-index-expansion-oct31", "stated": 0.83, "referenceClass": { "p": 0.65, "className": "qtype:threshold", "n": 19, "note": "most specific ledger class with n >= 8; Resolved predictions of type threshold: disclosure = requires a named primary document or named company; threshold = numeric bar without a named-disclosure requirement; timing = event-by-date without a numeric bar; other = none of the above." }, "calibrated": { "p": 0.5571210813920046, "method": "hierarchical-beta-binomial-by-lens", "ci80": [ 0.5162955940008747, 0.5977448750578769 ], "variants": { "platt-on-stated-logit": 0.6608 } }, "model": { "p": 0.5086000736466604, "method": "logistic-features-l2", "ci80": [ 0.2681618388701692, 0.7596374649414853 ], "features": { "conjuncts": 1, "hasNumericThreshold": true, "horizonDays": 84, "isMarketPriceQuestion": false, "isTimingQuestion": true, "requiresNamedPrimaryDisclosure": false, "statedConfidence": 0.83, "thresholdMagnitudeLog": 0, "weekIndex": 15 }, "gbm": 0.7422 }, "council": { "p": 0.550258619807539, "aggregation": "log-odds-mean", "extremization": 1, "votes": [ { "persona": "grid-and-power-economist", "model": "openai/gpt-5.6-sol", "p": 0.58, "rationale": "A 55-day window is sufficient for an established benchmark publisher to add and publish endpoint evaluations, but the resolution standard requires the complete stated expansion; partial pages explicitly miss. No digest evidence confirms that the expanded coverage is already live, in testing, or scheduled. Relevant ledger rates cluster around 0.45 for software and 0.62–0.65 for this horizon and threshold structure. The forecaster’s 0.83 confidence band has realized only 0.50, albeit on four cases. I therefore place this near the middle of those reference classes rather than at 0.83.", "keyEvidence": [], "whatWouldChangeMyMind": "A dated Artificial Analysis announcement listing every required endpoint, or a live index page already containing the full stated expansion, would raise the estimate substantially." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.58, "rationale": "A 55-day window is sufficient for a software-index expansion, and Artificial Analysis can publish without external hardware or OEM dependencies. However, the supplied digest gives no direct evidence of implementation progress, preview results, or a dated commitment. The resolution rule requires complete coverage; partial pages explicitly miss, increasing execution and scope risk. Relevant ledger rates cluster around 0.45 for software and 0.62–0.65 for this horizon/threshold type. The forecaster’s 0.8–0.9 band has realized only 0.50, so 0.83 appears too aggressive.", "keyEvidence": [ "https://artificialanalysis.ai/" ], "whatWouldChangeMyMind": "A dated Artificial Analysis preview showing every required endpoint already tested would raise it sharply; evidence that required providers remain unsupported or methodology is unfinished would lower it." }, { "persona": "capital-markets-credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.57, "rationale": "Artificial Analysis regularly updates public model and provider benchmarks, and continued model-release activity supports demand for broader endpoint coverage. However, the digest gives no direct evidence that the specific expansion is already live, announced, or scheduled. The exact required expansion is also not reproduced here; because partial pages miss and ambiguity resolves against the forecaster, an ordinary incremental update may not qualify. With only 55 days remaining, I place this near the relevant threshold/horizon base rates but well below the forecaster’s 0.83, especially given weak historical outcomes in its 0.8–0.9 confidence band.", "keyEvidence": [ "https://artificialanalysis.ai/" ], "whatWouldChangeMyMind": "A dated Artificial Analysis announcement, changelog, or visible index fields showing that every element of the required expansion is implemented would raise this sharply." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.45, "rationale": "I use the software-ledger base rate of 0.452, with the 31–90-day and threshold classes offering modest upward support. However, the supplied record does not specify the required expansion or show a primary commitment from Artificial Analysis. The rule requires complete published index coverage by a hard date; partial pages miss and ambiguity resolves against the forecast. Publication timing is an additional failure mode. The forecaster’s 0.8–0.9 confidence bin has realized only 0.50 across four cases, so 0.83 appears overconfident absent specific evidence.", "keyEvidence": [ "https://artificialanalysis.ai/" ], "whatWouldChangeMyMind": "A dated Artificial Analysis announcement defining the expansion and committing to full Endpoint Accuracy Index publication before October 31, or visible complete coverage well ahead of the deadline." }, { "persona": "geopolitics-regulation-analyst", "model": "openai/gpt-5.6-sol", "p": 0.57, "rationale": "The 55-day window is plausible for a software-index update, and the threshold and 31–90-day reference classes are relatively favorable. However, the software base rate is only 0.452, while the forecaster’s 0.8–0.9 confidence bin has realized just 0.5. Resolution requires complete published coverage—not an announcement, preview, or partial page—and ambiguity resolves against the forecast. The supplied record also does not specify the exact expansion, increasing verification risk. I therefore discount the stated 0.83 substantially.", "keyEvidence": [ "https://artificialanalysis.ai/" ], "whatWouldChangeMyMind": "A dated Artificial Analysis announcement naming the full required endpoint expansion and committing to publication before October 31 would raise the estimate; evidence of phased or delayed coverage would lower it." } ], "dissent": null, "spread": 0.13 }, "final": { "p": 0.54403175087588, "method": "stacked-fixed-weights (resolved n=57 < 80)", "weights": { "calibrated": 0.3, "council": 0.5, "model": 0.2 }, "ci80": [ 0.43006953637252837, 0.6248881765037753 ] }, "direction": "down", "drivers": [ "Reference class qtype:threshold runs 0.65 (n=19) vs stated 0.83.", "Hierarchical calibration moves stated 0.83 to 0.56 (lens software, kappa_stated=24.0, kappa_lens=200.0).", "Feature model at 0.51: conjuncts=1, named disclosure=False, horizon=84d.", "Council of 5 at 0.55." ], "warnings": [ "all 5 votes came from one vendor (openai); extremization disabled, persona diversity only" ], "evidence": [ { "url": "https://artificialanalysis.ai/", "title": "AI Model & API Providers Analysis | Artificial Analysis", "publishedAt": null, "source": "direct", "excerpt": "Cited by semiconductor-supply-chain-analyst: A 55-day window is sufficient for a software-index expansion, and Artificial Analysis can publish without external hardware or OEM dependencies. However, the supplied digest gives no direct evidence of implementation progress, preview results, or a dated " } ], "bubbleBasket": null, "daysRemaining": 55, "deadline": "2026-10-31", "lens": "software" } }, { "id": "lq-p91", "weeklyPredictionId": "p91-anthropic-skills-case-study-oct31", "text": "Anthropic or a named enterprise customer publishes an independently attributed Skills case study by October 31, 2026.", "lens": "software", "deadline": "2026-10-31", "resolutionSource": "Anthropic or named-customer publication.", "resolutionRule": "Hit requires independent attribution (customer-named outcomes), not an Anthropic-only anecdote.", "ambiguityPolicy": "resolves_against_forecaster", "stated": 0.58, "current": 0.4378640000045455, "direction": "down", "history": [ { "asOf": "2026-08-22", "p": 0.58, "source": "stated", "runId": null }, { "asOf": "2026-09-06", "p": 0.4378640000045455, "source": "engine", "runId": "2026-09-06T22-42-59Z-51a6cc" } ], "engine": { "id": "lq-p91", "weeklyPredictionId": "p91-anthropic-skills-case-study-oct31", "stated": 0.58, "referenceClass": { "p": 0.5, "className": "qtype:disclosure", "n": 29, "note": "most specific ledger class with n >= 8; Resolved predictions of type disclosure: disclosure = requires a named primary document or named company; threshold = numeric bar without a named-disclosure requirement; timing = event-by-date without a numeric bar; other = none of the above." }, "calibrated": { "p": 0.5325309174575783, "method": "hierarchical-beta-binomial-by-lens", "ci80": [ 0.49156641716087535, 0.5733805511859074 ], "variants": { "platt-on-stated-logit": 0.4876 } }, "model": { "p": 0.3372076236511158, "method": "logistic-features-l2", "ci80": [ 0.20007713717385403, 0.558119186759893 ], "features": { "conjuncts": 1, "hasNumericThreshold": false, "horizonDays": 70, "isMarketPriceQuestion": false, "isTimingQuestion": true, "requiresNamedPrimaryDisclosure": true, "statedConfidence": 0.58, "thresholdMagnitudeLog": 0, "weekIndex": 17 }, "gbm": 0.6725 }, "council": { "p": 0.4237127771080126, "aggregation": "log-odds-mean", "extremization": 1, "votes": [ { "persona": "grid-and-power-economist", "model": "openai/gpt-5.6-sol", "p": 0.43, "rationale": "The 55-day window permits a marketing publication, and Anthropic routinely has named enterprise relationships. But the hit is narrower than a generic customer story: it must specifically concern Skills and provide customer-attributed outcomes. Product announcements, anonymous examples, quotes without outcomes, or Anthropic-only anecdotes fail under the ambiguity rule. The software and disclosure reference classes center near 0.45–0.50, while the forecaster’s 0.50–0.59 band has materially underperformed. With no supplied evidence of a scheduled case study or customer commitment, 0.58 is too high.", "keyEvidence": [ "https://www.anthropic.com/customers" ], "whatWouldChangeMyMind": "A dated preview, event agenda, or named customer statement promising publication before October 31 would raise it; evidence that no customer has completed a Skills deployment would lower it." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.43, "rationale": "The 55-day window is long enough for a product-marketing case study, and Anthropic has incentives to demonstrate enterprise adoption of Skills. However, the resolution bar is materially narrower than a launch post, testimonial, or Anthropic-authored usage anecdote: it requires a named customer and independently attributed outcomes. The supplied evidence contains no scheduled publication, identified customer, or imminent joint announcement. Relevant software and disclosure reference classes center near 0.45–0.50, while the forecaster’s 0.50–0.59 band has historically underperformed its stated confidence. Strict adverse ambiguity further lowers the estimate below 0.58.", "keyEvidence": [], "whatWouldChangeMyMind": "A named customer previewing measurable Skills results, an Anthropic event agenda promising a joint customer presentation, or a customer-authored draft/publication would raise the probability substantially." }, { "persona": "capital-markets-credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.45, "rationale": "The software and disclosure reference classes center near 0.45–0.50, while the stronger 31–90-day rate is less informative about this unusually specific content requirement. A qualifying customer story is plausible, but publication timing is discretionary rather than filing-driven. The rule excludes product announcements, demos, unnamed adoption claims, and Anthropic-only anecdotes; obtaining a named enterprise’s approval and attributable outcomes can take longer than 55 days. With no supplied evidence of a scheduled Skills customer proof point, 0.58 appears optimistic, especially given the forecaster’s weak realized rate in that confidence band.", "keyEvidence": [], "whatWouldChangeMyMind": "A dated customer-story teaser, event agenda naming a Skills user, or a named enterprise quote attributing measurable outcomes specifically to Skills would materially increase the estimate." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.37, "rationale": "I use the lens:software base rate of 0.452, cross-checked against the 0.500 disclosure base rate. The 55-day window helps, but resolution requires more than product promotion: a named enterprise must independently attribute outcomes specifically to Skills. Case-study approval and measurable-outcome disclosure often lag adoption, and ambiguity counts as failure. The forecaster’s 0.50–0.59 predictions realized only 0.367 overall, indicating material overconfidence around the stated 0.58. These factors lower my estimate below the software base rate.", "keyEvidence": [], "whatWouldChangeMyMind": "A scheduled Anthropic customer event, draft case study, or named enterprise announcement explicitly promising quantified Skills outcomes before October 31 would raise it substantially." }, { "persona": "geopolitics-regulation-analyst", "model": "openai/gpt-5.6-sol", "p": 0.44, "rationale": "The 55-day window is adequate for a product-marketing publication, and enterprise AI vendors frequently coordinate customer stories with fall announcements. However, the rule is narrower than a routine testimonial: it requires a named customer, outcomes specifically attributable to Skills, and publication by Anthropic or that customer. Generic adoption quotes, launch-partner mentions, or Anthropic-authored anecdotes would fail. The most relevant ledger classes cluster around 0.45–0.50, while the forecaster’s 0.50–0.59 band has materially underperformed. Strict ambiguity treatment and the limited window therefore put this below even odds.", "keyEvidence": [ "https://www.anthropic.com/customers" ], "whatWouldChangeMyMind": "A scheduled Anthropic enterprise event featuring Skills customers, or a named customer previewing quantified Skills results, would raise the estimate; no qualifying publication by mid-October would lower it sharply." } ], "dissent": null, "spread": 0.08 }, "final": { "p": 0.4378640000045455, "method": "stacked-fixed-weights (resolved n=57 < 80)", "weights": { "calibrated": 0.3, "council": 0.5, "model": 0.2 }, "ci80": [ 0.36507776737669045, 0.508766174824044 ] }, "direction": "down", "drivers": [ "Reference class qtype:disclosure runs 0.50 (n=29) vs stated 0.58.", "Hierarchical calibration moves stated 0.58 to 0.53 (lens software, kappa_stated=24.0, kappa_lens=200.0).", "Feature model at 0.34: conjuncts=1, named disclosure=True, horizon=70d.", "Council of 5 at 0.42." ], "warnings": [ "all 5 votes came from one vendor (openai); extremization disabled, persona diversity only" ], "evidence": [ { "url": "https://www.anthropic.com/customers", "title": "Customer Stories | Claude by Anthropic", "publishedAt": null, "source": "direct", "excerpt": "Cited by grid-and-power-economist: The 55-day window permits a marketing publication, and Anthropic routinely has named enterprise relationships. But the hit is narrower than a generic customer story: it must specifically concern Skills and provide customer-attributed outcomes. Product announcements" } ], "bubbleBasket": null, "daysRemaining": 55, "deadline": "2026-10-31", "lens": "software" } }, { "id": "lq-p95", "weeklyPredictionId": "p95-jalapeno-agentx-oct31", "text": "SemiAnalysis publishes AgentX v3 multi-turn results for OpenAI Jalapeño by October 31, 2026.", "lens": "hardware", "deadline": "2026-10-31", "resolutionSource": "SemiAnalysis.", "resolutionRule": "Hit if AgentX v3 includes Jalapeño by the deadline. Other benches do not substitute.", "ambiguityPolicy": "resolves_against_forecaster", "stated": 0.36, "current": 0.439706027230459, "direction": "up", "history": [ { "asOf": "2026-08-29", "p": 0.36, "source": "stated", "runId": null }, { "asOf": "2026-09-06", "p": 0.439706027230459, "source": "engine", "runId": "2026-09-06T22-42-59Z-51a6cc" } ], "engine": { "id": "lq-p95", "weeklyPredictionId": "p95-jalapeno-agentx-oct31", "stated": 0.36, "referenceClass": { "p": 0.3888888888888889, "className": "qtype:timing", "n": 8, "note": "most specific ledger class with n >= 8; Resolved predictions of type timing: disclosure = requires a named primary document or named company; threshold = numeric bar without a named-disclosure requirement; timing = event-by-date without a numeric bar; other = none of the above." }, "calibrated": { "p": 0.5218539994052929, "method": "hierarchical-beta-binomial-by-lens", "ci80": [ 0.4801716961012243, 0.5634565045123754 ], "variants": { "platt-on-stated-logit": 0.3638 } }, "model": { "p": 0.6324948922554998, "method": "logistic-features-l2", "ci80": [ 0.416303101432388, 0.828819913275697 ], "features": { "conjuncts": 1, "hasNumericThreshold": false, "horizonDays": 63, "isMarketPriceQuestion": false, "isTimingQuestion": true, "requiresNamedPrimaryDisclosure": false, "statedConfidence": 0.36, "thresholdMagnitudeLog": 0, "weekIndex": 18 }, "gbm": 0.6746 }, "council": { "p": 0.3198704042101956, "aggregation": "log-odds-mean", "extremization": 1, "votes": [ { "persona": "grid-and-power-economist", "model": "openai/gpt-5.6-sol", "p": 0.33, "rationale": "This requires an exact publication outcome: Jalapeño must become benchmarkable and SemiAnalysis must publish it specifically in AgentX v3 within 55 days. A model announcement, API release, preview, or result on another benchmark would all miss. The timing reference class is only 0.389, and the additional dependency on SemiAnalysis testing and publication pushes the estimate lower. Hardware and 31–90-day base rates are less diagnostic because they include less document-specific outcomes. The supplied public-series data provides no direct evidence of scheduled Jalapeño availability or an AgentX v3 update.", "keyEvidence": [ "https://semianalysis.com/" ], "whatWouldChangeMyMind": "A dated SemiAnalysis benchmark schedule naming Jalapeño, or public Jalapeño API access plus evidence that AgentX v3 testing is underway, would raise the estimate substantially." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.31, "rationale": "This requires a specific publisher, benchmark version, model name, and publication by a near-term deadline. Jalapeño must become testable early enough for SemiAnalysis to complete multi-turn runs and publish AgentX v3 results; an OpenAI announcement or results on another benchmark would not qualify. The 31–90-day and hardware base rates are favorable, but the more relevant timing base rate is only 0.389, and ambiguity resolves against a hit. With no pertinent public-series evidence of imminent benchmark availability, I place this modestly below the forecaster’s 0.36.", "keyEvidence": [ "https://semianalysis.com/" ], "whatWouldChangeMyMind": "Official Jalapeño availability or a SemiAnalysis statement showing active AgentX v3 testing would raise this sharply; evidence of delayed access or a postponed benchmark update would lower it." }, { "persona": "capital-markets-credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.31, "rationale": "The strict trigger requires SemiAnalysis to publish a specific AgentX v3 result for a specifically named OpenAI model within 55 days. This depends on Jalapeño being available to the benchmark team, AgentX v3 testing being completed, and publication occurring by the deadline. Benchmark-version and model-timing slippage are common, while results on any other benchmark receive no credit. The timing reference class is weak at 0.389, and ambiguity resolves against a hit. Hardware and 31–90-day base rates are less diagnostic than the exact publication requirement.", "keyEvidence": [ "https://semianalysis.com/" ], "whatWouldChangeMyMind": "A SemiAnalysis preview naming Jalapeño in the AgentX v3 test set, or a dated commitment to publish v3 before October 31, would raise the estimate materially." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.31, "rationale": "I use the timing-question base rate of 0.389 (n=8), rather than the broader hardware or 31–90-day rates. I adjust downward because resolution requires one exact benchmark/version and named system to appear in SemiAnalysis’s publication by a hard date; related benchmarks, testing progress, or later publication do not count. Access, successful multi-turn testing, editorial scheduling, and explicit inclusion each add failure risk despite the nominally single-leg classification. The digest provides no specific evidence of a scheduled Jalapeño result. The forecaster’s sub-0.50 calibration sample is only one case, so it supplies no reliable correction.", "keyEvidence": [ "https://semianalysis.com/" ], "whatWouldChangeMyMind": "A dated SemiAnalysis announcement, preview, or live AgentX v3 results page explicitly naming OpenAI Jalapeño and indicating publication by October 31 would raise this substantially." }, { "persona": "geopolitics-regulation-analyst", "model": "openai/gpt-5.6-sol", "p": 0.34, "rationale": "The 55-day window is plausible for a benchmark update, and the ledger’s 31–90-day and hardware base rates are favorable. However, resolution requires the exact combination of AgentX v3, multi-turn results, and Jalapeño—not merely a model launch or another benchmark. This effectively depends on timely model access, completed evaluation, and SemiAnalysis publication before a hard deadline. The timing reference class is only 0.389, and strict ambiguity treatment increases the chance that near-matching coverage fails. With no concrete publication schedule in the supplied evidence, I place it modestly below that timing base rate.", "keyEvidence": [ "https://semianalysis.com/" ], "whatWouldChangeMyMind": "A SemiAnalysis preview naming Jalapeño and AgentX v3, or public Jalapeño availability by early October, would raise it materially; delayed access or a stated later benchmark cycle would lower it." } ], "dissent": null, "spread": 0.03 }, "final": { "p": 0.439706027230459, "method": "stacked-fixed-weights (resolved n=57 < 80)", "weights": { "calibrated": 0.3, "council": 0.5, "model": 0.2 }, "ci80": [ 0.37954974900100713, 0.5150881378232476 ] }, "direction": "up", "drivers": [ "Reference class qtype:timing runs 0.39 (n=8) vs stated 0.36.", "Hierarchical calibration moves stated 0.36 to 0.52 (lens hardware, kappa_stated=24.0, kappa_lens=200.0).", "Feature model at 0.63: conjuncts=1, named disclosure=False, horizon=63d.", "Council of 5 at 0.32." ], "warnings": [ "all 5 votes came from one vendor (openai); extremization disabled, persona diversity only" ], "evidence": [ { "url": "https://semianalysis.com/", "title": "SemiAnalysis – Bridging the gap between the world's most important industry, semiconductors, and business.", "publishedAt": null, "source": "direct", "excerpt": "Cited by grid-and-power-economist: This requires an exact publication outcome: Jalapeño must become benchmarkable and SemiAnalysis must publish it specifically in AgentX v3 within 55 days. A model announcement, API release, preview, or result on another benchmark would all miss. The timing reference" } ], "bubbleBasket": null, "daysRemaining": 55, "deadline": "2026-10-31", "lens": "hardware" } }, { "id": "lq-p99", "weeklyPredictionId": "p99-metaroce-ocp-spec-oct31", "text": "Meta contributes MetaRoCE specification through OCP at the October 2026 Global Summit.", "lens": "networking", "deadline": "2026-10-31", "resolutionSource": "OCP Summit materials or Meta engineering blog pointing at OCP.", "resolutionRule": "Hit on an OCP contribution at the October 2026 summit. A GitHub dump without OCP miss.", "ambiguityPolicy": "resolves_against_forecaster", "stated": 0.44, "current": 0.4071912075704372, "direction": "down", "history": [ { "asOf": "2026-08-29", "p": 0.44, "source": "stated", "runId": null }, { "asOf": "2026-09-06", "p": 0.4071912075704372, "source": "engine", "runId": "2026-09-06T22-42-59Z-51a6cc" } ], "engine": { "id": "lq-p99", "weeklyPredictionId": "p99-metaroce-ocp-spec-oct31", "stated": 0.44, "referenceClass": { "p": 0.5681818181818182, "className": "conjunctive:single-leg", "n": 43, "note": "most specific ledger class with n >= 8; Resolved predictions that are single-leg: multi-leg means two or more conditions must all hold." }, "calibrated": { "p": 0.5152886489728595, "method": "hierarchical-beta-binomial-by-lens", "ci80": [ 0.4734202248223134, 0.557100767710763 ], "variants": { "platt-on-stated-logit": 0.4087 } }, "model": { "p": 0.3281640504318027, "method": "logistic-features-l2", "ci80": [ 0.10991644595334871, 0.6161179546047796 ], "features": { "conjuncts": 1, "hasNumericThreshold": false, "horizonDays": 63, "isMarketPriceQuestion": false, "isTimingQuestion": false, "requiresNamedPrimaryDisclosure": false, "statedConfidence": 0.44, "thresholdMagnitudeLog": 0, "weekIndex": 18 }, "gbm": 0.4579 }, "council": { "p": 0.3772394306212745, "aggregation": "log-odds-mean", "extremization": 1, "votes": [ { "persona": "grid-and-power-economist", "model": "openai/gpt-5.6-sol", "p": 0.39, "rationale": "Meta’s foundational OCP role and history of contributing open networking designs make the Global Summit a plausible venue for a RoCE specification. However, the resolution requires an identifiable MetaRoCE contribution through OCP, not merely a presentation, general RoCE discussion, or GitHub release. No summit agenda, OCP project record, or Meta engineering post in the supplied evidence confirms that path. With only 55 days remaining and ambiguity resolving against the forecast, the networking reference-class rate deserves substantial weight.", "keyEvidence": [ "https://www.opencompute.org/", "https://engineering.fb.com/" ], "whatWouldChangeMyMind": "An OCP agenda/session listing, project proposal, or Meta engineering preview explicitly naming MetaRoCE and describing an OCP specification contribution would raise the probability sharply." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.4, "rationale": "Meta’s long-standing OCP involvement and incentive to standardize AI-cluster Ethernet make a RoCE-profile contribution plausible. However, the resolution requires a specifically named MetaRoCE contribution at the October summit, documented in OCP materials or a Meta engineering post pointing to OCP. No supplied evidence shows an accepted specification, agenda slot, or public preview. Conference talks, generic RoCE material, or a standalone GitHub release would miss. The networking ledger’s 0.318 base rate and strict ambiguity policy support a modest discount from the forecaster’s 0.44.", "keyEvidence": [ "https://www.opencompute.org/", "https://engineering.fb.com/" ], "whatWouldChangeMyMind": "An OCP agenda, contribution repository entry, or Meta engineering preview explicitly naming MetaRoCE and the October 2026 Global Summit would raise the estimate sharply." }, { "persona": "capital-markets-credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.38, "rationale": "Meta’s longstanding OCP participation and the summit’s relevance to AI-networking standards make a contribution plausible. However, the claim requires the specifically named MetaRoCE specification to be formally associated with OCP at the October summit, not merely discussed or published on GitHub. With only 55 days remaining, a summit-ready specification likely already needs working-group coordination and scheduled materials; none is identified in the supplied evidence. The networking reference-class rate of 0.318 and the against-forecaster ambiguity rule support an estimate below the stated 0.44.", "keyEvidence": [ "https://www.opencompute.org/", "https://engineering.fb.com/" ], "whatWouldChangeMyMind": "An OCP agenda, working-group repository, or Meta engineering post explicitly naming MetaRoCE and tying its specification contribution to the October 2026 Global Summit." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.31, "rationale": "Base rate: 0.318 for the forecaster’s networking questions (n=10). Meta’s longstanding OCP participation and the Global Summit’s relevance make the venue plausible, but no digest evidence specifically supports a MetaRoCE contribution. Resolution requires the named specification to be contributed through OCP at the October summit; an adjacent announcement, later contribution, or standalone GitHub release fails, and ambiguity resolves against the forecast. The 31–90-day rate is higher, but it is less specific than the networking class. Calibration offers no reliable uplift: the 0.4–0.5 bin has only one resolved case, while the broader record shows some overconfidence.", "keyEvidence": [ "https://www.opencompute.org/", "https://engineering.fb.com/" ], "whatWouldChangeMyMind": "An official summit agenda naming MetaRoCE, an OCP project proposal, or a Meta engineering post explicitly promising an October OCP contribution would raise this sharply." }, { "persona": "geopolitics-regulation-analyst", "model": "openai/gpt-5.6-sol", "p": 0.41, "rationale": "Meta is a longstanding OCP contributor, and Global Summit timing is a natural venue for publishing AI-networking specifications. RoCE scaling is strategically relevant, making a Meta-authored contribution plausible. However, the resolution requires the specifically named MetaRoCE specification to be contributed through OCP by October 31—not merely presented, discussed, or released on GitHub. Networking questions in the ledger have a low 0.318 base rate, and the exact document-and-venue requirement creates substantial miss risk. Alternative publication through another consortium or a delayed formal contribution would not count.", "keyEvidence": [ "https://www.opencompute.org/", "https://engineering.fb.com/" ], "whatWouldChangeMyMind": "An OCP summit agenda naming MetaRoCE or a Meta/OCP pre-summit contribution notice would raise this sharply; a complete agenda omitting it or publication through another body would lower it." } ], "dissent": null, "spread": 0.1 }, "final": { "p": 0.4071912075704372, "method": "stacked-fixed-weights (resolved n=57 < 80)", "weights": { "calibrated": 0.3, "council": 0.5, "model": 0.2 }, "ci80": [ 0.2993689970759738, 0.49536490484513784 ] }, "direction": "down", "drivers": [ "Reference class conjunctive:single-leg runs 0.57 (n=43) vs stated 0.44.", "Hierarchical calibration moves stated 0.44 to 0.52 (lens networking, kappa_stated=24.0, kappa_lens=200.0).", "Feature model at 0.33: conjuncts=1, named disclosure=False, horizon=63d.", "Council of 5 at 0.38." ], "warnings": [ "all 5 votes came from one vendor (openai); extremization disabled, persona diversity only" ], "evidence": [ { "url": "https://www.opencompute.org/", "title": "Home » Open Compute Project", "publishedAt": null, "source": "direct", "excerpt": "Cited by grid-and-power-economist: Meta’s foundational OCP role and history of contributing open networking designs make the Global Summit a plausible venue for a RoCE specification. However, the resolution requires an identifiable MetaRoCE contribution through OCP, not merely a presentation, genera" }, { "url": "https://engineering.fb.com/", "title": "Engineering at Meta - Engineering at Meta Blog", "publishedAt": null, "source": "direct", "excerpt": "Cited by grid-and-power-economist: Meta’s foundational OCP role and history of contributing open networking designs make the Global Summit a plausible venue for a RoCE specification. However, the resolution requires an identifiable MetaRoCE contribution through OCP, not merely a presentation, genera" } ], "bubbleBasket": null, "daysRemaining": 55, "deadline": "2026-10-31", "lens": "networking" } }, { "id": "lq-p74", "weeklyPredictionId": "p74-interconnect-revenue-q3-nov15", "text": "No publicly listed global colocation operator reports Q3 2026 interconnection revenue declining sequentially.", "lens": "networking", "deadline": "2026-11-15", "resolutionSource": "Listed colo 10-Q / earnings.", "resolutionRule": "Miss if any listed global colo prints sequential interconnect/revenue decline for Q3 2026 by Nov 15. Silence is not a hit until the deadline.", "ambiguityPolicy": "resolves_against_forecaster", "bubbleBasket": "interconnect_conversion", "stated": 0.46, "current": 0.4590013965235719, "direction": "flat", "history": [ { "asOf": "2026-08-01", "p": 0.46, "source": "stated", "runId": null }, { "asOf": "2026-09-06", "p": 0.4590013965235719, "source": "engine", "runId": "2026-09-06T22-42-59Z-51a6cc" } ], "engine": { "id": "lq-p74", "weeklyPredictionId": "p74-interconnect-revenue-q3-nov15", "stated": 0.46, "referenceClass": { "p": 0.5, "className": "qtype:disclosure", "n": 29, "note": "most specific ledger class with n >= 8; Resolved predictions of type disclosure: disclosure = requires a named primary document or named company; threshold = numeric bar without a named-disclosure requirement; timing = event-by-date without a numeric bar; other = none of the above." }, "calibrated": { "p": 0.5173399310241416, "method": "hierarchical-beta-binomial-by-lens", "ci80": [ 0.47547336461443984, 0.5591426375154679 ], "variants": { "platt-on-stated-logit": 0.4198 } }, "model": { "p": 0.2594040836618523, "method": "logistic-features-l2", "ci80": [ 0.10206840316291343, 0.534966194711354 ], "features": { "conjuncts": 1, "hasNumericThreshold": false, "horizonDays": 106, "isMarketPriceQuestion": false, "isTimingQuestion": false, "requiresNamedPrimaryDisclosure": true, "statedConfidence": 0.46, "thresholdMagnitudeLog": 0, "weekIndex": 14 }, "gbm": 0.437 }, "council": { "p": 0.5123144681829787, "aggregation": "log-odds-mean", "extremization": 1, "votes": [ { "persona": "grid-and-power-economist", "model": "openai/gpt-5.6-sol", "p": 0.54, "rationale": "Interconnection revenue is predominantly recurring and usually changes slowly; EQIX’s 8% trailing revenue growth and 23% RPO growth support continued sequential expansion. Strong networking demand also lowers the fundamental risk of a Q3 decline. However, this is an “any issuer” test: one reported decline from EQIX or another qualifying listed global operator causes a miss. Currency translation, churn, pricing adjustments, or disclosure reclassification can produce a small sequential decline despite healthy demand. The undefined operator universe and resolves-against-forecaster policy add meaningful downside. I place the result only modestly above even odds and above the forecaster’s 0.46.", "keyEvidence": [ "https://www.sec.gov/edgar/search/" ], "whatWouldChangeMyMind": "Q3 filings showing positive sequential interconnection revenue from EQIX and other clear qualifiers would raise it sharply; an announced revenue decline, changed classification, or delayed filing would lower it." }, { "persona": "capital-markets-credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.54, "rationale": "Interconnection revenue is recurring, diversified, and generally benefits from contractual pricing and rising traffic, making sequential declines uncommon. EQIX’s disclosed Q2 RPO growth of 23% and trailing revenue growth of 8% support a stable demand backdrop, while benign credit spreads and volatility reduce near-term macro risk. However, the event requires every qualifying listed global operator to avoid a decline; FX translation, customer churn, portfolio changes, or inconsistent reporting definitions could produce one adverse print. The broad issuer scope and ambiguity-against-forecaster policy materially limit confidence.", "keyEvidence": [ "https://www.sec.gov/edgar/browse/?CIK=1101239&owner=exclude", "https://www.sec.gov/edgar/browse/?CIK=1297996&owner=exclude" ], "whatWouldChangeMyMind": "Q2 filings showing weakening interconnection additions, elevated churn, or a Q3 guide implying flat reported-currency revenue would lower it; explicit sequential growth guidance across major operators would raise it." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.4, "rationale": "I use the networking-ledger base rate of 0.318, with the broader disclosure base rate of 0.500 as a secondary reference. EQIX’s 8% trailing revenue growth and 23% RPO growth support generally resilient demand and justify an upward adjustment. However, this is effectively a universal conjunction across an ambiguously defined operator set: one sequential decline, including from currency or reporting effects, causes failure. The ambiguity policy further lowers the odds. The forecaster’s sub-0.50 calibration bin has only one observation, so it provides no credible correction; higher-confidence bins show overconfidence but are less relevant here.", "keyEvidence": [ "https://www.sec.gov/edgar/browse/?CIK=1101239&owner=exclude" ], "whatWouldChangeMyMind": "Clear Q3 filings from all plausibly covered operators showing sequentially stable or growing interconnection revenue; conversely, guidance or preliminary results indicating any sequential decline." }, { "persona": "geopolitics-regulation-analyst", "model": "openai/gpt-5.6-sol", "p": 0.57, "rationale": "Interconnection is predominantly recurring, and the digest’s EQIX indicators—8% trailing revenue growth and 23% RPO growth—favor continued expansion. Strong networking-vendor growth also supports port and cross-connect demand. However, this is a universal claim: one listed global operator’s sequential decline causes a miss. Quarter-to-quarter reported revenue can fall because of currency translation, customer churn, pricing adjustments, or metric reclassification even when underlying demand remains healthy. Most relevant Q3 earnings should arrive before November 15, so silence offers little protection.", "keyEvidence": [ "https://www.sec.gov/edgar/search/" ], "whatWouldChangeMyMind": "Q3 preliminary results showing constant-currency interconnection contraction would lower p sharply; disclosed sequential growth from all operators that separately report interconnection would raise it." } ], "dissent": null, "spread": 0.17 }, "final": { "p": 0.4590013965235719, "method": "stacked-fixed-weights (resolved n=57 < 80)", "weights": { "calibrated": 0.3, "council": 0.5, "model": 0.2 }, "ci80": [ 0.3391513679754738, 0.5597750682227807 ] }, "direction": "flat", "drivers": [ "Reference class qtype:disclosure runs 0.50 (n=29) vs stated 0.46.", "Hierarchical calibration moves stated 0.46 to 0.52 (lens networking, kappa_stated=24.0, kappa_lens=200.0).", "Feature model at 0.26: conjuncts=1, named disclosure=True, horizon=106d.", "Council of 4 at 0.51." ], "warnings": [ "semiconductor-supply-chain-analyst vote dropped: recusal pattern [digital_realty]", "all 4 votes came from one vendor (openai); extremization disabled, persona diversity only" ], "evidence": [ { "url": "https://www.sec.gov/edgar/search/", "title": "SEC.gov | EDGAR Full Text Search", "publishedAt": null, "source": "direct", "excerpt": "Cited by grid-and-power-economist: Interconnection revenue is predominantly recurring and usually changes slowly; EQIX’s 8% trailing revenue growth and 23% RPO growth support continued sequential expansion. Strong networking demand also lowers the fundamental risk of a Q3 decline. However, this is a" }, { "url": "https://www.sec.gov/edgar/browse/?CIK=1101239&owner=exclude", "title": "EDGAR Entity Landing Page", "publishedAt": null, "source": "direct", "excerpt": "Cited by capital-markets-credit-analyst: Interconnection revenue is recurring, diversified, and generally benefits from contractual pricing and rising traffic, making sequential declines uncommon. EQIX’s disclosed Q2 RPO growth of 23% and trailing revenue growth of 8% support a stable demand backdro" }, { "url": "https://www.sec.gov/edgar/browse/?CIK=1297996&owner=exclude", "title": "EDGAR Entity Landing Page", "publishedAt": null, "source": "direct", "excerpt": "Cited by capital-markets-credit-analyst: Interconnection revenue is recurring, diversified, and generally benefits from contractual pricing and rising traffic, making sequential declines uncommon. EQIX’s disclosed Q2 RPO growth of 23% and trailing revenue growth of 8% support a stable demand backdro" } ], "bubbleBasket": "interconnect_conversion", "daysRemaining": 70, "deadline": "2026-11-15", "lens": "networking" } }, { "id": "lq-p89", "weeklyPredictionId": "p89-marvell-google-first-tranche-nov30", "text": "Marvell discloses the first attributable Custom Products revenue tranche under the Google relationship by November 30, 2026.", "lens": "hardware", "deadline": "2026-11-30", "resolutionSource": "Marvell 10-Q/K or earnings call.", "resolutionRule": "Hit on attributable Custom Products dollars tied to Google. 'Custom growing' without attribution misses.", "ambiguityPolicy": "resolves_against_forecaster", "stated": 0.34, "current": 0.23476783233214954, "direction": "down", "history": [ { "asOf": "2026-08-22", "p": 0.34, "source": "stated", "runId": null }, { "asOf": "2026-09-06", "p": 0.23476783233214954, "source": "engine", "runId": "2026-09-06T22-42-59Z-51a6cc" } ], "engine": { "id": "lq-p89", "weeklyPredictionId": "p89-marvell-google-first-tranche-nov30", "stated": 0.34, "referenceClass": { "p": 0.5, "className": "qtype:disclosure", "n": 29, "note": "most specific ledger class with n >= 8; Resolved predictions of type disclosure: disclosure = requires a named primary document or named company; threshold = numeric bar without a named-disclosure requirement; timing = event-by-date without a numeric bar; other = none of the above." }, "calibrated": { "p": 0.5198201011002082, "method": "hierarchical-beta-binomial-by-lens", "ci80": [ 0.47813440309378535, 0.5614334275238325 ], "variants": { "platt-on-stated-logit": 0.3524 } }, "model": { "p": 0.5593265824614625, "method": "logistic-features-l2", "ci80": [ 0.3635502925212694, 0.8097434051305782 ], "features": { "conjuncts": 1, "hasNumericThreshold": false, "horizonDays": 100, "isMarketPriceQuestion": false, "isTimingQuestion": true, "requiresNamedPrimaryDisclosure": true, "statedConfidence": 0.34, "thresholdMagnitudeLog": 0, "weekIndex": 17 }, "gbm": 0.6558 }, "council": { "p": 0.07542960421499215, "aggregation": "log-odds-mean", "extremization": 1, "votes": [ { "persona": "grid-and-power-economist", "model": "openai/gpt-5.6-sol", "p": 0.07, "rationale": "The strict resolution requires both a dollar amount and explicit Google attribution in a 10-Q/K or earnings call. Marvell’s ordinary fiscal-third-quarter reporting cadence typically reaches early December, creating substantial risk that no qualifying source appears by the November 30 deadline. Even if an earnings call occurs in time, semiconductor vendors usually discuss custom-silicon growth by product category or unnamed customer rather than disclose customer-specific revenue. The public digest shows strong companywide growth but no Google-attributable dollars. These timing and disclosure hurdles make the forecaster’s 0.34 too high.", "keyEvidence": [ "https://investor.marvell.com/", "https://www.sec.gov/edgar/browse/?CIK=1835632&owner=exclude" ], "whatWouldChangeMyMind": "A confirmed earnings date on or before November 30, plus company guidance indicating it will quantify Google-specific custom-product revenue." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.12, "rationale": "The underlying custom-silicon ramp is plausible—MRVL’s trailing revenue growth is 31%—but the resolution bar is much harder than evidence of growth. Marvell and peers usually discuss aggregate custom/data-center revenue and anonymize hyperscaler concentration, making explicit Google-attributed dollars uncommon. Timing also hurts: Marvell’s late-October fiscal quarter has typically been reported about a month later, putting the next 10-Q or call around December 1, just outside the cutoff. A qualifying statement in the late-August call or a slightly earlier next report remains possible. I am therefore well below the forecaster’s 0.34.", "keyEvidence": [ "https://investor.marvell.com/" ], "whatWouldChangeMyMind": "A verified late-August transcript giving a Google-specific dollar contribution, or an announced Q3 call by November 30 with guidance that named-customer revenue will be quantified." }, { "persona": "capital-markets-credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.1, "rationale": "The rule requires both customer attribution and Custom Products dollars; Marvell’s customary discussion of aggregate custom-silicon growth or unnamed hyperscalers would fail. Customer-specific revenue is rarely disclosed beyond anonymous concentration data, especially where confidentiality applies. Timing further lowers the odds: the deadline likely precedes or barely coincides with Marvell’s normal fiscal third-quarter reporting window, leaving limited qualifying disclosure opportunities. The digest’s 31% trailing revenue growth supports a ramp but provides no Google attribution. Ambiguity resolving against the forecaster makes inferred dollars insufficient.", "keyEvidence": [ "https://www.sec.gov/edgar/search/", "https://investor.marvell.com/" ], "whatWouldChangeMyMind": "A Marvell earnings date on or before November 30, plus guidance promising quantified Google-linked Custom Products revenue, or a filing explicitly identifying Google and the associated dollars." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.11, "rationale": "I use the disclosure-question base rate of 0.50 (29 cases), rather than the noisier 0.615 hardware rate. This rule is much stricter than ordinary segment commentary: Marvell must both tie revenue to Google and state attributable Custom Products dollars. Customer-specific revenue is often anonymized or discussed qualitatively, so even a genuine ramp can miss. Timing further hurts: only 85 days remain, and the deadline may precede the next routine quarterly reporting window. The forecaster’s calibration supplies no credible low-confidence correction because that band has only one observation.", "keyEvidence": [ "https://www.sec.gov/edgar/browse/?CIK=1835632&owner=exclude" ], "whatWouldChangeMyMind": "A Marvell filing or call transcript explicitly naming Google and quantifying associated Custom Products revenue would settle it; a confirmed pre-deadline earnings call with promised customer-level disclosure would materially raise the probability." }, { "persona": "geopolitics-regulation-analyst", "model": "openai/gpt-5.6-sol", "p": 0.025, "rationale": "The literal hurdle is unusually strict: Marvell must state dollar revenue and explicitly attribute it to Google; aggregate Custom Products growth or anonymous customer concentration does not qualify. Marvell generally reports its fiscal third quarter around the first Tuesday of December. Calendar progression from December 3, 2024 and December 2, 2025 points to December 1, 2026—one day after the deadline—with the 10-Q later still. An earlier-than-usual earnings call is possible, but even then customer-specific dollar attribution is unlikely.", "keyEvidence": [ "https://investor.marvell.com/", "https://www.sec.gov/edgar/browse/?CIK=1835632&owner=exclude" ], "whatWouldChangeMyMind": "A Marvell announcement scheduling fiscal Q3 earnings for November 30 or earlier, combined with prior management guidance promising customer-specific Custom Products revenue disclosure." } ], "dissent": null, "spread": 0.095 }, "final": { "p": 0.23476783233214954, "method": "stacked-fixed-weights (resolved n=57 < 80)", "weights": { "calibrated": 0.3, "council": 0.5, "model": 0.2 }, "ci80": [ 0.12238525250412052, 0.34695479501615917 ] }, "direction": "down", "drivers": [ "Reference class qtype:disclosure runs 0.50 (n=29) vs stated 0.34.", "Hierarchical calibration moves stated 0.34 to 0.52 (lens hardware, kappa_stated=24.0, kappa_lens=200.0).", "Feature model at 0.56: conjuncts=1, named disclosure=True, horizon=100d.", "Council of 5 at 0.08." ], "warnings": [ "all 5 votes came from one vendor (openai); extremization disabled, persona diversity only" ], "evidence": [ { "url": "https://investor.marvell.com/", "title": "Investor Relations | Marvell Technology, Inc. (MRVL)", "publishedAt": null, "source": "direct", "excerpt": "Cited by grid-and-power-economist: The strict resolution requires both a dollar amount and explicit Google attribution in a 10-Q/K or earnings call. Marvell’s ordinary fiscal-third-quarter reporting cadence typically reaches early December, creating substantial risk that no qualifying source appears" }, { "url": "https://www.sec.gov/edgar/browse/?CIK=1835632&owner=exclude", "title": "EDGAR Entity Landing Page", "publishedAt": null, "source": "direct", "excerpt": "Cited by grid-and-power-economist: The strict resolution requires both a dollar amount and explicit Google attribution in a 10-Q/K or earnings call. Marvell’s ordinary fiscal-third-quarter reporting cadence typically reaches early December, creating substantial risk that no qualifying source appears" }, { "url": "https://www.sec.gov/edgar/search/", "title": "SEC.gov | EDGAR Full Text Search", "publishedAt": null, "source": "direct", "excerpt": "Cited by capital-markets-credit-analyst: The rule requires both customer attribution and Custom Products dollars; Marvell’s customary discussion of aggregate custom-silicon growth or unnamed hyperscalers would fail. Customer-specific revenue is rarely disclosed beyond anonymous concentration data, e" } ], "bubbleBasket": null, "daysRemaining": 85, "deadline": "2026-11-30", "lens": "hardware" } }, { "id": "lq-p98", "weeklyPredictionId": "p98-nvidia-rubin-mix-q3-earnings", "text": "NVIDIA Q3 FY2027 earnings disclosure states Vera Rubin contributed more than 25% of datacenter revenue.", "lens": "hardware", "deadline": "2026-11-30", "resolutionSource": "NVIDIA 10-Q or earnings call for quarter ended October 26, 2026.", "resolutionRule": "Hit if the mix is stated above 25%. If mix is not disclosed, miss (ambiguity against the forecaster).", "ambiguityPolicy": "resolves_against_forecaster", "stated": 0.74, "current": 0.44378638815176197, "direction": "down", "history": [ { "asOf": "2026-08-29", "p": 0.74, "source": "stated", "runId": null }, { "asOf": "2026-09-06", "p": 0.44378638815176197, "source": "engine", "runId": "2026-09-06T22-42-59Z-51a6cc" } ], "engine": { "id": "lq-p98", "weeklyPredictionId": "p98-nvidia-rubin-mix-q3-earnings", "stated": 0.74, "referenceClass": { "p": 0.5, "className": "qtype:disclosure", "n": 29, "note": "most specific ledger class with n >= 8; Resolved predictions of type disclosure: disclosure = requires a named primary document or named company; threshold = numeric bar without a named-disclosure requirement; timing = event-by-date without a numeric bar; other = none of the above." }, "calibrated": { "p": 0.560498067201903, "method": "hierarchical-beta-binomial-by-lens", "ci80": [ 0.5190129713079354, 0.6017622600985648 ], "variants": { "platt-on-stated-logit": 0.5892 } }, "model": { "p": 0.5399524523136026, "method": "logistic-features-l2", "ci80": [ 0.3724461295272785, 0.7697858680664276 ], "features": { "conjuncts": 1, "hasNumericThreshold": true, "horizonDays": 93, "isMarketPriceQuestion": false, "isTimingQuestion": false, "requiresNamedPrimaryDisclosure": true, "statedConfidence": 0.74, "thresholdMagnitudeLog": 1.398, "weekIndex": 18 }, "gbm": 0.4014 }, "council": { "p": 0.340382788134062, "aggregation": "log-odds-mean", "extremization": 1, "votes": [ { "persona": "grid-and-power-economist", "model": "openai/gpt-5.6-sol", "p": 0.42, "rationale": "Rubin could ramp rapidly, as NVIDIA’s annual platform cadence and strong datacenter growth support substantial early shipments. However, Q3 FY2027 is likely an early production quarter, while HBM4, advanced packaging, networking, and rack integration constrain first-quarter volume. Exceeding 25% also requires Rubin to outgrow a rapidly expanding datacenter denominator. Most importantly, the rule requires an explicit mix disclosure; NVIDIA may discuss availability, shipments, or demand without quantifying Rubin’s share. That disclosure risk makes 0.74 too high.", "keyEvidence": [ "https://investor.nvidia.com/", "https://www.sec.gov/edgar/browse/?CIK=1045810&owner=exclude" ], "whatWouldChangeMyMind": "A pre-quarter filing or official NVIDIA statement quantifying Rubin production revenue, especially above one-quarter of the latest datacenter run rate, would raise the estimate substantially." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.29, "rationale": "Rubin was roadmapped for second-half 2026, but exceeding 25% of datacenter revenue by the October quarter would require an unusually steep initial ramp through HBM4 qualification, CoWoS-L capacity, substrates, and complete-system availability. Blackwell demonstrated that NVIDIA can ramp a new architecture rapidly, supporting some upside. The larger obstacle is the resolution rule: NVIDIA often discusses ramp progress or product revenue in dollars but does not consistently state architecture mix as a percentage. A qualitative Rubin update, or no separable mix disclosure, resolves as a miss. The forecaster’s 0.74 appears too high for this joint production-and-disclosure requirement.", "keyEvidence": [ "https://www.sec.gov/edgar/browse/?CIK=1045810&owner=exclude" ], "whatWouldChangeMyMind": "Official pre-quarter evidence of Rubin volume shipments plus guidance or reporting practice indicating separately quantified Rubin revenue would raise the estimate substantially." }, { "persona": "capital-markets-credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.31, "rationale": "This requires both an unusually rapid Rubin ramp and explicit product-mix disclosure. The quarter likely captures an early transition period, when incumbent Blackwell-family shipments could still dominate datacenter revenue. NVIDIA sometimes quantifies revenue from a newly launched architecture, which supports the case, but it more commonly discloses datacenter revenue by compute and networking rather than a precise architecture percentage. The reported 83% trailing revenue growth indicates strong accelerator demand but says little about Rubin’s share. Given the ambiguity-against-forecaster rule, even strong qualitative Rubin commentary would miss. Thus 0.74 appears too high.", "keyEvidence": [ "https://www.sec.gov/edgar/browse/?CIK=1045810&owner=exclude", "https://investor.nvidia.com/" ], "whatWouldChangeMyMind": "Official evidence that Rubin entered volume shipments early in the quarter, or guidance explicitly forecasting Rubin revenue comfortably above one-quarter of datacenter revenue." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.31, "rationale": "I start from the 0.50 disclosure-question base rate (n=29), with the 0.615 hardware rate as secondary context. This requires both Rubin exceeding 25% and NVIDIA explicitly quantifying that mix in the specified 10-Q or call. A rapid new-platform ramp is plausible, and NVIDIA’s strong companywide growth supports demand, but it does not establish Rubin-specific shipments. NVIDIA could discuss Rubin growth or disclose dollar revenue without stating the required percentage; that is an automatic miss. The forecaster’s 0.70–0.80 calibration is 0.70, slightly below stated confidence, and does not overcome the disclosure penalty.", "keyEvidence": [ "https://investor.nvidia.com/", "https://www.sec.gov/edgar/browse/?CIK=1045810" ], "whatWouldChangeMyMind": "Official evidence of Rubin volume shipments early in the quarter, plus guidance or precedent indicating NVIDIA will explicitly report Rubin’s percentage of Data Center revenue." }, { "persona": "geopolitics-regulation-analyst", "model": "openai/gpt-5.6-sol", "p": 0.38, "rationale": "Rubin’s planned 2026 ramp makes material Q3 revenue plausible, and NVIDIA has previously quantified first-quarter revenue for major architecture launches. However, exceeding 25% of rapidly growing datacenter revenue would require a fast production, systems-integration, and revenue-recognition ramp within the quarter. The stricter obstacle is disclosure: NVIDIA often discusses demand and shipment progress without stating architecture mix, and nondisclosure is an automatic miss. The digest’s 83% trailing revenue growth supports strong platform demand but also raises the dollar amount Rubin must contribute to cross the threshold.", "keyEvidence": [ "https://investor.nvidia.com/", "https://www.sec.gov/edgar/browse/?CIK=1045810&owner=exclude" ], "whatWouldChangeMyMind": "Raise materially if NVIDIA confirms volume Rubin shipments early in the quarter or guides to architecture-specific revenue; lower if shipments are weighted toward late 2026 or management avoids product-level revenue figures." } ], "dissent": null, "spread": 0.13 }, "final": { "p": 0.44378638815176197, "method": "stacked-fixed-weights (resolved n=57 < 80)", "weights": { "calibrated": 0.3, "council": 0.5, "model": 0.2 }, "ci80": [ 0.3707004915833979, 0.5507936278865512 ] }, "direction": "down", "drivers": [ "Reference class qtype:disclosure runs 0.50 (n=29) vs stated 0.74.", "Hierarchical calibration moves stated 0.74 to 0.56 (lens hardware, kappa_stated=24.0, kappa_lens=200.0).", "Feature model at 0.54: conjuncts=1, named disclosure=True, horizon=93d.", "Council of 5 at 0.34." ], "warnings": [ "all 5 votes came from one vendor (openai); extremization disabled, persona diversity only" ], "evidence": [ { "url": "https://investor.nvidia.com/", "title": "NVIDIA Corporation - Home", "publishedAt": null, "source": "direct", "excerpt": "Cited by grid-and-power-economist: Rubin could ramp rapidly, as NVIDIA’s annual platform cadence and strong datacenter growth support substantial early shipments. However, Q3 FY2027 is likely an early production quarter, while HBM4, advanced packaging, networking, and rack integration constrain firs" }, { "url": "https://www.sec.gov/edgar/browse/?CIK=1045810&owner=exclude", "title": "EDGAR Entity Landing Page", "publishedAt": null, "source": "direct", "excerpt": "Cited by grid-and-power-economist: Rubin could ramp rapidly, as NVIDIA’s annual platform cadence and strong datacenter growth support substantial early shipments. However, Q3 FY2027 is likely an early production quarter, while HBM4, advanced packaging, networking, and rack integration constrain firs" }, { "url": "https://www.sec.gov/edgar/browse/?CIK=1045810", "title": "EDGAR Entity Landing Page", "publishedAt": null, "source": "direct", "excerpt": "Cited by superforecaster-statistician: I start from the 0.50 disclosure-question base rate (n=29), with the 0.615 hardware rate as secondary context. This requires both Rubin exceeding 25% and NVIDIA explicitly quantifying that mix in the specified 10-Q or call. A rapid new-platform ramp is plausibl" } ], "bubbleBasket": null, "daysRemaining": 85, "deadline": "2026-11-30", "lens": "hardware" } }, { "id": "lq-p75", "weeklyPredictionId": "p75-vendor-backstop-documented-dec31", "text": "A definitive agreement of at least $100B in vendor-guaranteed AI data-center financing is documented by December 31, 2026.", "lens": "capital", "deadline": "2026-12-31", "resolutionSource": "Definitive agreement 8-K or equivalent.", "resolutionRule": "Hit on a signed agreement ≥$100B with vendor guarantee language. MOUs miss.", "ambiguityPolicy": "resolves_against_forecaster", "bubbleBasket": "capex_durability", "stated": 0.34, "current": 0.27229890086120917, "direction": "down", "history": [ { "asOf": "2026-08-01", "p": 0.34, "source": "stated", "runId": null }, { "asOf": "2026-09-06", "p": 0.27229890086120917, "source": "engine", "runId": "2026-09-06T22-42-59Z-51a6cc" } ], "engine": { "id": "lq-p75", "weeklyPredictionId": "p75-vendor-backstop-documented-dec31", "stated": 0.34, "referenceClass": { "p": 0.5, "className": "qtype:disclosure", "n": 29, "note": "most specific ledger class with n >= 8; Resolved predictions of type disclosure: disclosure = requires a named primary document or named company; threshold = numeric bar without a named-disclosure requirement; timing = event-by-date without a numeric bar; other = none of the above." }, "calibrated": { "p": 0.5241597611048898, "method": "hierarchical-beta-binomial-by-lens", "ci80": [ 0.4823935933928894, 0.5658373336349926 ], "variants": { "platt-on-stated-logit": 0.3524 } }, "model": { "p": 0.4097786792564665, "method": "logistic-features-l2", "ci80": [ 0.19193705110944945, 0.7041575428913229 ], "features": { "conjuncts": 1, "hasNumericThreshold": true, "horizonDays": 152, "isMarketPriceQuestion": false, "isTimingQuestion": true, "requiresNamedPrimaryDisclosure": true, "statedConfidence": 0.34, "thresholdMagnitudeLog": 11, "weekIndex": 14 }, "gbm": 0.3348 }, "council": { "p": 0.13261226638300513, "aggregation": "log-odds-mean", "extremization": 1, "votes": [ { "persona": "grid-and-power-economist", "model": "openai/gpt-5.6-sol", "p": 0.14, "rationale": "I am below the forecaster’s 0.34. Large and rapidly growing RPOs support extraordinary AI infrastructure demand, while low broad high-yield spreads and subdued volatility make financing feasible. But a single signed financing of at least $100B with explicit vendor-guarantee language would be unprecedented in scale, create major contingent-liability and concentration issues, and likely require extensive syndication and approvals. Transactions are more likely to be staged across facilities, projects, or counterparties. The strict requirement for a definitive 8-K or equivalent—and rejection of MOUs—materially lowers the four-month hit probability.", "keyEvidence": [ "https://www.sec.gov/edgar/search/", "https://fred.stlouisfed.org/series/BAMLH0A0HYM2", "https://fred.stlouisfed.org/series/VIXCLS" ], "whatWouldChangeMyMind": "A filed commitment letter or definitive agreement identifying at least $100B, the vendor guarantor, covered obligations, and binding guarantee terms would raise my estimate sharply." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.21, "rationale": "Capital markets are supportive: investment-grade spreads and volatility are low, while public-company RPO and capex figures confirm exceptional AI infrastructure demand. But a single $100B financing is extraordinarily large, and explicit vendor-guarantee language would require complex credit approval, documentation, and likely material disclosure. Large customer commitments or vendor investments would not qualify. With only 116 days remaining, an announced framework or MOU could easily miss the definitive-agreement test. The strict 8-K-or-equivalent requirement and adverse ambiguity policy put this below the forecaster’s 0.34.", "keyEvidence": [ "https://www.sec.gov/edgar/search/", "https://fred.stlouisfed.org/series/BAMLC0A0CM", "https://fred.stlouisfed.org/series/DGS10" ], "whatWouldChangeMyMind": "A primary filing identifying committed financing above $100B, explicit vendor guarantee or backstop terms, and a stated definitive-signing timetable before year-end." }, { "persona": "capital-markets-credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.12, "rationale": "Capital demand is plainly large: the digest shows multiple public issuers with RPO above $100B and rapid capex growth, while broad high-yield spreads remain benign. But the threshold concerns one signed financing—not projected spending or RPO—and requires explicit vendor-guarantee language. A $100B guarantee would create extraordinary contingent exposure, requiring extensive approvals, covenants, and material disclosure. Only 116 days remain; a project announcement, customer commitment, consortium, term sheet, or MOU would fail. Single-B OAS at 10.51% also complicates execution for weaker borrowers. The exact-document requirement and adverse ambiguity policy make late or vague disclosure a miss. This is materially less likely than the forecaster’s 0.34.", "keyEvidence": [ "https://www.sec.gov/edgar/search/", "https://fred.stlouisfed.org/series/BAMLH0A0HYM2", "https://fred.stlouisfed.org/series/BAMLH0A3HYC" ], "whatWouldChangeMyMind": "An issuer filing showing executed commitments approaching $100B with an explicit vendor guarantee, or a filed near-final agreement naming financing parties, would raise the probability sharply." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.08, "rationale": "I use 0.40, the ledger’s >90-day base rate, rather than the 0.708 capital-lens rate because that class mixes much easier outcomes. This proposition has several necessary gates: exceptional scale, AI-data-center purpose, explicit vendor guarantee, signature rather than an MOU, and qualifying disclosure within 116 days. Large RPOs and accelerating capex show financing demand; low broad credit spreads and VIX aid execution. Conversely, the 4.77 10-year yield and 10.51 single-B spread make giant risk-bearing guarantees harder, and the digest identifies no pending definitive agreement. Hard-date disclosure further reduces odds. Low-confidence calibration has only one observation, providing no credible uplift.", "keyEvidence": [ "https://www.sec.gov/edgar/search/", "https://fred.stlouisfed.org/series/DGS10", "https://fred.stlouisfed.org/series/BAMLH0A3HYC" ], "whatWouldChangeMyMind": "A filed announcement of committed financing near $100B, identifying guarantor obligations and showing that only customary closing conditions remain, would raise the estimate sharply." }, { "persona": "geopolitics-regulation-analyst", "model": "openai/gpt-5.6-sol", "p": 0.14, "rationale": "The capital-spending and RPO figures show extraordinary AI infrastructure demand, but they do not evidence a qualifying financing agreement. A single $100B financing with an explicit vendor guarantee would be unusually large, require substantial underwriting and risk allocation, and likely generate conspicuous issuer disclosures. Only 116 days remain. Financing conditions are mixed: broad high-yield spreads are tight, but the 10-year Treasury yield is 4.77% and single-B spreads are elevated. Most importantly, the rule excludes MOUs and resolves ambiguity against a hit; vendor investment, purchase commitments, backstops, or several smaller facilities would not suffice without signed guarantee language in a definitive filing.", "keyEvidence": [ "https://www.sec.gov/edgar/search/", "https://fred.stlouisfed.org/series/DGS10", "https://fred.stlouisfed.org/series/BAMLH0A3HYC" ], "whatWouldChangeMyMind": "A filed commitment letter or near-final definitive agreement identifying at least $100B, the guaranteeing vendor, guarantee scope, committed lenders, and a closing timetable before year-end." } ], "dissent": null, "spread": 0.13 }, "final": { "p": 0.27229890086120917, "method": "stacked-fixed-weights (resolved n=57 < 80)", "weights": { "calibrated": 0.3, "council": 0.5, "model": 0.2 }, "ci80": [ 0.17802229960366087, 0.3990172726487121 ] }, "direction": "down", "drivers": [ "Reference class qtype:disclosure runs 0.50 (n=29) vs stated 0.34.", "Hierarchical calibration moves stated 0.34 to 0.52 (lens capital, kappa_stated=24.0, kappa_lens=200.0).", "Feature model at 0.41: conjuncts=1, named disclosure=True, horizon=152d.", "Council of 5 at 0.13." ], "warnings": [ "all 5 votes came from one vendor (openai); extremization disabled, persona diversity only" ], "evidence": [ { "url": "https://www.sec.gov/edgar/search/", "title": "SEC.gov | EDGAR Full Text Search", "publishedAt": null, "source": "direct", "excerpt": "Cited by grid-and-power-economist: I am below the forecaster’s 0.34. Large and rapidly growing RPOs support extraordinary AI infrastructure demand, while low broad high-yield spreads and subdued volatility make financing feasible. But a single signed financing of at least $100B with explicit vendor-" }, { "url": "https://fred.stlouisfed.org/series/BAMLH0A0HYM2", "title": "ICE BofA US High Yield Index Option-Adjusted Spread (BAMLH0A0HYM2) | FRED | St. Louis Fed", "publishedAt": null, "source": "direct", "excerpt": "Cited by grid-and-power-economist: I am below the forecaster’s 0.34. Large and rapidly growing RPOs support extraordinary AI infrastructure demand, while low broad high-yield spreads and subdued volatility make financing feasible. But a single signed financing of at least $100B with explicit vendor-" }, { "url": "https://fred.stlouisfed.org/series/VIXCLS", "title": "CBOE Volatility Index: VIX (VIXCLS) | FRED | St. Louis Fed", "publishedAt": null, "source": "direct", "excerpt": "Cited by grid-and-power-economist: I am below the forecaster’s 0.34. Large and rapidly growing RPOs support extraordinary AI infrastructure demand, while low broad high-yield spreads and subdued volatility make financing feasible. But a single signed financing of at least $100B with explicit vendor-" }, { "url": "https://fred.stlouisfed.org/series/BAMLC0A0CM", "title": "ICE BofA US Corporate Index Option-Adjusted Spread (BAMLC0A0CM) | FRED | St. Louis Fed", "publishedAt": null, "source": "direct", "excerpt": "Cited by semiconductor-supply-chain-analyst: Capital markets are supportive: investment-grade spreads and volatility are low, while public-company RPO and capex figures confirm exceptional AI infrastructure demand. But a single $100B financing is extraordinarily large, and explicit vendor-guarantee " }, { "url": "https://fred.stlouisfed.org/series/DGS10", "title": "Market Yield on U.S. Treasury Securities at 10-Year Constant Maturity, Quoted on an Investment Basis (DGS10) | FRED | St. Louis Fed", "publishedAt": null, "source": "direct", "excerpt": "Cited by semiconductor-supply-chain-analyst: Capital markets are supportive: investment-grade spreads and volatility are low, while public-company RPO and capex figures confirm exceptional AI infrastructure demand. But a single $100B financing is extraordinarily large, and explicit vendor-guarantee " }, { "url": "https://fred.stlouisfed.org/series/BAMLH0A3HYC", "title": "ICE BofA CCC & Lower US High Yield Index Option-Adjusted Spread (BAMLH0A3HYC) | FRED | St. Louis Fed", "publishedAt": null, "source": "direct", "excerpt": "Cited by capital-markets-credit-analyst: Capital demand is plainly large: the digest shows multiple public issuers with RPO above $100B and rapid capex growth, while broad high-yield spreads remain benign. But the threshold concerns one signed financing—not projected spending or RPO—and requires exp" } ], "bubbleBasket": "capex_durability", "daysRemaining": 116, "deadline": "2026-12-31", "lens": "capital" } }, { "id": "lq-p76", "weeklyPredictionId": "p76-paducah-psc-filing-dec31", "text": "A power service agreement for the Paducah AI campus is filed with the Kentucky PSC by December 31, 2026.", "lens": "power", "deadline": "2026-12-31", "resolutionSource": "Kentucky PSC docket.", "resolutionRule": "Hit if a PSA or equivalent for the Paducah campus is file-stamped by 2026-12-31.", "ambiguityPolicy": "resolves_against_forecaster", "bubbleBasket": "energization", "stated": 0.63, "current": 0.5200150534728737, "direction": "down", "history": [ { "asOf": "2026-08-01", "p": 0.63, "source": "stated", "runId": null }, { "asOf": "2026-09-06", "p": 0.5200150534728737, "source": "engine", "runId": "2026-09-06T22-42-59Z-51a6cc" } ], "engine": { "id": "lq-p76", "weeklyPredictionId": "p76-paducah-psc-filing-dec31", "stated": 0.63, "referenceClass": { "p": 0.5, "className": "qtype:disclosure", "n": 29, "note": "most specific ledger class with n >= 8; Resolved predictions of type disclosure: disclosure = requires a named primary document or named company; threshold = numeric bar without a named-disclosure requirement; timing = event-by-date without a numeric bar; other = none of the above." }, "calibrated": { "p": 0.5488488765774084, "method": "hierarchical-beta-binomial-by-lens", "ci80": [ 0.5065540909189947, 0.5909589969214164 ], "variants": { "platt-on-stated-logit": 0.5173 } }, "model": { "p": 0.6432747939985151, "method": "logistic-features-l2", "ci80": [ 0.29222930977903394, 0.9313491070524849 ], "features": { "conjuncts": 1, "hasNumericThreshold": false, "horizonDays": 152, "isMarketPriceQuestion": false, "isTimingQuestion": true, "requiresNamedPrimaryDisclosure": true, "statedConfidence": 0.63, "thresholdMagnitudeLog": 0, "weekIndex": 14 }, "gbm": 0.7293 }, "council": { "p": 0.45183756315035256, "aggregation": "log-odds-mean", "extremization": 1, "votes": [ { "persona": "grid-and-power-economist", "model": "openai/gpt-5.6-sol", "p": 0.43, "rationale": "The exact-document rule is a meaningful hurdle: negotiation or announcement does not count without a file-stamped PSA or equivalent. No docket number, executed agreement, or published filing schedule is identified. The question also does not establish that the serving utility must submit the campus agreement to the PSC rather than handle it through another approval path. Four months is enough for an advanced special-contract filing, but large-load agreements typically require load, security, cost-allocation, and infrastructure terms first. The forecaster’s >90-day and disclosure base rates are only 0.4 and 0.5, and its 0.60–0.69 forecasts have realized below stated confidence.", "keyEvidence": [ "https://psc.ky.gov/" ], "whatWouldChangeMyMind": "A PSC case number, utility board authorization, signed term sheet, or public procedural schedule explicitly targeting a 2026 special-contract filing would raise the probability substantially." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.48, "rationale": "A year-end filing remains plausible, but the resolution requires a file-stamped PSA or equivalent in a specific Kentucky PSC docket, not merely a utility announcement or signed private agreement. With only 116 days left, negotiation, load-study, tariff, and commission-filing preparation create meaningful schedule risk. The power reference class is favorable but very small; the broader >90-day and disclosure classes are near or below 0.5. The forecaster’s 0.60–0.69 band has also realized below stated confidence. No Paducah-specific filing milestone is supplied, and ambiguity resolves against a hit.", "keyEvidence": [ "https://psc.ky.gov/" ], "whatWouldChangeMyMind": "A Kentucky PSC docket entry, utility application, agenda item, or public statement giving a firm 2026 PSA filing date would raise this materially." }, { "persona": "capital-markets-credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.46, "rationale": "A filing requires an executed or filing-ready agreement, utility approval processes, and a docket stamp within 116 days; general AI-power demand does not establish those project-specific steps. Paducah-area service may involve municipal or TVA-related arrangements, so Kentucky PSC filing jurisdiction is not automatic. The deadline is still achievable if negotiations are already advanced, because the rule requires filing rather than approval or energization. I weight the >90-day and disclosure base rates near 0.4–0.5 more heavily than the small power sample, putting this below the forecaster’s 0.63.", "keyEvidence": [ "https://psc.ky.gov/" ], "whatWouldChangeMyMind": "A Kentucky PSC notice, docket application, or utility announcement identifying the Paducah campus and an executed or imminent power-service agreement would raise the probability substantially." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.41, "rationale": "I use a 0.45 base rate, weighting the >90-day class (0.40) and disclosure class (0.50); the 0.70 power rate has only four observations. The forecaster’s 0.60–0.69 band realized 0.556, indicating overconfidence relative to 0.63. This requires a particular regulatory filing within 116 days, so agreement timing alone is insufficient, and ambiguity over an “equivalent” resolves against a hit. The digest provides no project-specific, verifiable filing milestone warranting an upward adjustment.", "keyEvidence": [ "https://psc.ky.gov/" ], "whatWouldChangeMyMind": "A Kentucky PSC docket entry, utility application notice, or public meeting agenda explicitly identifying an imminent Paducah power-service agreement filing." }, { "persona": "geopolitics-regulation-analyst", "model": "openai/gpt-5.6-sol", "p": 0.48, "rationale": "The filing requires two milestones within 116 days: execution of a qualifying service arrangement and a Kentucky PSC file stamp clearly tied to the Paducah campus. Special-contract filings can follow execution quickly, so construction or interconnection completion is unnecessary. However, the supplied evidence contains no project-specific docket, executed agreement, or announced filing schedule. The forecaster’s comparable disclosure and >90-day cases resolved near 0.5 and 0.4, respectively; the stronger power-lens record has only four observations. Year-end timing and the rule resolving unclear equivalents or campus attribution against the forecast further reduce the chance of a hit.", "keyEvidence": [ "https://psc.ky.gov/" ], "whatWouldChangeMyMind": "A PSC case number, utility board approval, executed term sheet, or public statement that a Paducah service contract will be submitted before year-end would raise the estimate materially." } ], "dissent": null, "spread": 0.07 }, "final": { "p": 0.5200150534728737, "method": "stacked-fixed-weights (resolved n=57 < 80)", "weights": { "calibrated": 0.3, "council": 0.5, "model": 0.2 }, "ci80": [ 0.41313179819757356, 0.6437959358755122 ] }, "direction": "down", "drivers": [ "Reference class qtype:disclosure runs 0.50 (n=29) vs stated 0.63.", "Hierarchical calibration moves stated 0.63 to 0.55 (lens power, kappa_stated=24.0, kappa_lens=200.0).", "Feature model at 0.64: conjuncts=1, named disclosure=True, horizon=152d.", "Council of 5 at 0.45." ], "warnings": [ "all 5 votes came from one vendor (openai); extremization disabled, persona diversity only" ], "evidence": [ { "url": "https://psc.ky.gov/", "title": "KY PSC Home", "publishedAt": null, "source": "direct", "excerpt": "Cited by grid-and-power-economist: The exact-document rule is a meaningful hurdle: negotiation or announcement does not count without a file-stamped PSA or equivalent. No docket number, executed agreement, or published filing schedule is identified. The question also does not establish that the serv" } ], "bubbleBasket": "energization", "daysRemaining": 116, "deadline": "2026-12-31", "lens": "power" } }, { "id": "lq-p84", "weeklyPredictionId": "p84-second-neocloud-tenor-gap-dec31", "text": "A second publicly traded neocloud discloses a GPU contract tenor gap in an SEC filing by December 31, 2026.", "lens": "capital", "deadline": "2026-12-31", "resolutionSource": "SEC filing.", "resolutionRule": "Hit if a second named neocloud (beyond the first already counted) discloses GPU useful-life vs contract tenor mismatch.", "ambiguityPolicy": "resolves_against_forecaster", "bubbleBasket": "residual_duration", "stated": 0.69, "current": 0.5355399996153618, "direction": "down", "history": [ { "asOf": "2026-08-15", "p": 0.69, "source": "stated", "runId": null }, { "asOf": "2026-09-06", "p": 0.5355399996153618, "source": "engine", "runId": "2026-09-06T22-42-59Z-51a6cc" } ], "engine": { "id": "lq-p84", "weeklyPredictionId": "p84-second-neocloud-tenor-gap-dec31", "stated": 0.69, "referenceClass": { "p": 0.5, "className": "qtype:disclosure", "n": 29, "note": "most specific ledger class with n >= 8; Resolved predictions of type disclosure: disclosure = requires a named primary document or named company; threshold = numeric bar without a named-disclosure requirement; timing = event-by-date without a numeric bar; other = none of the above." }, "calibrated": { "p": 0.5599044419559537, "method": "hierarchical-beta-binomial-by-lens", "ci80": [ 0.5183261061802915, 0.6012631058058022 ], "variants": { "platt-on-stated-logit": 0.5551 } }, "model": { "p": 0.7728018255394563, "method": "logistic-features-l2", "ci80": [ 0.6289863652874564, 0.8881932039736512 ], "features": { "conjuncts": 1, "hasNumericThreshold": false, "horizonDays": 138, "isMarketPriceQuestion": false, "isTimingQuestion": true, "requiresNamedPrimaryDisclosure": true, "statedConfidence": 0.69, "thresholdMagnitudeLog": 0, "weekIndex": 16 }, "gbm": 0.8055 }, "council": { "p": 0.41354144451179253, "aggregation": "log-odds-mean", "extremization": 1, "votes": [ { "persona": "grid-and-power-economist", "model": "openai/gpt-5.6-sol", "p": 0.39, "rationale": "The event requires more than separate depreciation and RPO disclosures: a second named public neocloud must disclose the GPU-life/customer-contract-tenor mismatch in an SEC filing. Remaining opportunities are mainly third-quarter reports and financing or registration filings. CoreWeave’s capex, debt, and RPO profile makes the issue salient, but the digest identifies Nebius as the clearest additional pure-play, and foreign-private-issuer reporting can provide less quarterly detail. Miner-to-AI firms may not qualify as neoclouds, while inferred gaps fail under the ambiguity rule. The >90-day ledger base rate is 0.4; the narrow issuer set and required wording outweigh the stronger capital-lens history, making 0.69 too high.", "keyEvidence": [ "https://www.sec.gov/edgar/search/" ], "whatWouldChangeMyMind": "An SEC filing from Nebius or another clearly qualifying public neocloud explicitly comparing GPU depreciation life with signed customer-contract duration would sharply increase this; relevant year-end filings without that comparison would reduce it." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.43, "rationale": "The deadline allows roughly one regular quarterly filing cycle. Rising RPO, debt, and capital intensity make GPU depreciation assumptions increasingly material, but issuers commonly disclose useful lives without explicitly comparing them with customer-contract tenor. The candidate pool is also narrow: CoreWeave is presumably the first counted issuer, while Nebius has foreign-issuer reporting cadence and other GPU-cloud operators may not unambiguously qualify as neoclouds. A qualifying filing must identify both measures clearly enough to establish a mismatch; the adverse ambiguity policy materially reduces the chance of resolution. This supports a probability well below the forecaster’s 0.69.", "keyEvidence": [ "https://www.sec.gov/edgar/search/" ], "whatWouldChangeMyMind": "A candidate’s upcoming earnings materials previewing both GPU depreciation life and weighted-average customer contract duration, with those materials furnished on Form 6-K or 8-K." }, { "persona": "capital-markets-credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.43, "rationale": "Only one normal quarterly filing cycle remains, and the plausible public-neocloud pool beyond CoreWeave is small. SEC accounting notes commonly disclose GPU depreciation lives, while customer-contract disclosures may provide duration, but issuers rarely make the explicit comparison needed to establish a tenor mismatch. Nebius or another qualifying issuer could add it amid rapid RPO growth, leverage, and financing scrutiny, yet foreign-issuer reporting and year-end filing calendars reduce opportunities before December 31. Ambiguity resolving against the forecaster further lowers the chance that separate or inferential disclosures qualify. This is below the forecaster’s 0.69 and closer to the >90-day reference-class rate.", "keyEvidence": [ "https://www.sec.gov/edgar/search/" ], "whatWouldChangeMyMind": "A filed draft, earnings release, or financing document previewing both GPU depreciation life and weighted-average customer-contract tenor for Nebius or another clearly qualifying public neocloud." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.36, "rationale": "I use 0.50, the ledger’s disclosure base rate; the >90-day base rate is lower at 0.40. The 0.708 capital-lens rate is less diagnostic because resolution requires another eligible public neocloud, an explicit GPU useful-life-versus-contract-tenor mismatch, an SEC filing, and year-end timing. The digest shows substantial capital commitments but no qualifying tenor disclosure. Only about one normal quarterly filing cycle remains, and ambiguity resolves adversely. The forecaster also shows overconfidence: its 0.60–0.69 bin realized 0.556 versus 0.663 stated. These disclosure and timing hurdles put the event below the generic disclosure base rate.", "keyEvidence": [ "https://www.sec.gov/edgar/search/" ], "whatWouldChangeMyMind": "Issuer guidance that an upcoming 10-Q or 6-K will explicitly compare GPU useful life with customer-contract duration would raise my estimate; a narrower eligible-issuer definition would lower it." }, { "persona": "geopolitics-regulation-analyst", "model": "openai/gpt-5.6-sol", "p": 0.46, "rationale": "Only one normal quarterly filing window remains, and several plausible candidates are foreign issuers whose 6-K disclosures are often less detailed than a 10-Q. Rising capital intensity and the digest’s large NBIS debt/RPO figures increase pressure to explain whether GPU depreciation extends beyond contracted revenue. However, an SEC filing must disclose the mismatch clearly enough to satisfy the literal rule; merely listing useful lives and contract durations separately may not qualify. Classification as a named neocloud is another ambiguity that resolves against a hit. These constraints outweigh the favorable capital-lens base rate and put the forecast near the >90-day and disclosure reference rates.", "keyEvidence": [ "https://www.sec.gov/edgar/search/", "https://fred.stlouisfed.org/series/BAMLH0A3HYC" ], "whatWouldChangeMyMind": "A candidate’s draft registration statement, SEC comment response, or earnings materials signaling an imminent filing that explicitly compares GPU useful lives with weighted-average customer contract tenor." } ], "dissent": null, "spread": 0.1 }, "final": { "p": 0.5355399996153618, "method": "stacked-fixed-weights (resolved n=57 < 80)", "weights": { "calibrated": 0.3, "council": 0.5, "model": 0.2 }, "ci80": [ 0.4600586015616395, 0.6124283719953265 ] }, "direction": "down", "drivers": [ "Reference class qtype:disclosure runs 0.50 (n=29) vs stated 0.69.", "Hierarchical calibration moves stated 0.69 to 0.56 (lens capital, kappa_stated=24.0, kappa_lens=200.0).", "Feature model at 0.77: conjuncts=1, named disclosure=True, horizon=138d.", "Council of 5 at 0.41." ], "warnings": [ "all 5 votes came from one vendor (openai); extremization disabled, persona diversity only" ], "evidence": [ { "url": "https://www.sec.gov/edgar/search/", "title": "SEC.gov | EDGAR Full Text Search", "publishedAt": null, "source": "direct", "excerpt": "Cited by grid-and-power-economist: The event requires more than separate depreciation and RPO disclosures: a second named public neocloud must disclose the GPU-life/customer-contract-tenor mismatch in an SEC filing. Remaining opportunities are mainly third-quarter reports and financing or registrati" }, { "url": "https://fred.stlouisfed.org/series/BAMLH0A3HYC", "title": "ICE BofA CCC & Lower US High Yield Index Option-Adjusted Spread (BAMLH0A3HYC) | FRED | St. Louis Fed", "publishedAt": null, "source": "direct", "excerpt": "Cited by geopolitics-regulation-analyst: Only one normal quarterly filing window remains, and several plausible candidates are foreign issuers whose 6-K disclosures are often less detailed than a 10-Q. Rising capital intensity and the digest’s large NBIS debt/RPO figures increase pressure to explain" } ], "bubbleBasket": "residual_duration", "daysRemaining": 116, "deadline": "2026-12-31", "lens": "capital" } }, { "id": "lq-p86", "weeklyPredictionId": "p86-ocp-silicon-photonics-spec-dec31", "text": "OCP Open Silicon Photonics for AI Systems submits its first specification by December 31, 2026.", "lens": "networking", "deadline": "2026-12-31", "resolutionSource": "OCP publication.", "resolutionRule": "Hit on a submitted spec, not a workstream announcement.", "ambiguityPolicy": "resolves_against_forecaster", "stated": 0.61, "current": 0.4546037678858174, "direction": "down", "history": [ { "asOf": "2026-08-15", "p": 0.61, "source": "stated", "runId": null }, { "asOf": "2026-09-06", "p": 0.4546037678858174, "source": "engine", "runId": "2026-09-06T22-42-59Z-51a6cc" } ], "engine": { "id": "lq-p86", "weeklyPredictionId": "p86-ocp-silicon-photonics-spec-dec31", "stated": 0.61, "referenceClass": { "p": 0.3888888888888889, "className": "qtype:timing", "n": 8, "note": "most specific ledger class with n >= 8; Resolved predictions of type timing: disclosure = requires a named primary document or named company; threshold = numeric bar without a named-disclosure requirement; timing = event-by-date without a numeric bar; other = none of the above." }, "calibrated": { "p": 0.532724546408757, "method": "hierarchical-beta-binomial-by-lens", "ci80": [ 0.4908945518101038, 0.5744340223913876 ], "variants": { "platt-on-stated-logit": 0.5053 } }, "model": { "p": 0.39708792178128355, "method": "logistic-features-l2", "ci80": [ 0.18353378857695185, 0.6215881301335289 ], "features": { "conjuncts": 1, "hasNumericThreshold": false, "horizonDays": 138, "isMarketPriceQuestion": false, "isTimingQuestion": true, "requiresNamedPrimaryDisclosure": false, "statedConfidence": 0.61, "thresholdMagnitudeLog": 0, "weekIndex": 16 }, "gbm": 0.395 }, "council": { "p": 0.43148900977101473, "aggregation": "log-odds-mean", "extremization": 1, "votes": [ { "persona": "grid-and-power-economist", "model": "openai/gpt-5.6-sol", "p": 0.44, "rationale": "The remaining window includes OCP’s usual year-end standards activity, and strong AI-networking demand gives participants reason to complete an initial deliverable. However, the rule requires an actually submitted specification documented by OCP; a project launch, roadmap, draft discussion, or workstream announcement does not qualify. First specifications often slip because scope, interoperability, and contributor-IP issues require consensus. With only 116 days and no qualifying submission identified in the supplied digest, 0.61 is too optimistic; the networking and timing reference classes also favor a below-even estimate.", "keyEvidence": [ "https://www.opencompute.org/" ], "whatWouldChangeMyMind": "An OCP project page naming a completed draft, formal review ballot, or scheduled specification submission before year-end would raise the estimate materially." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.48, "rationale": "The remaining 116 days include OCP’s usual fall summit window, providing a natural milestone, and the rule requires submission rather than approval or adoption. However, first specifications from new standards workstreams commonly slip as scope, contributor IP terms, interoperability requirements, and draft consensus develop. The strict requirement for an OCP publication also creates documentary risk: partner announcements or roadmap presentations alone would fail. Networking and timing reference classes are below 0.4, while the forecaster’s 0.60–0.69 calls have realized near 0.56, supporting a material discount from 0.61.", "keyEvidence": [ "https://www.opencompute.org/" ], "whatWouldChangeMyMind": "An OCP agenda or project update naming a completed draft and a 2026 submission vote would raise this substantially; evidence that requirements remain unsettled by the fall summit would lower it." }, { "persona": "capital-markets-credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.44, "rationale": "The remaining 116 days include OCP’s typical autumn publication window, and submission is a lower hurdle than ratification or final adoption. However, the rule requires an actual specification submission documented by OCP; a project launch, roadmap, draft discussion, or workstream announcement does not qualify. The digest provides no evidence of a completed draft, review milestone, or firm submission date. Networking and timing reference classes also resolve below 0.4, while the forecaster’s 0.60–0.69 band has historically underperformed its stated confidence. I therefore put this below even odds despite a plausible year-end push.", "keyEvidence": [ "https://www.opencompute.org/" ], "whatWouldChangeMyMind": "An OCP agenda or project update naming a completed draft, formal review, or scheduled specification submission before year-end would raise the estimate materially." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.36, "rationale": "I use the networking-ledger base rate of 0.318, cross-checked against timing (0.389) and >90-day (0.400) classes. This is single-leg, and 116 days is meaningful runway, warranting a small upward adjustment. However, the rule requires an actual submitted specification—not an announcement—and verifiable OCP publication by a hard date. The digest provides no project-specific evidence that a draft is complete or entering formal submission. Standards schedules commonly slip, while ambiguity resolves against a hit. The forecaster’s 0.60–0.69 band has realized 0.556, indicating overconfidence and offering no basis for 0.61 here.", "keyEvidence": [ "https://www.opencompute.org/" ], "whatWouldChangeMyMind": "A dated OCP draft, meeting record, or official roadmap showing completed technical text and a scheduled formal submission before December 31 would raise the estimate substantially." }, { "persona": "geopolitics-regulation-analyst", "model": "openai/gpt-5.6-sol", "p": 0.44, "rationale": "The remaining window includes OCP’s usual fall summit cycle, when working groups often publish or advance specifications. However, the supplied record shows no draft, review milestone, ballot, or firm submission schedule. This is a first specification, which carries added coordination and technical-consensus risk, and the rule excludes a workstream announcement. Strict reliance on an OCP publication also raises documentation risk near the deadline. The most relevant ledger classes—networking, timing, and >90-day horizon—cluster around 0.32–0.40, supporting a below-even estimate despite the summit catalyst.", "keyEvidence": [ "https://www.opencompute.org/" ], "whatWouldChangeMyMind": "An OCP draft repository, formal review notice, or agenda naming the specification and a 2026 submission vote would raise the estimate materially." } ], "dissent": null, "spread": 0.12 }, "final": { "p": 0.4546037678858174, "method": "stacked-fixed-weights (resolved n=57 < 80)", "weights": { "calibrated": 0.3, "council": 0.5, "model": 0.2 }, "ci80": [ 0.3550012272948749, 0.5372377029429284 ] }, "direction": "down", "drivers": [ "Reference class qtype:timing runs 0.39 (n=8) vs stated 0.61.", "Hierarchical calibration moves stated 0.61 to 0.53 (lens networking, kappa_stated=24.0, kappa_lens=200.0).", "Feature model at 0.40: conjuncts=1, named disclosure=False, horizon=138d.", "Council of 5 at 0.43." ], "warnings": [ "all 5 votes came from one vendor (openai); extremization disabled, persona diversity only" ], "evidence": [ { "url": "https://www.opencompute.org/", "title": "Home » Open Compute Project", "publishedAt": null, "source": "direct", "excerpt": "Cited by grid-and-power-economist: The remaining window includes OCP’s usual year-end standards activity, and strong AI-networking demand gives participants reason to complete an initial deliverable. However, the rule requires an actually submitted specification documented by OCP; a project launch, " } ], "bubbleBasket": null, "daysRemaining": 116, "deadline": "2026-12-31", "lens": "networking" } }, { "id": "lq-p92", "weeklyPredictionId": "p92-second-state-dc-grid-rule-dec31", "text": "A second US state with major hyperscale exposure publishes a data-center grid rule by December 31, 2026.", "lens": "power", "deadline": "2026-12-31", "resolutionSource": "State PUC or legislature.", "resolutionRule": "Hit for a second state among VA, TX, GA, OR, IL, AZ, OH beyond those already counted at prediction time.", "ambiguityPolicy": "resolves_against_forecaster", "bubbleBasket": "power_policy", "stated": 0.39, "current": 0.48675909098172265, "direction": "up", "history": [ { "asOf": "2026-08-22", "p": 0.39, "source": "stated", "runId": null }, { "asOf": "2026-09-06", "p": 0.48675909098172265, "source": "engine", "runId": "2026-09-06T22-42-59Z-51a6cc" } ], "engine": { "id": "lq-p92", "weeklyPredictionId": "p92-second-state-dc-grid-rule-dec31", "stated": 0.39, "referenceClass": { "p": 0.3888888888888889, "className": "qtype:timing", "n": 8, "note": "most specific ledger class with n >= 8; Resolved predictions of type timing: disclosure = requires a named primary document or named company; threshold = numeric bar without a named-disclosure requirement; timing = event-by-date without a numeric bar; other = none of the above." }, "calibrated": { "p": 0.5235857186826716, "method": "hierarchical-beta-binomial-by-lens", "ci80": [ 0.48118220776689863, 0.5659000673369444 ], "variants": { "platt-on-stated-logit": 0.3808 } }, "model": { "p": 0.6946398970126398, "method": "logistic-features-l2", "ci80": [ 0.3973951897219935, 0.929293559911551 ], "features": { "conjuncts": 1, "hasNumericThreshold": false, "horizonDays": 131, "isMarketPriceQuestion": false, "isTimingQuestion": true, "requiresNamedPrimaryDisclosure": false, "statedConfidence": 0.39, "thresholdMagnitudeLog": 0, "weekIndex": 17 }, "gbm": 0.6157 }, "council": { "p": 0.3795731108232087, "aggregation": "log-odds-mean", "extremization": 1, "votes": [ { "persona": "grid-and-power-economist", "model": "openai/gpt-5.6-sol", "p": 0.36, "rationale": "The multi-state opportunity set helps, but only 116 days remain and most legislatures are out of regular session. A qualifying second state therefore likely requires an already-advanced PUC docket or a scheduled final order, not merely a proposal, utility tariff filing, or policy announcement. Notice-and-comment timelines and year-end commission calendars make slippage material. The strict, ambiguity-against-forecaster rule also leaves uncertainty over what constitutes a “data-center grid rule” and which state was already counted. I am slightly below the forecaster’s 0.39.", "keyEvidence": [], "whatWouldChangeMyMind": "A docket showing a proposed or final qualifying rule with comments completed and an order date before December 31 would raise this substantially; absence of any active docket by October would lower it." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.44, "rationale": "Multiple candidate states face acute large-load interconnection and cost-allocation pressure, giving PUCs a plausible path to publish tariffs or rules before year-end. Texas and Virginia are the strongest procedural candidates, while Ohio and Georgia provide additional opportunities. However, only 116 days remain, most legislatures are out of regular session, and commission dockets frequently slip into the next calendar year. The requirement that this be a second qualifying state—and ambiguity resolving against the forecast—also means a proposal, utility tariff, or generic large-load action may not count. I place the probability modestly above the prior forecaster estimate but below even odds.", "keyEvidence": [], "whatWouldChangeMyMind": "A scheduled final-order date in a qualifying state would raise it materially; evidence that all active proceedings target 2027, or cover generic loads without data-center applicability, would lower it." }, { "persona": "capital-markets-credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.38, "rationale": "Only one additional qualifying state is needed, and several listed PUCs face acute large-load interconnection and cost-allocation issues. However, just 116 days remain; most 2026 legislatures have adjourned, leaving PUC proceedings as the main path. Final tariffs or orders commonly slip across year-end, while proposals, utility filings, studies, or broadly applicable grid reforms may not satisfy the strict “data-center grid rule” language. The timing reference-class rate near 0.39 is more relevant than the small power sample. I therefore place the probability close to, but slightly below, the forecaster’s 0.39.", "keyEvidence": [ "https://www.scc.virginia.gov/", "https://puco.ohio.gov/" ], "whatWouldChangeMyMind": "A scheduled final vote or order in an active data-center-specific tariff or interconnection rulemaking would raise this substantially; evidence that all candidate proceedings extend into 2027 would lower it." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.36, "rationale": "Base rate: 0.40, combining the forecaster’s >90-day (0.40) and timing-question (0.389) classes; the 0.70 power-lens rate has only four observations and receives little weight. Multiple eligible states create opportunity, but only 116 days remain and resolution requires formal publication by a state PUC or legislature—not proposals, workshops, utility tariffs, or post-deadline adoption. Legislative calendars, rulemaking notice periods, and the ambiguity-against-forecaster policy warrant a modest discount. The forecaster’s 0.39 is close; their low-confidence calibration bin has only one case and provides no reliable correction.", "keyEvidence": [], "whatWouldChangeMyMind": "A qualifying state PUC’s final-order agenda, adopted rule awaiting publication, or enacted bill with an effective publication date before December 31 would raise it substantially." }, { "persona": "geopolitics-regulation-analyst", "model": "openai/gpt-5.6-sol", "p": 0.36, "rationale": "The strongest reference classes are timing (0.389) and >90-day horizon (0.400). By September, most relevant legislatures have finished regular sessions, leaving fall PUC orders, Ohio’s continuing session, Illinois’s veto session, or special-session action. Grid-capacity pressure and policy clustering create several plausible paths, but the rule must be officially published by year-end and involve a newly qualifying state; proposals, studies, utility filings, or ambiguous general large-load measures may not count. With only 116 days and ambiguity resolving against the forecast, I place this modestly below the timing base rate.", "keyEvidence": [ "https://www.scc.virginia.gov/", "https://puco.ohio.gov/", "https://www.icc.illinois.gov/" ], "whatWouldChangeMyMind": "A scheduled final PUC vote, proposed final order, or enrolled fall-session bill in an eligible uncounted state would raise this materially; docket delays into 2027 would lower it." } ], "dissent": null, "spread": 0.08 }, "final": { "p": 0.48675909098172265, "method": "stacked-fixed-weights (resolved n=57 < 80)", "weights": { "calibrated": 0.3, "council": 0.5, "model": 0.2 }, "ci80": [ 0.40286536798066325, 0.616363027137231 ] }, "direction": "up", "drivers": [ "Reference class qtype:timing runs 0.39 (n=8) vs stated 0.39.", "Hierarchical calibration moves stated 0.39 to 0.52 (lens power, kappa_stated=24.0, kappa_lens=200.0).", "Feature model at 0.69: conjuncts=1, named disclosure=False, horizon=131d.", "Council of 5 at 0.38." ], "warnings": [ "all 5 votes came from one vendor (openai); extremization disabled, persona diversity only", "2 cited URL(s) failed verification and were dropped." ], "evidence": [ { "url": "https://www.scc.virginia.gov/", "title": "Virginia SCC - Home", "publishedAt": null, "source": "direct", "excerpt": "Cited by capital-markets-credit-analyst: Only one additional qualifying state is needed, and several listed PUCs face acute large-load interconnection and cost-allocation issues. However, just 116 days remain; most 2026 legislatures have adjourned, leaving PUC proceedings as the main path. Final tar" }, { "url": "https://puco.ohio.gov/", "title": "Public Utilities Commission of Ohio", "publishedAt": null, "source": "direct", "excerpt": "Cited by capital-markets-credit-analyst: Only one additional qualifying state is needed, and several listed PUCs face acute large-load interconnection and cost-allocation issues. However, just 116 days remain; most 2026 legislatures have adjourned, leaving PUC proceedings as the main path. Final tar" }, { "url": "https://www.icc.illinois.gov/", "title": "Illinois Commerce Commission", "publishedAt": null, "source": "direct", "excerpt": "Cited by geopolitics-regulation-analyst: The strongest reference classes are timing (0.389) and >90-day horizon (0.400). By September, most relevant legislatures have finished regular sessions, leaving fall PUC orders, Ohio’s continuing session, Illinois’s veto session, or special-session action. Gr" } ], "bubbleBasket": "power_policy", "daysRemaining": 116, "deadline": "2026-12-31", "lens": "power" } }, { "id": "lq-p102", "weeklyPredictionId": "p102-second-queue-contract-dec31", "text": "A second AI infrastructure contract above $500M discloses both a future service date and a queue position by December 31, 2026.", "lens": "capital", "deadline": "2026-12-31", "resolutionSource": "8-K, PSA, or ISO queue filing cross-walked in the same disclosure.", "resolutionRule": "Hit requires both a service date and a queue identifier. Either alone is partial.", "ambiguityPolicy": "partial_credit_if_one_leg_hits", "bubbleBasket": "interconnect_conversion", "stated": 0.68, "current": 0.461362737805638, "direction": "down", "history": [ { "asOf": "2026-09-05", "p": 0.68, "source": "stated", "runId": null }, { "asOf": "2026-09-06", "p": 0.461362737805638, "source": "engine", "runId": "2026-09-06T22-42-59Z-51a6cc" } ], "engine": { "id": "lq-p102", "weeklyPredictionId": "p102-second-queue-contract-dec31", "stated": 0.68, "referenceClass": { "p": 0.5, "className": "qtype:disclosure", "n": 29, "note": "most specific ledger class with n >= 8; Resolved predictions of type disclosure: disclosure = requires a named primary document or named company; threshold = numeric bar without a named-disclosure requirement; timing = event-by-date without a numeric bar; other = none of the above." }, "calibrated": { "p": 0.558883165360209, "method": "hierarchical-beta-binomial-by-lens", "ci80": [ 0.5172964427726934, 0.600253961043671 ], "variants": { "platt-on-stated-logit": 0.5486 } }, "model": { "p": 0.29978723740562, "method": "logistic-features-l2", "ci80": [ 0.17095381686254466, 0.49525133642259583 ], "features": { "conjuncts": 3, "hasNumericThreshold": true, "horizonDays": 117, "isMarketPriceQuestion": false, "isTimingQuestion": true, "requiresNamedPrimaryDisclosure": true, "statedConfidence": 0.68, "thresholdMagnitudeLog": 8.699, "weekIndex": 19 }, "gbm": 0.442 }, "council": { "p": 0.4719371548297334, "aggregation": "log-odds-mean", "extremization": 1, "votes": [ { "persona": "grid-and-power-economist", "model": "openai/gpt-5.6-sol", "p": 0.46, "rationale": "AI capex and RPO growth create many opportunities for another contract above $500M, but the resolution standard is unusually document-specific. Corporate filings commonly provide contract value or expected service timing; they rarely include an ISO queue identifier, while queue records often cannot be unambiguously cross-walked to the contract in the same disclosure. The “second” occurrence and 116-day window add hurdles. Partial credit remains plausible because a service date is substantially more likely than both legs. The multi-leg and >90-day ledger rates also support a forecast materially below 0.68.", "keyEvidence": [ "https://www.sec.gov/edgar/search/" ], "whatWouldChangeMyMind": "A filed 8-K or PSA announcing a qualifying contract with a dated energization milestone, or an imminent ISO/PUC filing explicitly linking the customer contract to a queue project." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.47, "rationale": "AI capex and RPO growth support additional contracts above $500M, but the documentary conjunction is difficult. Service dates often appear in PSAs or customer filings; ISO queue identifiers usually remain in separate interconnection records and may not be cross-walked in the same disclosure. Requiring a second qualifying contract within 116 days further reduces the full-hit chance. Partial credit materially supports the estimate because a service date alone is plausible. The multi-leg and >90-day ledger rates also argue below the forecaster’s 0.68.", "keyEvidence": [ "https://www.sec.gov/edgar/search/" ], "whatWouldChangeMyMind": "A filed $500M-plus agreement naming a future service date, or an announced transaction explicitly linking its site to a public ISO queue identifier, would raise the estimate sharply." }, { "persona": "capital-markets-credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.44, "rationale": "The AI-capex and RPO backdrop supports additional contracts above $500M, and the deadline includes a quarterly filing cycle. However, the test requires a second qualifying contract and an unusually specific cross-walk: future service date plus ISO queue identifier in the same 8-K, PSA, or queue filing. Commercial disclosures commonly provide timing but omit or redact queue identifiers. Thus partial credit is materially more likely than a full hit. The conjunctive and >90-day ledger base rates near 0.4 are more relevant than the stronger capital-lens rate. This makes 0.68 too high despite favorable financing markets.", "keyEvidence": [ "https://www.sec.gov/edgar/search/" ], "whatWouldChangeMyMind": "A filed agreement naming a >$500M commitment with either the service date or queue ID, or an announced near-term filing explicitly promising the missing cross-walk." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.46, "rationale": "I use the 0.433 multi-leg base rate, which already reflects partial scoring; the disclosure and >90-day classes are similarly modest at 0.500 and 0.400. Exceptional capex and RPO growth supports more large AI contracts, modestly raising the outside-view estimate. But resolution requires a second contract above $500M plus two unusually specific details—a future service date and queue identifier—in a qualifying, cross-walked primary disclosure by a hard date. Contract occurrence is likelier than timely disclosure. The forecaster’s 0.60–0.69 bin realized only 0.556, indicating overconfidence, and this question is stricter than that average.", "keyEvidence": [ "https://www.sec.gov/edgar/search/" ], "whatWouldChangeMyMind": "A qualifying filing identifying a candidate second contract above $500M and either required leg, with an official cross-walk making the remaining leg likely before December 31." }, { "persona": "geopolitics-regulation-analyst", "model": "openai/gpt-5.6-sol", "p": 0.53, "rationale": "Capital intensity supports additional large AI contracts: several issuers show exceptional RPO and capex growth. But the decisive documentary requirement is unusually strict. A second contract must exceed $500M and pair a future service date with an identifiable ISO queue position in the same cross-walkable disclosure. Companies commonly disclose timing without queue identifiers, while ISO filings may omit contract value or customer linkage. The relevant ledger rates are only 0.433 for multi-leg questions and 0.400 beyond 90 days, versus 0.708 for capital. Partial credit and one remaining quarterly filing cycle keep the expected outcome slightly above 0.5, but 0.68 appears too high.", "keyEvidence": [ "https://www.sec.gov/edgar/search/" ], "whatWouldChangeMyMind": "A filed contract announcement naming an ISO, project, or interconnection study—or evidence that one qualifying contract already exists—would materially raise the estimate." } ], "dissent": null, "spread": 0.09 }, "final": { "p": 0.461362737805638, "method": "stacked-fixed-weights (resolved n=57 < 80)", "weights": { "calibrated": 0.3, "council": 0.5, "model": 0.2 }, "ci80": [ 0.3975716672078132, 0.5444399543973634 ] }, "direction": "down", "drivers": [ "Reference class qtype:disclosure runs 0.50 (n=29) vs stated 0.68.", "Hierarchical calibration moves stated 0.68 to 0.56 (lens capital, kappa_stated=24.0, kappa_lens=200.0).", "Feature model at 0.30: conjuncts=3, named disclosure=True, horizon=117d.", "Council of 5 at 0.47." ], "warnings": [ "all 5 votes came from one vendor (openai); extremization disabled, persona diversity only", "1 cited URL(s) failed verification and were dropped." ], "evidence": [ { "url": "https://www.sec.gov/edgar/search/", "title": "SEC.gov | EDGAR Full Text Search", "publishedAt": null, "source": "direct", "excerpt": "Cited by grid-and-power-economist: AI capex and RPO growth create many opportunities for another contract above $500M, but the resolution standard is unusually document-specific. Corporate filings commonly provide contract value or expected service timing; they rarely include an ISO queue identifier" } ], "bubbleBasket": "interconnect_conversion", "daysRemaining": 116, "deadline": "2026-12-31", "lens": "capital" } }, { "id": "lq-p103", "weeklyPredictionId": "p103-fabric-one-control-schema-dec31", "text": "Equinix publishes Fabric One beta documentation exposing approval, rollback, and schema controls by December 31, 2026.", "lens": "networking", "deadline": "2026-12-31", "resolutionSource": "Equinix docs.", "resolutionRule": "Hit if public docs expose approval, rollback, and schema. Marketing pages without those three miss.", "ambiguityPolicy": "resolves_against_forecaster", "stated": 0.57, "current": 0.3782631430805243, "direction": "down", "history": [ { "asOf": "2026-09-05", "p": 0.57, "source": "stated", "runId": null }, { "asOf": "2026-09-06", "p": 0.3782631430805243, "source": "engine", "runId": "2026-09-06T22-42-59Z-51a6cc" } ], "engine": { "id": "lq-p103", "weeklyPredictionId": "p103-fabric-one-control-schema-dec31", "stated": 0.57, "referenceClass": { "p": 0.3888888888888889, "className": "qtype:timing", "n": 8, "note": "most specific ledger class with n >= 8; Resolved predictions of type timing: disclosure = requires a named primary document or named company; threshold = numeric bar without a named-disclosure requirement; timing = event-by-date without a numeric bar; other = none of the above." }, "calibrated": { "p": 0.5286219823061928, "method": "hierarchical-beta-binomial-by-lens", "ci80": [ 0.48677832399414167, 0.5703602309683993 ], "variants": { "platt-on-stated-logit": 0.4819 } }, "model": { "p": 0.2934703459896904, "method": "logistic-features-l2", "ci80": [ 0.1261546217316851, 0.5075707200842156 ], "features": { "conjuncts": 2, "hasNumericThreshold": false, "horizonDays": 117, "isMarketPriceQuestion": false, "isTimingQuestion": true, "requiresNamedPrimaryDisclosure": false, "statedConfidence": 0.57, "thresholdMagnitudeLog": 0, "weekIndex": 19 }, "gbm": 0.3563 }, "council": { "p": 0.32933678715053505, "aggregation": "log-odds-mean", "extremization": 1, "votes": [ { "persona": "grid-and-power-economist", "model": "openai/gpt-5.6-sol", "p": 0.36, "rationale": "The rule is a strict three-leg documentation test, not merely a Fabric One beta launch. Public beta documentation could appear within 116 days, but approval workflows, rollback behavior, and schema controls must all be exposed explicitly; omission of any one is a miss. No supplied evidence identifies an existing documentation page or dated publication commitment. Relevant ledger classes are weak: networking 0.318, timing 0.389, and multi-leg 0.433. The forecaster’s 0.50–0.59 band has also realized only 0.367, supporting a substantial discount from 0.57.", "keyEvidence": [], "whatWouldChangeMyMind": "A live Equinix documentation branch or dated beta roadmap explicitly covering approval workflows, rollback semantics, and schema/version controls would raise the estimate sharply." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.32, "rationale": "The deadline leaves only 116 days, and the supplied public digest provides no product-roadmap evidence that Fabric One documentation or these controls are imminent. Beta documentation could accompany a late-year launch, but the rule requires public technical documentation explicitly exposing all three capabilities. Approval and schema controls are plausible early features; documented rollback is a harder leg and may arrive later or remain operational rather than user-facing. The strict conjunctive test and ambiguity policy materially lower the odds. Networking, timing, and similar-confidence reference classes also resolve near 0.29–0.40, making 0.57 look optimistic.", "keyEvidence": [], "whatWouldChangeMyMind": "Official preview documentation or release notes naming Fabric One and explicitly describing approval workflows, rollback/version restoration, and schema validation would raise the estimate sharply." }, { "persona": "capital-markets-credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.34, "rationale": "The deadline allows roughly four months, but the rule requires a public beta documentation release plus three distinct capabilities. Approval and schema controls are plausible for governed network automation; documented rollback is the harder leg and may arrive later or be described too vaguely. Beta documentation often trails product announcements, and marketing material cannot qualify. The networking, timing, and multi-leg reference classes cluster around 0.32–0.43, while the forecaster’s 0.50–0.59 calls have realized materially below their stated confidence. These factors put the estimate well below 0.57.", "keyEvidence": [], "whatWouldChangeMyMind": "A dated Equinix documentation preview or changelog naming Fabric One and explicitly demonstrating all three controls would raise the estimate substantially." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.28, "rationale": "I use the networking-ledger base rate of 0.318. The 116-day window is plausible, but resolution requires public beta documentation—not marketing—and all three distinct controls: approval, rollback, and schema. That conjunction plus ambiguity resolving against the forecast makes documentation timing a substantial failure mode. No digest evidence specifically verifies Fabric One documentation or these controls. The forecaster’s 0.50–0.59 calls have realized only 0.367 overall, and the narrower confidence-class base rate is 0.286, indicating material overconfidence at 0.57.", "keyEvidence": [], "whatWouldChangeMyMind": "Official Equinix documentation or a dated documentation roadmap explicitly showing all three controls would raise this sharply; evidence that any control is deferred beyond 2026 would lower it." }, { "persona": "geopolitics-regulation-analyst", "model": "openai/gpt-5.6-sol", "p": 0.35, "rationale": "The strict three-part documentation test is harder than a general beta launch: approval workflows, rollback behavior, and schema controls must all appear in public technical docs, while marketing or partial documentation fails. Only 116 days remain, spanning a year-end release window where documentation and governance features commonly slip. The networking (0.318), timing (0.389), and multi-leg (0.433) reference classes all argue below 0.57; the forecaster’s 0.50–0.59 band has also realized at 0.367.", "keyEvidence": [], "whatWouldChangeMyMind": "A dated Equinix documentation roadmap or live Fabric One beta reference explicitly covering all three controls would raise the estimate sharply; evidence that any control is deferred beyond 2026 would lower it." } ], "dissent": null, "spread": 0.08 }, "final": { "p": 0.3782631430805243, "method": "stacked-fixed-weights (resolved n=57 < 80)", "weights": { "calibrated": 0.3, "council": 0.5, "model": 0.2 }, "ci80": [ 0.2941776910878261, 0.4510005889892484 ] }, "direction": "down", "drivers": [ "Reference class qtype:timing runs 0.39 (n=8) vs stated 0.57.", "Hierarchical calibration moves stated 0.57 to 0.53 (lens networking, kappa_stated=24.0, kappa_lens=200.0).", "Feature model at 0.29: conjuncts=2, named disclosure=False, horizon=117d.", "Council of 5 at 0.33." ], "warnings": [ "all 5 votes came from one vendor (openai); extremization disabled, persona diversity only", "3 cited URL(s) failed verification and were dropped." ], "evidence": [], "bubbleBasket": null, "daysRemaining": 116, "deadline": "2026-12-31", "lens": "networking" } }, { "id": "lq-p69", "weeklyPredictionId": "p69-camellia-curtailment-template-jan31", "text": "A second US multi-gigawatt AI campus publicly commits to at least 250 MW of utility curtailment flexibility by January 31, 2027.", "lens": "power", "deadline": "2027-01-31", "resolutionSource": "Utility filing or campus IR.", "resolutionRule": "Hit on a second campus besides Camellia, ≥250 MW curtailment commitment.", "ambiguityPolicy": "resolves_against_forecaster", "bubbleBasket": "energization", "stated": 0.49, "current": 0.4561350514532453, "direction": "down", "history": [ { "asOf": "2026-07-25", "p": 0.49, "source": "stated", "runId": null }, { "asOf": "2026-09-06", "p": 0.4561350514532453, "source": "engine", "runId": "2026-09-06T22-42-59Z-51a6cc" } ], "engine": { "id": "lq-p69", "weeklyPredictionId": "p69-camellia-curtailment-template-jan31", "stated": 0.49, "referenceClass": { "p": 0.5, "className": "qtype:disclosure", "n": 29, "note": "most specific ledger class with n >= 8; Resolved predictions of type disclosure: disclosure = requires a named primary document or named company; threshold = numeric bar without a named-disclosure requirement; timing = event-by-date without a numeric bar; other = none of the above." }, "calibrated": { "p": 0.5341120344721453, "method": "hierarchical-beta-binomial-by-lens", "ci80": [ 0.4917404968464171, 0.576354616801809 ], "variants": { "platt-on-stated-logit": 0.4365 } }, "model": { "p": 0.5332676328012661, "method": "logistic-features-l2", "ci80": [ 0.24202531343501052, 0.8911311362078943 ], "features": { "conjuncts": 1, "hasNumericThreshold": true, "horizonDays": 190, "isMarketPriceQuestion": false, "isTimingQuestion": true, "requiresNamedPrimaryDisclosure": true, "statedConfidence": 0.49, "thresholdMagnitudeLog": 2.398, "weekIndex": 13 }, "gbm": 0.2466 }, "council": { "p": 0.3805705392903609, "aggregation": "log-odds-mean", "extremization": 1, "votes": [ { "persona": "grid-and-power-economist", "model": "openai/gpt-5.6-sol", "p": 0.38, "rationale": "Large-load tariffs increasingly seek dispatchability, and Camellia provides a precedent, but the resolution requires another specifically identified multi-gigawatt campus, a quantified commitment of at least 250 MW, and publication through utility filing or campus IR within 147 days. Generic demand-response eligibility, proposed tariff authority, or an unquantified promise would fail under the ambiguity rule. The supplied digest identifies no second candidate, docket, or scheduled filing. Queue delays do not directly preclude a paper commitment, but they reduce utilities’ need to finalize campus-specific operating terms quickly. Disclosure and >90-day reference classes support a middling prior, with a haircut for the exact threshold and document requirements.", "keyEvidence": [], "whatWouldChangeMyMind": "A pending utility docket naming another multi-gigawatt campus and proposing at least 250 MW of mandatory or contracted curtailment, especially with an order or filing deadline before January 31." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.42, "rationale": "Grid constraints increasingly push utilities to require large AI loads to offer interruptibility, and several multi-gigawatt projects could reach regulatory approval within 147 days. However, the resolution standard is demanding: a second campus must be identified as multi-gigawatt, explicitly commit at least 250 MW—not merely discuss demand response—and do so in a utility filing or campus IR. Special contracts often redact operational terms, while announcements use vague language such as “flexible load.” The short window and ambiguity-against policy therefore keep the probability below even odds.", "keyEvidence": [], "whatWouldChangeMyMind": "A pending utility docket containing an unredacted special contract with a quantified 250 MW-plus interruptible tranche would raise it sharply; confidentiality orders or sub-threshold flexibility terms would lower it." }, { "persona": "capital-markets-credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.41, "rationale": "Large-load tariffs and grid-reliability concerns make substantial curtailment provisions increasingly plausible. However, resolution requires a second distinct campus to be publicly established as multi-gigawatt, commit at least 250 MW—not merely discuss flexible load—and have that commitment appear in a utility filing or campus IR within 147 days. Such agreements are often confidential, redacted, expressed without a firm MW amount, or disclosed only at the utility-program level. The strict source, threshold, and ambiguity rules put this below the forecaster’s 0.49 despite favorable power-sector momentum.", "keyEvidence": [], "whatWouldChangeMyMind": "A pending utility docket naming another multi-gigawatt AI campus and proposing a binding interruptible-load amount of at least 250 MW would raise my estimate materially." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.29, "rationale": "I use the >90-day base rate of 0.40; the power-lens rate of 0.70 has only four observations and receives little weight. I adjust downward because success requires a second qualifying multi-gigawatt campus, a large explicit ≥250 MW commitment, and disclosure in a utility filing or campus IR within 147 days. Disclosure timing and ambiguity materially increase failure risk, while the digest provides no specific evidence of an imminent qualifying filing. The forecaster’s broader calibration shows overconfidence near and above 0.50, reinforcing a discount from 0.49.", "keyEvidence": [], "whatWouldChangeMyMind": "A pending utility docket, tariff agreement, or official campus announcement explicitly quantifying at least 250 MW of curtailable load would raise the estimate substantially." }, { "persona": "geopolitics-regulation-analyst", "model": "openai/gpt-5.6-sol", "p": 0.41, "rationale": "Grid-stress concerns and emerging large-load tariffs make another sizable curtailment obligation plausible, especially in utility commission proceedings this fall. However, resolution requires a second identifiable multi-gigawatt campus, an explicit commitment of at least 250 MW, and disclosure through a utility filing or campus IR by a near-term deadline. Special contracts often redact customer identity or describe interruptibility without a firm MW amount, which would fail under the ambiguity rule. The 147-day window includes several filing cycles but is short for negotiating and approving a campus-specific agreement. Six weeks elapsed since the prior forecast also modestly reduces the remaining probability.", "keyEvidence": [], "whatWouldChangeMyMind": "A pending large-load tariff or special-contract docket naming a multi-gigawatt campus and proposing at least 250 MW of mandatory interruptible load would raise the estimate substantially." } ], "dissent": null, "spread": 0.13 }, "final": { "p": 0.4561350514532453, "method": "stacked-fixed-weights (resolved n=57 < 80)", "weights": { "calibrated": 0.3, "council": 0.5, "model": 0.2 }, "ci80": [ 0.33493999736522695, 0.5869644823222192 ] }, "direction": "down", "drivers": [ "Reference class qtype:disclosure runs 0.50 (n=29) vs stated 0.49.", "Hierarchical calibration moves stated 0.49 to 0.53 (lens power, kappa_stated=24.0, kappa_lens=200.0).", "Feature model at 0.53: conjuncts=1, named disclosure=True, horizon=190d.", "Council of 5 at 0.38." ], "warnings": [ "all 5 votes came from one vendor (openai); extremization disabled, persona diversity only" ], "evidence": [], "bubbleBasket": "energization", "daysRemaining": 147, "deadline": "2027-01-31", "lens": "power" } }, { "id": "lq-p79", "weeklyPredictionId": "p79-gateway-endpoint-fidelity-jan31", "text": "A major model-serving platform publishes per-endpoint accuracy, price, and latency by January 31, 2027.", "lens": "networking", "deadline": "2027-01-31", "resolutionSource": "Platform docs or status page.", "resolutionRule": "Hit requires all three metrics per endpoint. Two of three is partial.", "ambiguityPolicy": "partial_credit_if_one_leg_hits", "stated": 0.31, "current": 0.42886966859528997, "direction": "up", "history": [ { "asOf": "2026-08-08", "p": 0.31, "source": "stated", "runId": null }, { "asOf": "2026-09-06", "p": 0.42886966859528997, "source": "engine", "runId": "2026-09-06T22-42-59Z-51a6cc" } ], "engine": { "id": "lq-p79", "weeklyPredictionId": "p79-gateway-endpoint-fidelity-jan31", "stated": 0.31, "referenceClass": { "p": 0.3888888888888889, "className": "qtype:timing", "n": 8, "note": "most specific ledger class with n >= 8; Resolved predictions of type timing: disclosure = requires a named primary document or named company; threshold = numeric bar without a named-disclosure requirement; timing = event-by-date without a numeric bar; other = none of the above." }, "calibrated": { "p": 0.5019553156395261, "method": "hierarchical-beta-binomial-by-lens", "ci80": [ 0.46009209208749124, 0.5438113381045794 ], "variants": { "platt-on-stated-logit": 0.3349 } }, "model": { "p": 0.3801877187221175, "method": "logistic-features-l2", "ci80": [ 0.10850054582761932, 0.7236509838568924 ], "features": { "conjuncts": 2, "hasNumericThreshold": false, "horizonDays": 176, "isMarketPriceQuestion": true, "isTimingQuestion": true, "requiresNamedPrimaryDisclosure": false, "statedConfidence": 0.31, "thresholdMagnitudeLog": 0, "weekIndex": 15 }, "gbm": 0.3563 }, "council": { "p": 0.40561546427886525, "aggregation": "log-odds-mean", "extremization": 1, "votes": [ { "persona": "grid-and-power-economist", "model": "openai/gpt-5.6-sol", "p": 0.29, "rationale": "Price and latency are already natural endpoint-level routing attributes, but accuracy is harder: it requires a disclosed benchmark, reproducible methodology, and endpoint-specific results rather than model-level scores. The four-month window permits a dashboard or documentation update, yet the literal all-three requirement makes a full hit materially less likely than publication of only price and latency. The networking and timing reference classes center near 0.32–0.40, but the strict conjunction and likely accuracy bottleneck justify a discount. The forecaster’s 0.31 is therefore reasonable but slightly high.", "keyEvidence": [], "whatWouldChangeMyMind": "A dated platform roadmap, public beta, or documentation schema showing endpoint-level benchmark accuracy alongside price and measured latency would raise the estimate substantially." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.43, "rationale": "Price is commonly documented per model or provider endpoint, and latency is increasingly exposed in routing dashboards, making partial credit plausible within 147 days. The binding leg is accuracy: it is workload-dependent, difficult to standardize, and usually reported per model or benchmark rather than per serving endpoint. Requiring a platform-controlled docs or status-page publication further reduces the full-hit probability. Accounting for a meaningful chance of partial credit puts the expected outcome above the forecaster’s 0.31, but below the overall base rate.", "keyEvidence": [], "whatWouldChangeMyMind": "A named major platform announcing endpoint-level quality benchmarks alongside existing pricing and latency would raise p sharply; clarification that model-level accuracy qualifies would also raise it." }, { "persona": "capital-markets-credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.53, "rationale": "Price and observed latency are already natural endpoint-level routing inputs for major model-serving platforms, making at least partial resolution plausible within 147 days. Accuracy is the difficult leg: it requires repeatable evaluations tied to each provider endpoint, version control, and public documentation rather than model-level benchmarks. The strict per-endpoint wording therefore keeps a full hit materially below even odds. However, the explicit partial-credit policy raises expected resolution value above the forecaster’s 0.31 and above the networking base rate.", "keyEvidence": [ "https://openrouter.ai/docs" ], "whatWouldChangeMyMind": "A dated platform roadmap promising endpoint-level benchmark accuracy would raise this sharply; evidence that latency and pricing remain only model-level, aggregated, or selectively disclosed would lower it." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.41, "rationale": "I use the timing base rate of 0.389, cross-checked against networking at 0.318 and multi-leg questions at 0.433. The existential wording (“a major” platform) and partial credit help: endpoint pricing is common, and latency reporting is increasingly plausible. But a full hit requires the harder accuracy metric at endpoint granularity, published in qualifying platform documentation by a fixed date. That conjunctive disclosure requirement offsets the four-month window. The digest provides no platform-specific evidence. The forecaster’s lone sub-0.50 calibration observation is uninformative, while overconfidence appears mainly in higher bands.", "keyEvidence": [], "whatWouldChangeMyMind": "A qualifying platform launching a live endpoint comparison table, or dated documentation committing to accuracy, price, and latency reporting, would raise the estimate substantially." }, { "persona": "geopolitics-regulation-analyst", "model": "openai/gpt-5.6-sol", "p": 0.38, "rationale": "The 147-day window and competition among several major serving platforms create multiple chances for a launch, while price and latency reporting are already natural platform features. Accuracy is the limiting leg: it requires benchmark definitions, version control, and comparability, and platforms often publish it by model rather than by endpoint. The strict requirement that all metrics be per endpoint and appear in platform documentation or a status page materially lowers the full-hit probability. Partial-credit treatment offsets some of that downside. The networking, timing, and multi-leg reference classes center around 0.32–0.43, supporting an estimate near their midpoint.", "keyEvidence": [], "whatWouldChangeMyMind": "A platform previewing a unified endpoint leaderboard with benchmark accuracy, posted price, and measured latency would raise this sharply; documentation showing accuracy remains model-only would lower it." } ], "dissent": null, "spread": 0.24 }, "final": { "p": 0.42886966859528997, "method": "stacked-fixed-weights (resolved n=57 < 80)", "weights": { "calibrated": 0.3, "council": 0.5, "model": 0.2 }, "ci80": [ 0.28558720262160636, 0.5757400383128694 ] }, "direction": "up", "drivers": [ "Reference class qtype:timing runs 0.39 (n=8) vs stated 0.31.", "Hierarchical calibration moves stated 0.31 to 0.50 (lens networking, kappa_stated=24.0, kappa_lens=200.0).", "Feature model at 0.38: conjuncts=2, named disclosure=False, horizon=176d.", "Council of 5 at 0.41." ], "warnings": [ "all 5 votes came from one vendor (openai); extremization disabled, persona diversity only" ], "evidence": [ { "url": "https://openrouter.ai/docs", "title": "OpenRouter Quickstart Guide", "publishedAt": null, "source": "direct", "excerpt": "Cited by capital-markets-credit-analyst: Price and observed latency are already natural endpoint-level routing inputs for major model-serving platforms, making at least partial resolution plausible within 147 days. Accuracy is the difficult leg: it requires repeatable evaluations tied to each provid" } ], "bubbleBasket": null, "daysRemaining": 147, "deadline": "2027-01-31", "lens": "networking" } }, { "id": "lq-p81", "weeklyPredictionId": "p81-eval-vendor-isolation-policy-jan31", "text": "At least two frontier labs publish network isolation or containment requirements for eval vendors by January 31, 2027.", "lens": "capital", "deadline": "2027-01-31", "resolutionSource": "Lab policy pages.", "resolutionRule": "Hit on two labs. One lab is a miss.", "ambiguityPolicy": "resolves_against_forecaster", "stated": 0.44, "current": 0.49310638828319703, "direction": "up", "history": [ { "asOf": "2026-08-08", "p": 0.44, "source": "stated", "runId": null }, { "asOf": "2026-09-06", "p": 0.49310638828319703, "source": "engine", "runId": "2026-09-06T22-42-59Z-51a6cc" } ], "engine": { "id": "lq-p81", "weeklyPredictionId": "p81-eval-vendor-isolation-policy-jan31", "stated": 0.44, "referenceClass": { "p": 0.65, "className": "qtype:threshold", "n": 19, "note": "most specific ledger class with n >= 8; Resolved predictions of type threshold: disclosure = requires a named primary document or named company; threshold = numeric bar without a named-disclosure requirement; timing = event-by-date without a numeric bar; other = none of the above." }, "calibrated": { "p": 0.5343725270623367, "method": "hierarchical-beta-binomial-by-lens", "ci80": [ 0.4926379160900053, 0.5759810921852361 ], "variants": { "platt-on-stated-logit": 0.4087 } }, "model": { "p": 0.7850485121140612, "method": "logistic-features-l2", "ci80": [ 0.5654627679582795, 0.9110732690580562 ], "features": { "conjuncts": 1, "hasNumericThreshold": true, "horizonDays": 176, "isMarketPriceQuestion": false, "isTimingQuestion": true, "requiresNamedPrimaryDisclosure": false, "statedConfidence": 0.44, "thresholdMagnitudeLog": 0, "weekIndex": 15 }, "gbm": 0.7113 }, "council": { "p": 0.3416589628887384, "aggregation": "log-odds-mean", "extremization": 1, "votes": [ { "persona": "grid-and-power-economist", "model": "openai/gpt-5.6-sol", "p": 0.34, "rationale": "The requirement is unusually specific: public lab policy pages must state network isolation or containment controls applicable to evaluation vendors, and two labs must qualify. Frontier-lab frameworks increasingly mention external evaluations and model security, but detailed vendor controls are usually confined to contracts, security questionnaires, or nonpublic operating procedures. Updating one policy within 147 days is plausible; two qualifying publications require either coordinated standards pressure or a shared security incident. Ambiguity against the forecaster also makes generic sandboxing, model-access, or cybersecurity language insufficient. I am therefore below the forecaster’s 0.44.", "keyEvidence": [ "https://www.anthropic.com/responsible-scaling-policy" ], "whatWouldChangeMyMind": "Draft or final policy language from a second lab explicitly requiring eval vendors to use isolated networks, air gaps, sandboxes, or equivalent containment would raise the estimate sharply." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.34, "rationale": "The requirement is unusually specific: two separate labs must publicly state network-isolation or containment controls applicable to evaluation vendors. OpenAI, Anthropic, and Google DeepMind periodically revise safety frameworks, creating plausible publication opportunities. However, operational controls for outside evaluators are commonly kept in contracts, security addenda, or confidential onboarding material rather than public policy pages. The short window, two-lab threshold, and ambiguity-against-forecaster rule substantially reduce the chance of a hit. This is below the forecaster’s 0.44.", "keyEvidence": [ "https://www.anthropic.com/responsible-scaling-policy" ], "whatWouldChangeMyMind": "A first qualifying lab policy, or a shared standard/regulatory commitment requiring labs to publish external-evaluator containment controls, would raise the estimate materially." }, { "persona": "capital-markets-credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.34, "rationale": "Frontier labs increasingly publish safety frameworks, but these usually specify outcome-level security principles rather than concrete network-isolation or containment controls imposed on external evaluation vendors. The rule requires qualifying language on official policy pages from two distinct labs; private contract terms, evaluator-authored standards, general sandboxing language, or publication by only one lab fail. Although 147 days permits policy revisions, obtaining two unambiguous public disclosures remains a meaningful conjunctive hurdle, especially under adverse ambiguity resolution. The capital-market series offers little direct signal for this operational-policy event.", "keyEvidence": [ "https://www.anthropic.com/responsible-scaling-policy" ], "whatWouldChangeMyMind": "A qualifying official policy page from one lab, plus a dated commitment or draft from a second lab explicitly covering eval-vendor network isolation or containment." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.31, "rationale": "I use the >90-day base rate of 0.40; the higher capital and threshold rates are less relevant to this narrow policy-disclosure question. Detailed network-isolation terms are often contractual or operational rather than published on lab policy pages. Requiring two separate labs compounds adoption and disclosure-timing risk, and ambiguity about what qualifies as an eval vendor or containment requirement resolves against a hit. A shared safety standard could generate correlated publication, partly offsetting that penalty. The forecaster’s 0.44 confidence band has only one resolved case, so it provides no meaningful calibration adjustment; broader bins show overconfidence rather than underconfidence.", "keyEvidence": [], "whatWouldChangeMyMind": "A qualifying official policy from one lab plus a dated public commitment or draft from a second would raise the forecast substantially." }, { "persona": "geopolitics-regulation-analyst", "model": "openai/gpt-5.6-sol", "p": 0.38, "rationale": "The requirement is unusually specific: two distinct frontier labs must publicly state network-isolation or containment controls applicable to evaluation vendors, not merely endorse external evaluations or describe internal model-security practices. Such vendor controls are commonly contractual, audit-based, or confidential, reducing incentives to place operational details on public policy pages. Labs may update safety frameworks around year-end, and one publication is plausible as third-party evaluation expands, but obtaining two qualifying documents within 147 days is materially harder. Ambiguity resolving against the forecaster further lowers the chance that general secure-environment language counts.", "keyEvidence": [], "whatWouldChangeMyMind": "A first lab publishing explicit vendor network controls, or coordinated standards/commitments requiring labs to disclose secure third-party evaluation environments, would raise the estimate substantially." } ], "dissent": null, "spread": 0.07 }, "final": { "p": 0.49310638828319703, "method": "stacked-fixed-weights (resolved n=57 < 80)", "weights": { "calibrated": 0.3, "council": 0.5, "model": 0.2 }, "ci80": [ 0.4118743654851236, 0.5774897608125298 ] }, "direction": "up", "drivers": [ "Reference class qtype:threshold runs 0.65 (n=19) vs stated 0.44.", "Hierarchical calibration moves stated 0.44 to 0.53 (lens capital, kappa_stated=24.0, kappa_lens=200.0).", "Feature model at 0.79: conjuncts=1, named disclosure=False, horizon=176d.", "Council of 5 at 0.34." ], "warnings": [ "all 5 votes came from one vendor (openai); extremization disabled, persona diversity only", "2 cited URL(s) failed verification and were dropped." ], "evidence": [ { "url": "https://www.anthropic.com/responsible-scaling-policy", "title": "Anthropic’s Responsible Scaling Policy \\ Anthropic", "publishedAt": null, "source": "direct", "excerpt": "Cited by grid-and-power-economist: The requirement is unusually specific: public lab policy pages must state network isolation or containment controls applicable to evaluation vendors, and two labs must qualify. Frontier-lab frameworks increasingly mention external evaluations and model security, bu" } ], "bubbleBasket": null, "daysRemaining": 147, "deadline": "2027-01-31", "lens": "capital" } }, { "id": "lq-p87", "weeklyPredictionId": "p87-turn-count-in-model-card-jan31", "text": "A frontier lab publishes turn count or task-completion cost as a headline model-card metric by January 31, 2027.", "lens": "software", "deadline": "2027-01-31", "resolutionSource": "Model card.", "resolutionRule": "Hit if turn count or task-completion cost is a headline field, not a footnote.", "ambiguityPolicy": "resolves_against_forecaster", "stated": 0.26, "current": 0.4045057264463835, "direction": "up", "history": [ { "asOf": "2026-08-15", "p": 0.26, "source": "stated", "runId": null }, { "asOf": "2026-09-06", "p": 0.4045057264463835, "source": "engine", "runId": "2026-09-06T22-42-59Z-51a6cc" } ], "engine": { "id": "lq-p87", "weeklyPredictionId": "p87-turn-count-in-model-card-jan31", "stated": 0.26, "referenceClass": { "p": 0.3888888888888889, "className": "qtype:timing", "n": 8, "note": "most specific ledger class with n >= 8; Resolved predictions of type timing: disclosure = requires a named primary document or named company; threshold = numeric bar without a named-disclosure requirement; timing = event-by-date without a numeric bar; other = none of the above." }, "calibrated": { "p": 0.5010555076215127, "method": "hierarchical-beta-binomial-by-lens", "ci80": [ 0.4600593865120491, 0.542047901731291 ], "variants": { "platt-on-stated-logit": 0.3045 } }, "model": { "p": 0.5564636770573331, "method": "logistic-features-l2", "ci80": [ 0.298398526223336, 0.8152133212399633 ], "features": { "conjuncts": 1, "hasNumericThreshold": false, "horizonDays": 169, "isMarketPriceQuestion": true, "isTimingQuestion": true, "requiresNamedPrimaryDisclosure": false, "statedConfidence": 0.26, "thresholdMagnitudeLog": 0, "weekIndex": 16 }, "gbm": 0.5791 }, "council": { "p": 0.2959385390668332, "aggregation": "log-odds-mean", "extremization": 1, "votes": [ { "persona": "grid-and-power-economist", "model": "openai/gpt-5.6-sol", "p": 0.29, "rationale": "Multiple frontier-lab model releases are likely within 147 days given the digest’s frequent 2026 release cadence. However, model cards usually headline capability, safety, latency, or token-efficiency results—not turn count or end-to-end task-completion cost. Agentic evaluations create a plausible path because longer trajectories make cost and turns decision-relevant, but merely including either metric in a table, appendix, or footnote will not resolve positively. The undefined boundary of “frontier lab” and the ambiguity-against-forecaster rule further reduce the chance. I am slightly above the prior 0.26 because numerous likely release opportunities provide multiple shots at adoption.", "keyEvidence": [ "https://huggingface.co/models" ], "whatWouldChangeMyMind": "A pre-deadline model-card template or launch preview from a major lab explicitly placing average turns or dollars per completed task in its top-level benchmark summary." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.29, "rationale": "Agentic evaluations increasingly expose inefficiency: two models can have similar completion rates but materially different tool-call counts, turns, and inference costs. Multiple active frontier labs and likely year-end model releases create several opportunities. However, model cards still foreground capability and safety benchmarks; pricing, token use, and task cost usually appear in API documentation, benchmark methodology, or footnotes. “Task-completion cost” also requires standardized task and pricing assumptions, while turn count can be harness-dependent. The strict requirement that either metric be a headline model-card field, plus ambiguity resolving against the forecast, keeps the probability below the software and timing base rates.", "keyEvidence": [], "whatWouldChangeMyMind": "A frontier lab previewing a model card with a top-level agent-efficiency table covering average turns, tool calls, tokens, or dollar cost per completed task." }, { "persona": "capital-markets-credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.31, "rationale": "Several frontier labs are likely to release models before the deadline, and agentic evaluations increasingly track interaction length and end-to-end economics. However, model cards still emphasize capability, safety, latency, and benchmark accuracy; turn count is usually supporting evaluation detail, while task-completion cost is complicated by pricing, scaffolding, and token-budget assumptions. The rule further requires the metric to be a headline field in the model card itself, excluding blog emphasis or footnotes. These constraints keep the probability below the software and timing base rates, though modestly above the forecaster’s 0.26.", "keyEvidence": [], "whatWouldChangeMyMind": "A frontier lab previewing standardized agent evaluations with average turns or dollars per completed task as a top-level model-card table column would raise the estimate materially." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.29, "rationale": "I use the 0.40 base rate for >90-day questions, consistent with the 0.389 timing rate. Multiple frontier labs and two qualifying metric types provide several paths to a hit, while the digest indicates continued model-release activity. However, the rule requires the metric in a model card and as a headline field by a hard date; ordinary benchmark tables, launch posts, or footnotes fail. “Frontier lab” and “headline” also create resolution risk under the adverse ambiguity policy. The forecaster’s low-confidence calibration has only one observation, while broader results show overconfidence, so it offers no basis for raising this much.", "keyEvidence": [], "whatWouldChangeMyMind": "A pre-deadline model card draft, evaluation standard, or lab announcement explicitly placing turns or cost per completed task in its top-line metrics would raise this sharply." }, { "persona": "geopolitics-regulation-analyst", "model": "openai/gpt-5.6-sol", "p": 0.3, "rationale": "Multiple frontier labs are likely to release models within 147 days, and agentic evaluation increasingly makes turns and end-to-end task cost decision-relevant. However, model cards still foreground capability, safety, context length, and benchmark accuracy; turns usually appear in evaluation methodology, while pricing is published separately. The strict requirement that either measure be a headline field—not a table detail or footnote—substantially reduces the chance of a hit. The release-cadence digest raises opportunity count but provides no evidence that headline model-card conventions have shifted.", "keyEvidence": [], "whatWouldChangeMyMind": "A frontier lab previewing standardized agent evaluations with cost-per-completed-task or turns as a top-level reporting field would raise this materially." } ], "dissent": null, "spread": 0.02 }, "final": { "p": 0.4045057264463835, "method": "stacked-fixed-weights (resolved n=57 < 80)", "weights": { "calibrated": 0.3, "council": 0.5, "model": 0.2 }, "ci80": [ 0.3392338687594011, 0.4868447458893411 ] }, "direction": "up", "drivers": [ "Reference class qtype:timing runs 0.39 (n=8) vs stated 0.26.", "Hierarchical calibration moves stated 0.26 to 0.50 (lens software, kappa_stated=24.0, kappa_lens=200.0).", "Feature model at 0.56: conjuncts=1, named disclosure=False, horizon=169d.", "Council of 5 at 0.30." ], "warnings": [ "all 5 votes came from one vendor (openai); extremization disabled, persona diversity only" ], "evidence": [ { "url": "https://huggingface.co/models", "title": "Models – Hugging Face", "publishedAt": null, "source": "direct", "excerpt": "Cited by grid-and-power-economist: Multiple frontier-lab model releases are likely within 147 days given the digest’s frequent 2026 release cadence. However, model cards usually headline capability, safety, latency, or token-efficiency results—not turn count or end-to-end task-completion cost. Agent" } ], "bubbleBasket": null, "daysRemaining": 147, "deadline": "2027-01-31", "lens": "software" } }, { "id": "lq-p85", "weeklyPredictionId": "p85-nvidia-rubin-ultra-memory-confirm-mar31", "text": "NVIDIA publicly confirms a Rubin Ultra memory configuration at or below 512GB, or explicitly walks it back, by March 31, 2027.", "lens": "hardware", "deadline": "2027-03-31", "resolutionSource": "NVIDIA GTC, IR, or datasheet.", "resolutionRule": "Hit on confirmation ≤512GB or an explicit walk-back of a prior higher figure. Silence at deadline is a miss.", "ambiguityPolicy": "resolves_against_forecaster", "stated": 0.37, "current": 0.4039606567515722, "direction": "up", "history": [ { "asOf": "2026-08-15", "p": 0.37, "source": "stated", "runId": null }, { "asOf": "2026-09-06", "p": 0.4039606567515722, "source": "engine", "runId": "2026-09-06T22-42-59Z-51a6cc" } ], "engine": { "id": "lq-p85", "weeklyPredictionId": "p85-nvidia-rubin-ultra-memory-confirm-mar31", "stated": 0.37, "referenceClass": { "p": 0.5, "className": "qtype:disclosure", "n": 29, "note": "most specific ledger class with n >= 8; Resolved predictions of type disclosure: disclosure = requires a named primary document or named company; threshold = numeric bar without a named-disclosure requirement; timing = event-by-date without a numeric bar; other = none of the above." }, "calibrated": { "p": 0.5228709485578352, "method": "hierarchical-beta-binomial-by-lens", "ci80": [ 0.4811906032847247, 0.564467782327935 ], "variants": { "platt-on-stated-logit": 0.3695 } }, "model": { "p": 0.4509756993264821, "method": "logistic-features-l2", "ci80": [ 0.24160457422188777, 0.7914950580048584 ], "features": { "conjuncts": 1, "hasNumericThreshold": true, "horizonDays": 228, "isMarketPriceQuestion": false, "isTimingQuestion": true, "requiresNamedPrimaryDisclosure": true, "statedConfidence": 0.37, "thresholdMagnitudeLog": 2.709, "weekIndex": 16 }, "gbm": 0.2466 }, "council": { "p": 0.31990350599919387, "aggregation": "log-odds-mean", "extremization": 1, "votes": [ { "persona": "grid-and-power-economist", "model": "openai/gpt-5.6-sol", "p": 0.31, "rationale": "The deadline likely captures GTC 2027, creating a plausible venue for Rubin Ultra specifications, but detailed memory disclosure could still await a later launch datasheet. The ≤512GB outcome also faces hardware risk: Ultra-class designs may aggregate multiple GPU dies and HBM stacks into a package whose advertised memory exceeds 512GB even if a per-die figure does not. Under the ambiguity rule, unclear die-versus-package labeling would not qualify. An explicit walk-back is a weak alternative because NVIDIA commonly replaces roadmap specifications without characterizing the change as a retraction. The supplied public-series data shows strong NVIDIA demand but does not resolve the memory architecture or disclosure timing.", "keyEvidence": [ "https://www.nvidia.com/gtc/", "https://investor.nvidia.com/" ], "whatWouldChangeMyMind": "An NVIDIA roadmap or datasheet defining Rubin Ultra’s relevant unit and listing ≤512GB would raise this sharply; a confirmed package configuration above 512GB or GTC scheduled after the deadline would lower it." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.34, "rationale": "Rubin Ultra’s previously indicated higher capacity is architecturally plausible for a multi-die, many-stack HBM4E package, so a reduction to 512GB or less is not the base case. HBM4E 16-high yield, thermals, CoWoS-L integration, and package availability create meaningful downgrade risk, but NVIDIA more often preserves headline specifications and shifts timing or volume than explicitly walks back a roadmap figure. GTC 2027 provides a natural disclosure window, yet NVIDIA could discuss system performance without publishing a qualifying memory configuration. The strict source requirement and silence-as-miss rule lower the probability below the hardware-led reference rate.", "keyEvidence": [ "https://www.nvidia.com/gtc/", "https://investor.nvidia.com/" ], "whatWouldChangeMyMind": "Supplier evidence of 8-high rather than 16-high HBM4E allocation, a revised NVIDIA roadmap slide, or an official Rubin Ultra datasheet listing package memory would move the estimate materially." }, { "persona": "capital-markets-credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.32, "rationale": "GTC in March is a natural venue for updated Rubin Ultra specifications, so there is a meaningful chance NVIDIA addresses memory before the deadline. However, reaching ≤512GB would be a substantial reduction from the previously discussed higher configuration. NVIDIA is more likely to preserve the roadmap target, revise other specifications, or defer exact capacity than explicitly acknowledge a large downgrade. The strict source requirement and silence-as-miss rule further reduce the probability. Hardware forecasts resolve relatively well, but the longer-horizon disclosure reference class is weaker.", "keyEvidence": [ "https://www.nvidia.com/gtc/", "https://investor.nvidia.com/" ], "whatWouldChangeMyMind": "Credible HBM4E packaging constraints or an official pre-GTC NVIDIA specification showing 512GB or less would raise it; continued higher-capacity language in 2027 GTC materials would sharply lower it." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.31, "rationale": "Base rate: 0.40 for >90-day questions, cross-checked against 0.50 for disclosures and 0.615 for hardware. I lower it because resolution requires a specific ≤512GB Rubin Ultra statement or explicit walk-back in NVIDIA GTC, IR, or a datasheet by a hard date; leaks, partner materials, vague platform-memory figures, and silence all fail. A likely GTC window before the deadline helps, but the digest provides no Rubin-specific evidence. The forecaster’s low-confidence calibration has only one observation, so it warrants no adjustment; broader results show overconfidence rather than grounds to raise the estimate.", "keyEvidence": [ "https://www.nvidia.com/gtc/", "https://investor.nvidia.com/" ], "whatWouldChangeMyMind": "An NVIDIA agenda, presentation, IR release, or datasheet explicitly listing Rubin Ultra memory per accelerator/package at ≤512GB, or directly correcting a previously published higher figure." }, { "persona": "geopolitics-regulation-analyst", "model": "openai/gpt-5.6-sol", "p": 0.32, "rationale": "A March GTC creates a plausible disclosure window before the deadline, but Rubin Ultra is positioned as a high-memory 2027 product, making a configuration above 512GB more likely than one at or below it. NVIDIA also rarely issues explicit walk-backs; it more often replaces roadmap slides without characterizing the change. Even if capacity has been reduced, architecture-level GTC material may omit final memory specifications, while a definitive datasheet could arrive after March. The strict source requirement and silence-as-miss rule push this below the hardware and disclosure base rates.", "keyEvidence": [ "https://www.nvidia.com/gtc/", "https://investor.nvidia.com/" ], "whatWouldChangeMyMind": "An NVIDIA GTC agenda promising Rubin Ultra specifications, or supplier evidence that production HBM4E capacity limits the package to 512GB, would raise the estimate materially." } ], "dissent": null, "spread": 0.03 }, "final": { "p": 0.4039606567515722, "method": "stacked-fixed-weights (resolved n=57 < 80)", "weights": { "calibrated": 0.3, "council": 0.5, "model": 0.2 }, "ci80": [ 0.3426691449289524, 0.5032348125517568 ] }, "direction": "up", "drivers": [ "Reference class qtype:disclosure runs 0.50 (n=29) vs stated 0.37.", "Hierarchical calibration moves stated 0.37 to 0.52 (lens hardware, kappa_stated=24.0, kappa_lens=200.0).", "Feature model at 0.45: conjuncts=1, named disclosure=True, horizon=228d.", "Council of 5 at 0.32." ], "warnings": [ "all 5 votes came from one vendor (openai); extremization disabled, persona diversity only" ], "evidence": [ { "url": "https://www.nvidia.com/gtc/", "title": "Best of NVIDIA GTC: AI Breakthroughs and Recommended Sessions | NVIDIA GTC", "publishedAt": null, "source": "direct", "excerpt": "Cited by grid-and-power-economist: The deadline likely captures GTC 2027, creating a plausible venue for Rubin Ultra specifications, but detailed memory disclosure could still await a later launch datasheet. The ≤512GB outcome also faces hardware risk: Ultra-class designs may aggregate multiple GPU " }, { "url": "https://investor.nvidia.com/", "title": "NVIDIA Corporation - Home", "publishedAt": null, "source": "direct", "excerpt": "Cited by grid-and-power-economist: The deadline likely captures GTC 2027, creating a plausible venue for Rubin Ultra specifications, but detailed memory disclosure could still await a later launch datasheet. The ≤512GB outcome also faces hardware risk: Ultra-class designs may aggregate multiple GPU " } ], "bubbleBasket": null, "daysRemaining": 206, "deadline": "2027-03-31", "lens": "hardware" } }, { "id": "lq-p94", "weeklyPredictionId": "p94-second-vendor-cpo-shipping-mar31", "text": "A second major networking or accelerator vendor ships co-packaged optics commercially by March 31, 2027.", "lens": "networking", "deadline": "2027-03-31", "resolutionSource": "Vendor shipment disclosure.", "resolutionRule": "Hit on a second vendor besides the first already-counted shipper. Sampling/eval units miss.", "ambiguityPolicy": "resolves_against_forecaster", "stated": 0.47, "current": 0.39537691521594115, "direction": "down", "history": [ { "asOf": "2026-08-22", "p": 0.47, "source": "stated", "runId": null }, { "asOf": "2026-09-06", "p": 0.39537691521594115, "source": "engine", "runId": "2026-09-06T22-42-59Z-51a6cc" } ], "engine": { "id": "lq-p94", "weeklyPredictionId": "p94-second-vendor-cpo-shipping-mar31", "stated": 0.47, "referenceClass": { "p": 0.5, "className": "qtype:disclosure", "n": 29, "note": "most specific ledger class with n >= 8; Resolved predictions of type disclosure: disclosure = requires a named primary document or named company; threshold = numeric bar without a named-disclosure requirement; timing = event-by-date without a numeric bar; other = none of the above." }, "calibrated": { "p": 0.5183655720497826, "method": "hierarchical-beta-binomial-by-lens", "ci80": [ 0.4765002005263141, 0.560163306402882 ], "variants": { "platt-on-stated-logit": 0.4254 } }, "model": { "p": 0.26743189878690726, "method": "logistic-features-l2", "ci80": [ 0.08836317595380072, 0.5898820440976973 ], "features": { "conjuncts": 1, "hasNumericThreshold": false, "horizonDays": 221, "isMarketPriceQuestion": false, "isTimingQuestion": true, "requiresNamedPrimaryDisclosure": true, "statedConfidence": 0.47, "thresholdMagnitudeLog": 0, "weekIndex": 17 }, "gbm": 0.437 }, "council": { "p": 0.3797635051716962, "aggregation": "log-odds-mean", "extremization": 1, "votes": [ { "persona": "grid-and-power-economist", "model": "openai/gpt-5.6-sol", "p": 0.43, "rationale": "The seven-month window is plausible because at least one major accelerator/networking vendor has publicly targeted CPO-based platforms around the 2026 product cycle. However, the rule requires commercial shipment by a second vendor, not announcement, qualification, sampling, or customer evaluation. CPO integration, optical-engine yield, thermal reliability, and switch-system qualification commonly delay volume availability. A vendor may also begin limited deliveries without issuing sufficiently explicit shipment disclosure. Strong networking demand and supplier growth improve incentives but do not establish production readiness. Given the strict documentation standard and the forecaster’s weak networking reference-class rate, I place this below even odds.", "keyEvidence": [], "whatWouldChangeMyMind": "An explicit vendor statement that production units are shipping to paying customers, or a dated general-availability commitment backed by named systems and customers, would raise it materially; schedule slips or evaluation-only language would lower it." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.38, "rationale": "The seven-month window is tight for a second vendor under a shipment-only rule. CPO programs often spend multiple quarters between sampling, qualification, production announcements, and revenue shipments because optical attach, thermal reliability, yield, and system serviceability remain gating issues. Strong networking demand and rapid revenue growth at relevant suppliers support continued investment, but do not establish commercial shipment readiness. The requirement for an explicit vendor disclosure, with evaluation units excluded and ambiguity resolved against the forecast, materially lowers the probability versus general roadmap execution. The networking reference-class rate also favors below-even odds.", "keyEvidence": [], "whatWouldChangeMyMind": "A named vendor announcing completed customer qualification, volume production, or immediate orderability with shipments scheduled before March 31 would raise the estimate substantially." }, { "persona": "capital-markets-credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.41, "rationale": "CPO has repeatedly moved from demonstrations and qualification into delayed production ramps. The remaining 206 days are enough for a previously announced platform to reach revenue shipment, but not much time to resolve packaging, yield, qualification, and customer deployment issues. The rule is demanding: a second major vendor must explicitly disclose commercial shipment; availability dates, demonstrations, design wins, and sampling do not count. Strong networking demand and vendor investment support the technology, but the networking reference rate and ambiguity policy favor a below-even estimate.", "keyEvidence": [], "whatWouldChangeMyMind": "A major vendor explicitly reporting production-volume CPO shipments to paying customers, or confirming completed qualification and a firm shipment date before March 31, would raise the estimate materially." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.27, "rationale": "I use the networking-ledger base rate of 0.318. I adjust downward because resolution requires both genuine commercial shipment—not sampling/evaluation—and a qualifying vendor disclosure within 206 days. “Second” and “major” add qualification risk, while ambiguity resolves against the forecast. Strong networking and semiconductor revenue growth supports demand but is not specific evidence of production CPO shipments. The forecaster’s 0.4–0.5 calibration bin has only one observation, so it provides no credible upward correction; broader results show some overconfidence.", "keyEvidence": [], "whatWouldChangeMyMind": "A dated official announcement from another major vendor explicitly stating volume commercial shipments, with products available to paying customers rather than evaluation units." }, { "persona": "geopolitics-regulation-analyst", "model": "openai/gpt-5.6-sol", "p": 0.42, "rationale": "Several major vendors have public CPO roadmaps, and strong networking revenue growth supports commercialization incentives. However, only 206 days remain, and the rule requires a second vendor’s explicit commercial-shipment disclosure; product announcements, availability targets, customer trials, and evaluation units do not qualify. CPO ramps have substantial packaging, yield, thermal, fiber-attach, and system-qualification risk, making roadmap slippage common. The networking reference-class rate of 0.318 and ambiguity-against treatment weigh against a near-even forecast, although having multiple plausible vendors raises the estimate above that base rate.", "keyEvidence": [], "whatWouldChangeMyMind": "A major vendor announcing volume production, general availability with fulfilled orders, or revenue shipments before year-end would raise this sharply; roadmap delays or continued sampling language would lower it." } ], "dissent": null, "spread": 0.16 }, "final": { "p": 0.39537691521594115, "method": "stacked-fixed-weights (resolved n=57 < 80)", "weights": { "calibrated": 0.3, "council": 0.5, "model": 0.2 }, "ci80": [ 0.27046017222625623, 0.5010793570580104 ] }, "direction": "down", "drivers": [ "Reference class qtype:disclosure runs 0.50 (n=29) vs stated 0.47.", "Hierarchical calibration moves stated 0.47 to 0.52 (lens networking, kappa_stated=24.0, kappa_lens=200.0).", "Feature model at 0.27: conjuncts=1, named disclosure=True, horizon=221d.", "Council of 5 at 0.38." ], "warnings": [ "all 5 votes came from one vendor (openai); extremization disabled, persona diversity only" ], "evidence": [], "bubbleBasket": null, "daysRemaining": 206, "deadline": "2027-03-31", "lens": "networking" } }, { "id": "lq-p68", "weeklyPredictionId": "p68-helios-production-rack-q2-2027", "text": "At least one named customer reports receiving a production AMD Helios rack by June 30, 2027.", "lens": "hardware", "deadline": "2027-06-30", "resolutionSource": "Named-customer or AMD primary disclosure.", "resolutionRule": "Hit on a named customer receiving a production rack. 'Qualified' without receipt misses.", "ambiguityPolicy": "resolves_against_forecaster", "stated": 0.68, "current": 0.5547551496706284, "direction": "down", "history": [ { "asOf": "2026-07-25", "p": 0.68, "source": "stated", "runId": null }, { "asOf": "2026-09-06", "p": 0.5547551496706284, "source": "engine", "runId": "2026-09-06T22-42-59Z-51a6cc" } ], "engine": { "id": "lq-p68", "weeklyPredictionId": "p68-helios-production-rack-q2-2027", "stated": 0.68, "referenceClass": { "p": 0.5, "className": "qtype:disclosure", "n": 29, "note": "most specific ledger class with n >= 8; Resolved predictions of type disclosure: disclosure = requires a named primary document or named company; threshold = numeric bar without a named-disclosure requirement; timing = event-by-date without a numeric bar; other = none of the above." }, "calibrated": { "p": 0.5543963722866488, "method": "hierarchical-beta-binomial-by-lens", "ci80": [ 0.5128632831174572, 0.5957308380151835 ], "variants": { "platt-on-stated-logit": 0.5486 } }, "model": { "p": 0.5560812721162408, "method": "logistic-features-l2", "ci80": [ 0.3755998953571308, 0.8108844481788784 ], "features": { "conjuncts": 1, "hasNumericThreshold": true, "horizonDays": 340, "isMarketPriceQuestion": false, "isTimingQuestion": true, "requiresNamedPrimaryDisclosure": true, "statedConfidence": 0.68, "thresholdMagnitudeLog": 0, "weekIndex": 13 }, "gbm": 0.7293 }, "council": { "p": 0.5544397707772538, "aggregation": "log-odds-mean", "extremization": 1, "votes": [ { "persona": "grid-and-power-economist", "model": "openai/gpt-5.6-sol", "p": 0.53, "rationale": "The deadline allows roughly ten months, enough for a planned rack-scale platform to enter initial production but not enough to absorb a substantial silicon, networking, cooling, or integration delay. The strict test is harder than shipment alone: a customer must be named and receipt of a production rack must be disclosed; qualification, sampling, orders, or future deployment plans do not count. Hardware forecasts have performed better than the forecaster’s overall ledger, but long-horizon and disclosure questions have weaker base rates. AMD’s strong investment backdrop supports execution, while the forecaster’s 0.60–0.69 band has historically been overconfident. I therefore put this modestly above even odds but below 0.68.", "keyEvidence": [], "whatWouldChangeMyMind": "A dated AMD or customer statement confirming production Helios deliveries would raise this sharply; disclosed schedule slippage, qualification-only language, or production targeted after June 2027 would lower it." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.61, "rationale": "Helios is a rack-scale integration of next-generation accelerators, HBM, CPUs, networking, and liquid cooling, so production receipt can lag silicon sampling and launch claims by several quarters. The deadline leaves enough time for an initial hyperscaler or AI customer deployment, and AMD would likely publicize such a milestone. However, the rule requires explicit receipt of a production rack: customer commitments, qualification, availability, or AMD saying production began would not suffice. That disclosure requirement and normal HBM/packaging/system-validation slip risk keep this below the forecaster’s 0.68.", "keyEvidence": [ "https://ir.amd.com/" ], "whatWouldChangeMyMind": "An AMD or named-customer statement confirming production Helios shipments or installation would raise it sharply; a MI450/Helios ramp delay beyond early 2027 would lower it materially." }, { "persona": "capital-markets-credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.57, "rationale": "Helios was positioned as a 2026 rack-scale platform, leaving meaningful schedule buffer through June 2027, and AMD has incentives to publicize an initial deployment. However, the rule requires more than qualification, sampling, an order, or planned deployment: a primary source must identify a customer that actually received a production rack. New rack architectures commonly face GPU, networking, cooling, firmware, and integration delays, while customer announcements often avoid confirming physical receipt. The hardware reference class is supportive, but long-horizon disclosure outcomes and the forecaster’s historical overstatement in this confidence band argue below 0.68.", "keyEvidence": [], "whatWouldChangeMyMind": "A dated AMD or customer statement confirming production Helios shipments or installation would raise it sharply; roadmap slippage into late 2027, or disclosures limited to qualification and planned deployment, would lower it." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.47, "rationale": "Base rate: 0.50 from the disclosure reference class, bracketed by hardware at 0.615 and >90-day forecasts at 0.40. The 297-day window helps, but resolution requires both actual production-rack receipt and a timely primary disclosure naming the customer; qualification, sampling, or an unnamed shipment fails. The digest provides no customer-specific shipment evidence, while AMD’s capex growth and nearly flat RPO are weak proxies for rack delivery. The forecaster’s 0.60–0.69 calibration band realized 0.556, supporting a substantial haircut from 0.68. Ambiguity resolving against the forecast further lowers the estimate.", "keyEvidence": [ "https://ir.amd.com/", "https://www.sec.gov/edgar/browse/?CIK=2488&owner=exclude" ], "whatWouldChangeMyMind": "A primary AMD or customer statement that production Helios shipments have begun, especially identifying a customer deployment and receipt schedule before the deadline." }, { "persona": "geopolitics-regulation-analyst", "model": "openai/gpt-5.6-sol", "p": 0.59, "rationale": "The hardware reference class supports an above-even estimate, and 297 days provides meaningful time for rack integration, validation, and customer deployment. Strong AI-infrastructure indicators at Dell and Supermicro also suggest a capable rack supply chain. However, the resolution standard is stricter than product availability: a customer or AMD must publicly name the customer and confirm receipt of a production Helios rack. Qualification, sampling, orders, or general availability all miss. Rack-scale launches commonly slip because of networking, cooling, firmware, and system-validation dependencies, while customers may delay disclosure. These factors put the forecast below 0.68.", "keyEvidence": [ "https://ir.amd.com/" ], "whatWouldChangeMyMind": "A dated AMD/customer statement confirming production shipments or imminent installation would raise it substantially; evidence that Helios remains in qualification or has slipped beyond mid-2027 would lower it." } ], "dissent": null, "spread": 0.14 }, "final": { "p": 0.5547551496706284, "method": "stacked-fixed-weights (resolved n=57 < 80)", "weights": { "calibrated": 0.3, "council": 0.5, "model": 0.2 }, "ci80": [ 0.46349353220659467, 0.6527438275318147 ] }, "direction": "down", "drivers": [ "Reference class qtype:disclosure runs 0.50 (n=29) vs stated 0.68.", "Hierarchical calibration moves stated 0.68 to 0.55 (lens hardware, kappa_stated=24.0, kappa_lens=200.0).", "Feature model at 0.56: conjuncts=1, named disclosure=True, horizon=340d.", "Council of 5 at 0.55." ], "warnings": [ "all 5 votes came from one vendor (openai); extremization disabled, persona diversity only", "1 cited URL(s) failed verification and were dropped." ], "evidence": [ { "url": "https://ir.amd.com/", "title": "Advanced Micro Devices, Inc. (AMD)", "publishedAt": null, "source": "direct", "excerpt": "Cited by semiconductor-supply-chain-analyst: Helios is a rack-scale integration of next-generation accelerators, HBM, CPUs, networking, and liquid cooling, so production receipt can lag silicon sampling and launch claims by several quarters. The deadline leaves enough time for an initial hyperscaler" }, { "url": "https://www.sec.gov/edgar/browse/?CIK=2488&owner=exclude", "title": "EDGAR Entity Landing Page", "publishedAt": null, "source": "direct", "excerpt": "Cited by superforecaster-statistician: Base rate: 0.50 from the disclosure reference class, bracketed by hardware at 0.615 and >90-day forecasts at 0.40. The 297-day window helps, but resolution requires both actual production-rack receipt and a timely primary disclosure naming the customer; qualifi" } ], "bubbleBasket": null, "daysRemaining": 297, "deadline": "2027-06-30", "lens": "hardware" } }, { "id": "lq-p80", "weeklyPredictionId": "p80-amd-taalas-roadmap-jun30", "text": "AMD publicly names a Taalas-derived product or roadmap item tied to a specific manufacturing node by June 30, 2027.", "lens": "hardware", "deadline": "2027-06-30", "resolutionSource": "AMD IR or product page.", "resolutionRule": "Hit requires Taalas-derived naming and a node. Acquisition mentions without a product miss.", "ambiguityPolicy": "resolves_against_forecaster", "stated": 0.46, "current": 0.44253908554308935, "direction": "flat", "history": [ { "asOf": "2026-08-08", "p": 0.46, "source": "stated", "runId": null }, { "asOf": "2026-09-06", "p": 0.44253908554308935, "source": "engine", "runId": "2026-09-06T22-42-59Z-51a6cc" } ], "engine": { "id": "lq-p80", "weeklyPredictionId": "p80-amd-taalas-roadmap-jun30", "stated": 0.46, "referenceClass": { "p": 0.5, "className": "qtype:disclosure", "n": 29, "note": "most specific ledger class with n >= 8; Resolved predictions of type disclosure: disclosure = requires a named primary document or named company; threshold = numeric bar without a named-disclosure requirement; timing = event-by-date without a numeric bar; other = none of the above." }, "calibrated": { "p": 0.5320234909307167, "method": "hierarchical-beta-binomial-by-lens", "ci80": [ 0.4903685976248335, 0.5735614530176301 ], "variants": { "platt-on-stated-logit": 0.4198 } }, "model": { "p": 0.5571908715103946, "method": "logistic-features-l2", "ci80": [ 0.3286811515492792, 0.8293656477611151 ], "features": { "conjuncts": 1, "hasNumericThreshold": false, "horizonDays": 326, "isMarketPriceQuestion": false, "isTimingQuestion": true, "requiresNamedPrimaryDisclosure": true, "statedConfidence": 0.46, "thresholdMagnitudeLog": 0, "weekIndex": 15 }, "gbm": 0.6558 }, "council": { "p": 0.3473752844020192, "aggregation": "log-odds-mean", "extremization": 1, "votes": [ { "persona": "grid-and-power-economist", "model": "openai/gpt-5.6-sol", "p": 0.38, "rationale": "The 297-day window includes likely AMD product-event opportunities, and manufacturing nodes are commonly disclosed once hardware reaches launch-level detail. However, the rule requires AMD itself to connect three elements: Taalas derivation, a named product or roadmap item, and a specific node. Acquisition updates or generic technology-integration language do not qualify. Acquired architectures often take longer than ten months to enter a public roadmap, and AMD may absorb the technology without preserving Taalas attribution. These strict documentation risks put me below the forecaster’s 0.46.", "keyEvidence": [ "https://ir.amd.com/" ], "whatWouldChangeMyMind": "An AMD event agenda, filing, or roadmap preview explicitly linking Taalas technology to a named accelerator would raise this sharply; continued integration-only language through early 2027 would lower it." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.34, "rationale": "The deadline allows only about ten months, short for AMD to convert acquired accelerator IP into a named roadmap item with a disclosed process node. A roadmap announcement is more plausible than volume productization, especially if Taalas already had working silicon. However, the resolution requires AMD itself to identify the item as Taalas-derived and specify its node; AMD could instead discuss generic integration, custom silicon, or an Instinct roadmap without publicly attributing the IP. The strict source and ambiguity rules therefore put this below the hardware and disclosure base rates.", "keyEvidence": [ "https://ir.amd.com/" ], "whatWouldChangeMyMind": "An AMD analyst-day agenda, earnings commentary, or product teaser explicitly preserving the Taalas brand, describing inherited silicon, or promising a node-specific inference roadmap before the deadline." }, { "persona": "capital-markets-credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.34, "rationale": "The deadline allows roughly one major AMD roadmap-event cycle. A Taalas design already near commercialization could surface, and AMD commonly identifies manufacturing nodes for sufficiently mature products. The larger obstacle is the literal attribution requirement: acquired technology is often folded into broader Instinct or custom-silicon roadmaps without naming its origin. AMD must publish both recognizable Taalas derivation and a specific node on an IR or product page; a general acquisition update, architecture discussion, or node-free product announcement misses. AMD’s elevated capex supports development activity but says little about whether this unusually specific provenance disclosure occurs by the deadline.", "keyEvidence": [ "https://ir.amd.com/" ], "whatWouldChangeMyMind": "An AMD event agenda, executive statement, or product-page placeholder explicitly linking a named accelerator to Taalas technology would raise it; integration-only language or delayed silicon milestones would lower it." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.3, "rationale": "I use 0.45 as the base rate, blending the ledger’s >90-day 0.40 and disclosure 0.50 rates; hardware’s 0.615 is secondary because this rule is unusually strict. A hit needs, by a hard date, an AMD IR or product-page item that both identifies Taalas derivation and states a manufacturing node. Ordinary acquisition or integration commentary explicitly fails, while product integration plus public roadmap attribution within 297 days creates substantial timing risk. AMD’s capex growth is not specific evidence of such naming. Calibration offers no reliable support for 0.46: its 0.4–0.5 bin has only one case, while broader mid-confidence bins show overstatement.", "keyEvidence": [], "whatWouldChangeMyMind": "An official AMD event agenda or executive statement promising a near-term Taalas-derived product roadmap, especially with foundry or process-node details, would raise the estimate." }, { "persona": "geopolitics-regulation-analyst", "model": "openai/gpt-5.6-sol", "p": 0.38, "rationale": "AMD has a natural disclosure window around its mid-2027 product and AI roadmap events, and manufacturing nodes are commonly specified for announced accelerators. However, the strict requirement is harder than a normal product disclosure: AMD must both identify a product or roadmap item as Taalas-derived and state its node on an official page. Acquired technology is often absorbed under AMD branding without lineage attribution, while integration and silicon-roadmap cycles commonly exceed ten months. A generic acquisition update, team integration, or unnamed technology reference would miss.", "keyEvidence": [ "https://ir.amd.com/" ], "whatWouldChangeMyMind": "An AMD statement preserving a named Taalas chip program, confirming tape-out, or placing Taalas technology on a pre-July 2027 roadmap with a foundry node." } ], "dissent": null, "spread": 0.08 }, "final": { "p": 0.44253908554308935, "method": "stacked-fixed-weights (resolved n=57 < 80)", "weights": { "calibrated": 0.3, "council": 0.5, "model": 0.2 }, "ci80": [ 0.35938388524558995, 0.5400074302380328 ] }, "direction": "flat", "drivers": [ "Reference class qtype:disclosure runs 0.50 (n=29) vs stated 0.46.", "Hierarchical calibration moves stated 0.46 to 0.53 (lens hardware, kappa_stated=24.0, kappa_lens=200.0).", "Feature model at 0.56: conjuncts=1, named disclosure=True, horizon=326d.", "Council of 5 at 0.35." ], "warnings": [ "all 5 votes came from one vendor (openai); extremization disabled, persona diversity only", "1 cited URL(s) failed verification and were dropped." ], "evidence": [ { "url": "https://ir.amd.com/", "title": "Advanced Micro Devices, Inc. (AMD)", "publishedAt": null, "source": "direct", "excerpt": "Cited by grid-and-power-economist: The 297-day window includes likely AMD product-event opportunities, and manufacturing nodes are commonly disclosed once hardware reaches launch-level detail. However, the rule requires AMD itself to connect three elements: Taalas derivation, a named product or road" } ], "bubbleBasket": null, "daysRemaining": 297, "deadline": "2027-06-30", "lens": "hardware" } }, { "id": "lq-p101", "weeklyPredictionId": "p101-fervo-expansion-firm-jun30", "text": "Google accepts at least 500 MW of Fervo's conditional expansion and the parties disclose it by June 30, 2027.", "lens": "power", "deadline": "2027-06-30", "resolutionSource": "Google or Fervo primary disclosure.", "resolutionRule": "Hit on ≥500 MW accepted expansion. Conditional MOUs without acceptance miss.", "ambiguityPolicy": "resolves_against_forecaster", "bubbleBasket": "energization", "stated": 0.41, "current": 0.39467929056347145, "direction": "flat", "history": [ { "asOf": "2026-09-05", "p": 0.41, "source": "stated", "runId": null }, { "asOf": "2026-09-06", "p": 0.39467929056347145, "source": "engine", "runId": "2026-09-06T22-42-59Z-51a6cc" } ], "engine": { "id": "lq-p101", "weeklyPredictionId": "p101-fervo-expansion-firm-jun30", "stated": 0.41, "referenceClass": { "p": 0.5, "className": "qtype:disclosure", "n": 29, "note": "most specific ledger class with n >= 8; Resolved predictions of type disclosure: disclosure = requires a named primary document or named company; threshold = numeric bar without a named-disclosure requirement; timing = event-by-date without a numeric bar; other = none of the above." }, "calibrated": { "p": 0.5256909818405664, "method": "hierarchical-beta-binomial-by-lens", "ci80": [ 0.48329234698825396, 0.5679924958159563 ], "variants": { "platt-on-stated-logit": 0.392 } }, "model": { "p": 0.3542435391026188, "method": "logistic-features-l2", "ci80": [ 0.11269921238373272, 0.838831126231056 ], "features": { "conjuncts": 2, "hasNumericThreshold": true, "horizonDays": 298, "isMarketPriceQuestion": false, "isTimingQuestion": true, "requiresNamedPrimaryDisclosure": true, "statedConfidence": 0.41, "thresholdMagnitudeLog": 2.699, "weekIndex": 19 }, "gbm": 0.2466 }, "council": { "p": 0.33695188982488233, "aggregation": "log-odds-mean", "extremization": 1, "votes": [ { "persona": "grid-and-power-economist", "model": "openai/gpt-5.6-sol", "p": 0.34, "rationale": "Google and Fervo have an established commercial relationship, so counterparty fit is stronger than for a cold-start offtake. But the rule requires acceptance—not another conditional MOU—and at least 500 MW, a very large increment. In 297 days, drilling results, financing, transmission deliverability, and contract terms must align enough for a primary disclosure. Queue COD statistics are not directly dispositive because energization is unnecessary, but long development timelines and low completion rates still reduce willingness to make a firm commitment. I therefore price this below the forecaster’s 0.41.", "keyEvidence": [], "whatWouldChangeMyMind": "A signed PPA, regulatory filing, or primary announcement identifying Google and at least 500 MW would sharply increase confidence; material drilling, financing, or transmission delays would lower it." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.35, "rationale": "The explicit conditional expansion creates a credible path, and Google’s rising data-center power demand makes additional geothermal contracting plausible. But accepting at least 500 MW would be a major commitment relative to Fervo’s demonstrated operating scale, with drilling, financing, transmission, and interconnection risks. The 297-day window is short for converting conditional capacity into an accepted expansion. The strict requirement for acceptance—not another MOU or option—plus primary disclosure makes this a multi-leg event. I therefore place it below the forecaster’s 0.41.", "keyEvidence": [ "https://blog.google/" ], "whatWouldChangeMyMind": "A primary update showing completed drilling milestones, committed financing, interconnection progress, or Google negotiations specifically covering at least 500 MW would raise my estimate." }, { "persona": "capital-markets-credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.36, "rationale": "The 297-day window is meaningful, and Google has strategic reasons to secure firm clean power. But ≥500 MW is a large commitment to an emerging geothermal platform, and the rule requires actual acceptance—not another conditional MOU—plus primary disclosure. Such an exercise likely depends on development, financing, drilling, transmission, and pricing milestones. Interconnection data show long completion timelines, while elevated rates can complicate project financing. Contracting can precede construction, but milestone slippage and the ambiguity-against-forecaster rule put this below the relevant multi-leg and long-horizon base rates.", "keyEvidence": [ "https://fervoenergy.com/", "https://blog.google/" ], "whatWouldChangeMyMind": "A disclosed definitive option exercise, PPA, board-approved commitment, or financing package explicitly covering at least 500 MW would raise this sharply; delayed drilling or financing milestones would lower it." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.27, "rationale": "I use the 0.433 multi-leg base rate, cross-checked against 0.400 for >90-day questions and 0.500 for disclosures. I adjust downward because this requires both a large ≥500 MW acceptance and qualifying primary disclosure within 297 days; an MOU or broader partnership announcement explicitly misses. The digest provides no company-specific evidence that acceptance is imminent. Low historical queue completion adds execution caution, though acceptance can precede construction. The forecaster’s lone sub-0.50 resolution is uninformative, while broader calibration shows overconfidence, so 0.41 receives no upward deference.", "keyEvidence": [], "whatWouldChangeMyMind": "A Google or Fervo primary disclosure showing an executed, accepted expansion of at least 500 MW, or concrete evidence that final approval and announcement are imminent." }, { "persona": "geopolitics-regulation-analyst", "model": "openai/gpt-5.6-sol", "p": 0.37, "rationale": "Google’s established relationship with Fervo and rising demand for firm clean power make an expansion plausible. However, accepting at least 500 MW is a large single commitment, and only 297 days remain for technical, commercial, and possibly utility or regulatory milestones. Queued-generation projects have long development timelines and high withdrawal rates, encouraging staged rather than immediate commitments. The strict resolution rule further requires an accepted expansion—not an option or conditional MOU—and a primary-party disclosure by the deadline. Those conjunctive requirements put this modestly below the multi-leg and long-horizon reference rates.", "keyEvidence": [ "https://fervoenergy.com/", "https://sustainability.google/" ], "whatWouldChangeMyMind": "A Google or Fervo disclosure of advanced definitive negotiations, milestone satisfaction, utility approval, or a binding option exercise covering at least 500 MW would raise the estimate substantially." } ], "dissent": null, "spread": 0.1 }, "final": { "p": 0.39467929056347145, "method": "stacked-fixed-weights (resolved n=57 < 80)", "weights": { "calibrated": 0.3, "council": 0.5, "model": 0.2 }, "ci80": [ 0.2829124895852321, 0.5364110572617554 ] }, "direction": "flat", "drivers": [ "Reference class qtype:disclosure runs 0.50 (n=29) vs stated 0.41.", "Hierarchical calibration moves stated 0.41 to 0.53 (lens power, kappa_stated=24.0, kappa_lens=200.0).", "Feature model at 0.35: conjuncts=2, named disclosure=True, horizon=298d.", "Council of 5 at 0.34." ], "warnings": [ "all 5 votes came from one vendor (openai); extremization disabled, persona diversity only", "3 cited URL(s) failed verification and were dropped." ], "evidence": [ { "url": "https://blog.google/", "title": "News from Google | Google Product and Technology News and Stories", "publishedAt": null, "source": "direct", "excerpt": "Cited by semiconductor-supply-chain-analyst: The explicit conditional expansion creates a credible path, and Google’s rising data-center power demand makes additional geothermal contracting plausible. But accepting at least 500 MW would be a major commitment relative to Fervo’s demonstrated operatin" }, { "url": "https://fervoenergy.com/", "title": "Fervo Energy - Next-Generation Geothermal Projects", "publishedAt": null, "source": "direct", "excerpt": "Cited by capital-markets-credit-analyst: The 297-day window is meaningful, and Google has strategic reasons to secure firm clean power. But ≥500 MW is a large commitment to an emerging geothermal platform, and the rule requires actual acceptance—not another conditional MOU—plus primary disclosure. S" }, { "url": "https://sustainability.google/", "title": "Google Sustainability", "publishedAt": null, "source": "direct", "excerpt": "Cited by geopolitics-regulation-analyst: Google’s established relationship with Fervo and rising demand for firm clean power make an expansion plausible. However, accepting at least 500 MW is a large single commitment, and only 297 days remain for technical, commercial, and possibly utility or regul" } ], "bubbleBasket": "energization", "daysRemaining": 297, "deadline": "2027-06-30", "lens": "power" } } ], "theses": [ { "id": "th-2026-01", "title": "Enterprise on-prem AI at scale stays undisclosed", "statement": "By 2027-06-30, fewer than three Fortune 500 companies outside the technology, semiconductor and cloud sectors will have disclosed, in an 8-K, annual report, or company-issued press release, an owned on-premises AI compute deployment of 50 MW or more.", "horizon": "2027-06-30", "p": 0.9181390938717987, "ci80": [ 0.885785807112705, 0.9419285414032552 ], "direction": "warning", "milestones": [ "lq-p2", "lq-p19" ], "namedEntities": [ "Fortune 500" ], "recusalCheck": { "passed": true, "keysChecked": [ "digital_realty", "dlr_ticker", "dxm", "named_employer_site_decision", "named_competitor_booking", "customer_pipeline", "mnpi" ] }, "rationale": "The ledger's disclosure-type questions (a named primary document by a hard date) resolve at 50% (n=29) against a book average of 53% (n=57); the software lens runs at 45% (n=20). Enterprises consume frontier compute through hyperscalers and neoclouds and have weak incentives to disclose owned megawatts. The thesis is falsified by three qualifying disclosures, not by trend pieces or vendor case studies.", "whatWouldChangeMyMind": [ "A sovereign or regulated-industry buyer publishing a capacity figure in a filing.", "A vendor 8-K naming a non-tech Fortune 500 customer with an MW-scale on-prem order." ], "councilVotes": [ { "persona": "grid-and-power-economist", "model": "openai/gpt-5.6-sol", "p": 0.92, "rationale": "A 50 MW owned on-premises AI installation is campus-scale infrastructure, well beyond ordinary enterprise clusters. Most non-technology companies can obtain frontier compute through cloud or hosted capacity without owning the power and cooling plant. The resolution also requires three companies and qualifying primary documents, not vendor case studies or inferred GPU power. With under ten months remaining and no qualifying disclosures identified in the supplied digest, lead times and weak incentives to publish megawatts strongly favor fewer than three." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.88, "rationale": "Only about ten months remain. A 50 MW owned AI deployment is campus-scale, while non-tech enterprises generally procure frontier compute from cloud providers; even operating projects often disclose GPUs or investment, not power and ownership in the specified primary documents. The supplied digest identifies hyperscaler capex acceleration but no qualifying non-tech issuer. Requiring three separate Fortune 500 disclosures is unlikely, though undiscovered existing deployments and rapid “AI factory” announcements create tail risk." }, { "persona": "capital-markets-credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.94, "rationale": "Fewer than ten months remain, and 50 MW is an exceptionally large enterprise-owned AI deployment. Nontechnology Fortune 500 companies generally procure frontier compute through hyperscalers, neoclouds, or colocation rather than disclose owned on-premises capacity in megawatts. The required primary-document language further narrows qualification: vendor case studies, plans, partnerships, and aggregate data-center spending do not count. Three qualifying disclosures by the deadline would require an unusually abrupt, publicly specified buildout wave." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.91, "rationale": "Base rate: 0.50 for the ledger’s named-document disclosure questions. I adjust upward because FALSE requires at least three separate non-tech Fortune 500 issuers to make unusually specific primary disclosures within ten months: owned, on-premises AI compute of at least 50 MW. The digest shows massive hyperscaler capex but no qualifying enterprise disclosure, supporting continued cloud concentration. I stop at 0.91 because the ledger’s highest-confidence forecasts have been materially overconfident." }, { "persona": "geopolitics-regulation-analyst", "model": "openai/gpt-5.6-sol", "p": 0.93, "rationale": "A 50 MW owned deployment is utility-scale infrastructure, requiring substantial capital, power procurement and long lead times. Non-technology enterprises generally obtain frontier AI capacity through cloud or colocation arrangements, while corporate disclosures usually report spending rather than facility megawatts. The strict requirement for three Fortune 500 companies, qualifying ownership, explicit scale and a primary company document within ten months makes FALSE unlikely even if enterprise AI adoption accelerates." } ], "publicSeries": [ { "id": "huggingface-hub-models", "latest": null, "asOf": "2026-09-06" }, { "id": "sec-xbrl-companyfacts", "latest": null, "asOf": "2026-09-06" } ], "dissent": null, "enginePrior": 0.7, "method": "council-log-odds-mean" }, { "id": "th-2026-02", "title": "HBM stays sold out into 2027", "statement": "By 2027-03-31, at least two of SK hynix, Samsung Electronics, and Micron state in an earnings release, earnings call, or company press release that their 2027 HBM supply is fully allocated, sold out, or substantially committed under long-term agreements.", "horizon": "2027-03-31", "p": 0.6448853577801388, "ci80": [ 0.5613473208546763, 0.7204373192934582 ], "direction": "call", "milestones": [ "lq-p73", "lq-p9", "lq-p98" ], "namedEntities": [ "SK hynix", "Samsung Electronics", "Micron", "NVIDIA" ], "recusalCheck": { "passed": true, "keysChecked": [ "digital_realty", "dlr_ticker", "dxm", "named_employer_site_decision", "named_competitor_booking", "customer_pipeline", "mnpi" ] }, "rationale": "Hardware-lens questions in the ledger resolve at 62% (n=12), the book's strongest lens, and the memory-allocation questions in particular have hit or partially hit. Memory makers have used 'sold out' language for the forward year once qualification is complete; two of three doing so for 2027 requires the supply-demand gap to persist through the Vera Rubin ramp. Falsified if fewer than two make the statement by the horizon.", "whatWouldChangeMyMind": [ "A memory maker guiding to HBM price declines or capacity additions outrunning orders.", "NVIDIA or a hyperscaler cutting 2027 accelerator unit guidance." ], "councilVotes": [ { "persona": "grid-and-power-economist", "model": "openai/gpt-5.6-sol", "p": 0.62, "rationale": "SK hynix and Micron are the likeliest pair: both have historically discussed forward HBM allocation, while Micron’s sharply higher reported RPO supports stronger contractual visibility. Continued AI-accelerator growth should keep advanced HBM constrained. However, the rule requires explicit 2027 allocation or long-term-commitment language by March, not merely strong demand. Capacity additions, qualification delays, customer dual-sourcing, and Samsung’s weaker likelihood of using “sold out” language keep this below two-thirds." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.68, "rationale": "AI-accelerator demand remains strong: Micron’s RPO expansion and downstream backlog growth support continued HBM tightness into 2027. SK hynix and Micron are the likeliest qualifying pair; both have historically pre-allocated forward HBM output, while Samsung is a weaker third path. Allowing “substantially committed” and long-term agreements helps. Risks are rapid HBM capacity expansion, qualification shifts, and guarded wording that leaves only one issuer making an explicit 2027 statement by March 31." }, { "persona": "capital-markets-credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.66, "rationale": "HBM capacity is typically reserved well ahead of delivery, and the digest’s sharp Micron RPO growth supports continued forward contracting. SK hynix and Micron are the likeliest qualifying pair, with several reporting opportunities remaining before March 31. However, expanded industry capacity, Samsung’s supply recovery, or cautious wording such as “strong visibility” could prevent literal qualification. Requiring two companies to publicly reference 2027—not merely current shortages—keeps the probability below 0.7." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.58, "rationale": "Base rate: 0.615 for hardware questions, tempered by the 0.400 rate for >90-day horizons. Micron’s sharply higher RPO and 31% revenue growth support sustained AI-memory demand. However, resolution requires two separate issuers to use qualifying 2027 commitment language by a hard date; tight supply alone is insufficient. Earnings calls broaden the disclosure path, but Samsung uncertainty and capacity expansion matter. The ledger’s 0.5–0.7 forecasts are overconfident, warranting a discount." }, { "persona": "geopolitics-regulation-analyst", "model": "openai/gpt-5.6-sol", "p": 0.68, "rationale": "SK hynix and Micron are the likeliest qualifying pair: both have precedent for describing forward HBM output as sold out or allocated, while Micron’s sharply higher RPO supports durable commitments. AI-system growth and the Rubin ramp favor continued tightness. The main risk is literal resolution: suppliers may discuss strong demand, negotiations, or 2026 allocation without explicitly characterizing 2027 supply as fully allocated or substantially committed by March 31." } ], "publicSeries": [ { "id": "sec-xbrl-companyfacts", "latest": null, "asOf": "2026-09-06" } ], "dissent": null, "enginePrior": 0.65, "method": "council-log-odds-mean" }, { "id": "th-2026-03", "title": "A listed neocloud has a credit event", "statement": "By 2027-09-30, at least one publicly listed GPU-cloud operator (CoreWeave, Nebius, or another listed neocloud) discloses in an SEC filing a covenant waiver or amendment, a going-concern qualification, a missed debt service payment, or the loss of a customer representing 10% or more of remaining performance obligations.", "horizon": "2027-09-30", "p": 0.4901314840765464, "ci80": [ 0.40384268124669986, 0.57701237970461 ], "direction": "warning", "milestones": [ "lq-p5", "lq-p84", "lq-p75" ], "namedEntities": [ "CoreWeave", "Nebius" ], "recusalCheck": { "passed": true, "keysChecked": [ "digital_realty", "dlr_ticker", "dxm", "named_employer_site_decision", "named_competitor_booking", "customer_pipeline", "mnpi" ] }, "rationale": "Capital-lens questions resolve at 71% (n=11). High-yield OAS is 2.65% as of 2026-09-03, so credit markets are not signalling stress today; the thesis is a duration-mismatch call, not a spread call: GPU useful lives, contract tenors, and amortising debt are misaligned and the first disclosure of a tenor gap is already in the ledger. Twelve months is enough for one renewal cycle to miss. Falsified if no listed neocloud files any of the four triggers.", "whatWouldChangeMyMind": [ "Hyperscaler contract extensions that lengthen tenors past debt maturities.", "Equity raises that retire GPU-backed debt at par.", "High-yield OAS compressing further alongside neocloud refinancing at lower coupons." ], "councilVotes": [ { "persona": "grid-and-power-economist", "model": "openai/gpt-5.6-sol", "p": 0.49, "rationale": "The trigger is broader than default: a routine covenant amendment or waiver qualifies, raising the odds across multiple issuers. CoreWeave’s $24.9B long-term debt and heavy quarterly capex create meaningful refinancing and execution exposure. Conversely, rapidly growing RPO, strong revenue growth, tight broad high-yield spreads, and benign equity volatility imply near-term funding remains available. The small listed-neocloud universe and requirement for explicit SEC disclosure keep this near even odds." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.46, "rationale": "CoreWeave’s $24.9B long-term debt and extraordinary capex create meaningful refinancing and covenant risk over 13 months, while customer concentration makes contract disruption consequential. However, $103.7B of RPO, rapid growth, and a 2.65% broad high-yield OAS argue against imminent distress. The literal SEC-disclosure requirement is restrictive: routine refinancing or weaker demand does not qualify unless a specified waiver, amendment, payment miss, going-concern language, or qualifying customer loss is filed." }, { "persona": "capital-markets-credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.45, "rationale": "CoreWeave’s $24.9B long-term debt and $7.7B quarterly capex create meaningful refinancing and covenant-management risk, while its $103.7B RPO provides substantial demand visibility. Broad high-yield OAS at 2.65% and low volatility argue against imminent systemic stress. The likeliest literal trigger is a disclosed covenant amendment; missed payment or going-concern language remains unlikely. Across several listed operators, the one-year union probability is material but below even odds." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.46, "rationale": "Base rate: 0.40 for >90-day ledger questions; the 0.708 capital-lens rate is small-sample, and calibration shows overprediction above 0.5. CoreWeave’s $24.9B long-term debt and heavy capex make an amendment plausible, while multiple issuers and qualifying triggers increase the OR-event probability. Conversely, rapidly growing RPO, calm broad high-yield spreads, and the requirement for explicit SEC disclosure by a hard date materially reduce it; customer loss must also be quantified against RPO." }, { "persona": "geopolitics-regulation-analyst", "model": "openai/gpt-5.6-sol", "p": 0.59, "rationale": "The easiest qualifying path is a disclosed covenant amendment, which can accompany refinancing or capacity-funded debt without missed payments. CoreWeave’s $24.9B long-term debt and continuing capital intensity create meaningful amendment risk over four reporting cycles. Conversely, $103.7B of RPO, rapid revenue growth, tight broad high-yield spreads, and low volatility argue against acute distress. Going-concern, missed-payment, and qualifying customer-loss disclosures remain individually unlikely; the broad issuer set and multiple alternative triggers lift the combined probability above even." } ], "publicSeries": [ { "id": "fred-BAMLH0A0HYM2", "latest": 2.65, "asOf": "2026-09-03" }, { "id": "fred-BAMLH0A3HYC", "latest": 10.51, "asOf": "2026-09-03" }, { "id": "sec-xbrl-companyfacts", "latest": null, "asOf": "2026-09-06" } ], "dissent": null, "enginePrior": 0.4, "method": "council-log-odds-mean" }, { "id": "th-2026-04", "title": "Power-first hyperscale campuses keep coming", "statement": "Between 2026-09-06 and 2027-06-30, hyperscalers (Microsoft, Amazon, Alphabet, Meta, Oracle) announce, each in a primary company disclosure, at least three distinct new AI campuses or generation blocks of 500 MW or more that are explicitly behind-the-meter, co-located with dedicated generation, or described as power-first.", "horizon": "2027-06-30", "p": 0.4350006232955849, "ci80": [ 0.34339221228002653, 0.5312739372451487 ], "direction": "call", "milestones": [ "lq-p36", "lq-p76" ], "namedEntities": [ "Microsoft", "Amazon", "Alphabet", "Meta", "Oracle" ], "recusalCheck": { "passed": true, "keysChecked": [ "digital_realty", "dlr_ticker", "dxm", "named_employer_site_decision", "named_competitor_booking", "customer_pipeline", "mnpi" ] }, "rationale": "Power-lens questions are the book's most reliable: 70% (n=4). Behind-the-meter announcements have repeated because grid interconnection is the binding constraint: 13.9% of 2015-2020 interconnection requests had reached commercial operation by end-2025 (LBNL Queued Up, n=11622). Three announcements in ten months is a modest bar against the recent cadence; the risk is in the 500 MW and primary-disclosure qualifiers, not in the trend.", "whatWouldChangeMyMind": [ "A federal or state rule that makes behind-the-meter generation uneconomic for large loads.", "Hyperscaler capex guidance cuts of 20% or more for 2027." ], "councilVotes": [ { "persona": "grid-and-power-economist", "model": "openai/gpt-5.6-sol", "p": 0.38, "rationale": "The physical trend is strong: long interconnection timelines increase demand for dedicated generation and power-first siting. But resolution requires three separate announcements in under ten months, each from a hyperscaler itself, each at least 500 MW, and each using qualifying power language. Many otherwise relevant projects are announced by utilities or developers, omit capacity, or involve grid-supplied generation. Those documentation and wording filters make three qualifying disclosures materially less likely than three underlying projects." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.42, "rationale": "Power scarcity strongly favors dedicated-generation designs, but resolution requires three qualifying announcements in under ten months, each from a named hyperscaler’s primary disclosure and explicitly at least 500 MW. Historically, capacity figures and behind-the-meter details often come from utilities, developers, regulators, or local officials rather than the hyperscaler. Projects may also be framed as expansions, procurement agreements, or future studies, failing the “distinct new campus or generation block” test. The power trend is strong; the documentation conjunction is restrictive." }, { "persona": "capital-markets-credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.48, "rationale": "Grid delays and strong generator-equipment backlogs support continued power-first development, but resolution requires three unusually specific events within ten months: each must exceed 500 MW, concern a distinct AI campus or generation block, explicitly use qualifying power language, and appear in a hyperscaler’s own disclosure. Many genuine projects will instead be announced by utilities, developers, or governments, or omit capacity and behind-the-meter details. The announcement threshold is therefore materially harder than the underlying construction trend." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.36, "rationale": "I use the 0.40 base rate for >90-day questions (n=14), not the 0.70 power-lens rate because that sample has only four cases. Grid delays make dedicated generation plausible, but this is conjunctive: three separate ≥500 MW projects, AI linkage, qualifying power language, and hyperscaler primary disclosures before a hard date. Utility or vendor announcements do not suffice. The forecaster’s 0.5–0.7 bins show overconfidence, supporting a below-base estimate." }, { "persona": "geopolitics-regulation-analyst", "model": "openai/gpt-5.6-sol", "p": 0.54, "rationale": "Grid delays strongly favor dedicated-generation and power-first designs, while supplier backlogs and benign credit conditions support continued hyperscale investment. The event concerns announcements, not completed interconnections, so low queue completion rates reinforce the thesis without imposing construction timing. However, three qualifying disclosures in under ten months is demanding: hyperscaler releases often omit an explicit 500 MW figure, use grid-connected language, or leave generation details to utilities and developers. The power-lens precedent is favorable but only n=4." } ], "publicSeries": [ { "id": "lbnl-queued-up", "latest": null, "asOf": "2026-09-06" }, { "id": "sec-xbrl-companyfacts", "latest": null, "asOf": "2026-09-06" } ], "dissent": null, "enginePrior": 0.7, "method": "council-log-odds-mean" }, { "id": "th-2026-05", "title": "State-level data-center policy friction spreads", "statement": "By 2027-06-30, at least three additional US states (beyond those counted at 2026-09-06) adopt, through enacted legislation or a public utility commission order, a statewide data-center moratorium, a discretionary-permit pause, or a large-load tariff or grid-interconnection rule.", "horizon": "2027-06-30", "p": 0.6301490130340821, "ci80": [ 0.5379812107543747, 0.7137146498609033 ], "direction": "call", "milestones": [ "lq-p65", "lq-p92" ], "namedEntities": [ "FERC", "PJM", "ERCOT" ], "recusalCheck": { "passed": true, "keysChecked": [ "digital_realty", "dlr_ticker", "dxm", "named_employer_site_decision", "named_competitor_booking", "customer_pipeline", "mnpi" ] }, "rationale": "State action clusters after visible ratepayer impact, and most 2027 legislative sessions convene in January, inside the horizon. The ledger's two live policy questions are single-leg threshold questions, a type that resolves at 65% (n=19). Three states in ten months is bold but measurable; county pauses do not count.", "whatWouldChangeMyMind": [ "Retail rate relief mechanisms that remove the ratepayer argument.", "Federal preemption of state large-load rules." ], "councilVotes": [ { "persona": "grid-and-power-economist", "model": "openai/gpt-5.6-sol", "p": 0.54, "rationale": "The qualifying category is broad: a statewide large-load tariff or interconnection rule can clear through a PUC order without legislation, and 2027 sessions provide another route. Visible cost-allocation disputes make clustered action plausible. Against that, three new states in under ten months is a demanding threshold; legislation can miss June adjournment, PUC schedules slip, and county restrictions do not count. Unclear baseline and qualifying scope also resolve against TRUE." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.72, "rationale": "The broad resolution language makes PUC action the likeliest route: a statewide large-load tariff or interconnection rule is easier to adopt than a moratorium. Accelerating load requests, long interconnection timelines, and ratepayer-cost concerns create pressure across multiple states. Most 2027 legislatures also have time to act before June 30. The main risks are that actions remain utility-specific or proposed rather than ordered, or that the baseline already includes the easiest states." }, { "persona": "capital-markets-credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.66, "rationale": "The criterion is broad: any qualifying statewide tariff or interconnection rule counts, making commission action more likely than three moratoria or legislative pauses. Long interconnection timelines and accelerating utility commitments create ratepayer and reliability pressure, while most 2027 legislatures meet before the deadline. Still, three net-new states in under ten months is demanding: proposals, utility filings, local restrictions, and orders issued after June 30 do not qualify." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.59, "rationale": "Base rate: 0.65 for threshold questions (n=19). I adjust downward because success requires three separate states within ten months, each through enacted legislation or a qualifying statewide commission order; local actions and pending proposals fail. Broad eligibility and 2027 legislative sessions partly offset that penalty. The forecaster’s 0.6–0.7 calibration bin realized only 0.556, supporting shrinkage toward 0.59." }, { "persona": "geopolitics-regulation-analyst", "model": "openai/gpt-5.6-sol", "p": 0.63, "rationale": "The broadest qualifying path is a PUC-approved large-load tariff or interconnection rule, which can advance outside legislative calendars and is increasingly prompted by grid-cost and ratepayer concerns. Most states also get a January-to-June 2027 legislative window, allowing clustered imitation after prominent disputes. However, three distinct new states is a meaningful threshold: bills often die, contested dockets slip, and local moratoria or utility proposals without final commission orders do not qualify." } ], "publicSeries": [ { "id": "lbnl-queued-up", "latest": null, "asOf": "2026-09-06" } ], "dissent": null, "enginePrior": 0.55, "method": "council-log-odds-mean" }, { "id": "th-2026-06", "title": "Vendor-dated milestones slip more than the book expects", "statement": "Of the ledger's live vendor-timing questions lq-p83, lq-p77, lq-p95, lq-p86, and lq-p33, at most two resolve as a hit by 2027-03-31 under their written resolution rules.", "horizon": "2027-03-31", "p": 0.6100535344396106, "ci80": [ 0.5243649998577111, 0.689447429298674 ], "direction": "contrarian", "milestones": [ "lq-p83", "lq-p77", "lq-p95", "lq-p86", "lq-p33" ], "namedEntities": [ "DeepSeek", "Artificial Analysis", "SemiAnalysis", "Open Compute Project", "NVIDIA" ], "recusalCheck": { "passed": true, "keysChecked": [ "digital_realty", "dlr_ticker", "dxm", "named_employer_site_decision", "named_competitor_booking", "customer_pipeline", "mnpi" ] }, "rationale": "Timing-type questions in the ledger resolve at 39% (n=8) and multi-leg questions at 43% (n=14) versus 57% for single-leg (n=43). The five questions carry stated probabilities averaging above 0.6, which implies roughly three hits; the thesis says the book is still too generous to vendor dates. Falsified by three or more hits.", "whatWouldChangeMyMind": [ "Two of the five resolving early as hits before 2026-11-30." ], "councilVotes": [ { "persona": "grid-and-power-economist", "model": "openai/gpt-5.6-sol", "p": 0.61, "rationale": "The strongest applicable base rates are 0.39 for timing questions and 0.40 for horizons over 90 days, though both samples are small. The book’s above-0.6 average appears optimistic: its 0.6–0.7 bin realized 0.556, while exact-document and multi-leg rules add failure modes. Constituent hit probabilities around 0.44–0.48 imply roughly 0.55–0.65 for at most two hits; vendor-common correlation tempers confidence." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.62, "rationale": "The most relevant ledger base rates—0.39 for timing questions and 0.43 for multi-leg questions—suggest per-question hit odds around 0.42–0.46. Five independent trials at those odds put the probability of at most two hits near 0.59–0.65. Shared qualification, packaging, and foundry dependencies create correlated slippage, modestly favoring the low-hit outcome. Small samples and six months remaining limit confidence." }, { "persona": "capital-markets-credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.61, "rationale": "The timing reference class’s 39% hit rate implies roughly 0.70 for at most two hits if applied independently, but n=8 is weak and these five were selected with materially higher ex-ante odds. Shrinking toward broader calibration gives individual hit odds near 0.45–0.50. Literal document, counterparty, and deadline requirements favor misses, while strong hyperscaler capex, RPO growth, and benign broad credit conditions support execution. Net, the threshold is more likely than not but not dominant." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.59, "rationale": "Base rate: timing questions hit 0.39 (n=8); treating five as independent gives 0.70 for at most two hits. I shrink that small sample toward the broader long-horizon evidence and allow correlated vendor execution, reducing the thesis probability. Calibration still favors skepticism: 0.6–0.7 forecasts realized only 0.556, while hard-date disclosure requirements add failure modes. Strong open-weight release cadence and hyperscaler spending argue against going much higher." }, { "persona": "geopolitics-regulation-analyst", "model": "openai/gpt-5.6-sol", "p": 0.62, "rationale": "The strongest applicable reference classes favor misses: timing questions hit 39% (n=8), multi-leg questions 43% (n=14), and >90-day questions 40% (n=14). Treating five questions as roughly 0.43–0.47 each gives an independent-case probability near 0.56–0.63 for no more than two hits. Strict document and deadline requirements further favor misses, though small samples and correlated vendor execution limit confidence." } ], "publicSeries": [ { "id": "huggingface-hub-models", "latest": null, "asOf": "2026-09-06" } ], "dissent": null, "enginePrior": 0.5, "method": "council-log-odds-mean" }, { "id": "th-2026-07", "title": "Interconnection conversion stays below one in five", "statement": "The next LBNL Queued Up edition (data through 2026), if published by 2027-09-30, reports that no more than 20% of the generation capacity requested in 2018-2020 had reached commercial operation by end-2026; if no edition is published by then the thesis resolves against the engine.", "horizon": "2027-09-30", "p": 0.8998112932049759, "ci80": [ 0.8635542464494671, 0.927245634110988 ], "direction": "call", "milestones": [ "lq-p102", "lq-p76" ], "namedEntities": [ "Lawrence Berkeley National Laboratory", "PJM", "MISO", "ERCOT" ], "recusalCheck": { "passed": true, "keysChecked": [ "digital_realty", "dlr_ticker", "dxm", "named_employer_site_decision", "named_competitor_booking", "customer_pipeline", "mnpi" ] }, "rationale": "13.9% of 2015-2020 interconnection requests had reached commercial operation by end-2025 (LBNL Queued Up, n=11622). The 2018-2020 vintages sit at 2018: 8.2% of MW, 2019: 8.0% of MW, 2020: 4.5% of MW today, and median request-to-COD time has been lengthening. Twelve more months adds single-digit points of completion for those vintages, so 20% is a comfortable ceiling.", "whatWouldChangeMyMind": [ "FERC Order 2023 cluster studies clearing a large backlog of executed agreements into service.", "A methodology change in the LBNL edition that redefines the vintage denominator." ], "councilVotes": [ { "persona": "grid-and-power-economist", "model": "openai/gpt-5.6-sol", "p": 0.89, "rationale": "The reported MW completion shares for the 2018-2020 vintages—8.2%, 8.0%, and 4.5% through 2025—leave substantial room below the 20% ceiling. Lengthening request-to-COD times and low cumulative completion incidence make a one-year surge above 20% very unlikely. The principal failure risk is procedural: no qualifying data-through-2026 edition appears by September 30, 2027, which automatically resolves false." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.92, "rationale": "The 2018-2020 vintages currently show only 4.5%-8.2% of requested MW operating. Reaching a weighted 20% by end-2026 would require an implausibly large one-year completion jump amid a 5.44-year median request-to-COD lag. The principal risk is procedural: Berkeley Lab may delay the data-through-2026 edition beyond September 2027 or revise the series methodology. Its established annual cadence makes timely publication likely." }, { "persona": "capital-markets-credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.91, "rationale": "Conditional on publication, the threshold is highly likely: end-2025 MW completion rates for the 2018, 2019, and 2020 vintages were only 8.2%, 8.0%, and 4.5%. Lengthening request-to-COD times and low late-vintage cumulative completion make a one-year jump above 20% unlikely. The principal failure risk is that LBNL does not publish the data-through-2026 edition by the hard deadline or materially revises its vintage methodology." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.86, "rationale": "Base rate: 0.65 for threshold questions; the >90-day class is only 0.40. Conditional on publication, TRUE is very likely: 2018–2020 vintages are only 4.5%–8.2% operational through 2025, while cohort COD incidence rises slowly and median request-to-COD reached 5.44 years. Crossing 20% in one year would require an unprecedented jump. I assign 0.96 to the metric condition and 0.90 to timely publication, then apply a small calibration penalty." }, { "persona": "geopolitics-regulation-analyst", "model": "openai/gpt-5.6-sol", "p": 0.91, "rationale": "The threshold leg is highly likely: 2018-2020 vintages were only 4.5%-8.2% operational by end-2025, while cumulative COD incidence reaches just 10% after seven years and completion times are lengthening. A one-year jump above 20% would be exceptional. The principal risk is procedural: the thesis automatically fails if LBNL does not publish a data-through-2026 edition by September 30, 2027." } ], "publicSeries": [ { "id": "lbnl-queued-up", "latest": null, "asOf": "2026-09-06" } ], "dissent": null, "enginePrior": 0.8, "method": "council-log-odds-mean" }, { "id": "th-2026-08", "title": "Second-source 1.6T and co-packaged optics ship", "statement": "By 2027-06-30, at least two vendors other than Broadcom (from Marvell, Credo, NVIDIA, Cisco, Ciena, Coherent, Lumentum) state in an earnings release or earnings call that 1.6T or co-packaged-optics products are shipping in production volume to at least one hyperscaler.", "horizon": "2027-06-30", "p": 0.4589350480148439, "ci80": [ 0.362401249794151, 0.5586526925230041 ], "direction": "call", "milestones": [ "lq-p45", "lq-p86", "lq-p103", "lq-p94" ], "namedEntities": [ "Broadcom", "Marvell", "Credo", "NVIDIA", "Cisco", "Ciena", "Coherent", "Lumentum" ], "recusalCheck": { "passed": true, "keysChecked": [ "digital_realty", "dlr_ticker", "dxm", "named_employer_site_decision", "named_competitor_booking", "customer_pipeline", "mnpi" ] }, "rationale": "The networking lens has a mixed record and the resolved optics questions have landed as partials or misses as often as hits: design wins are announced, production revenue lags. 'Production volume to a hyperscaler' in an earnings document is the strict version of the milestone questions. Two named vendors by mid-2027 aligns with the Vera Rubin ramp but is exposed to CPO reliability slips.", "whatWouldChangeMyMind": [ "A hyperscaler publicly delaying CPO adoption to a later switch generation.", "Broadcom guidance implying it holds essentially all 1.6T volume through 2027." ], "councilVotes": [ { "persona": "grid-and-power-economist", "model": "openai/gpt-5.6-sol", "p": 0.46, "rationale": "Seven eligible vendors create multiple paths, and strong revenue growth at Credo, Marvell, and Ciena supports an AI-networking ramp. But the rule requires two separate earnings-document confirmations—not sampling, qualification, design wins, or initial shipments. Vendors often avoid identifying customer class and describe ramps without saying “production volume”; 1.6T modules and CPO also face packaging and reliability risks. With under ten months and weak networking-question base rates, one qualifying disclosure looks likelier than two." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.55, "rationale": "The strongest paths are Marvell, Credo, Coherent, and Lumentum as 1.6T DSP, AEC, or optical-module ramps reach hyperscalers; rapid Credo and Ciena growth supports underlying demand. However, the rule requires two vendors to use production-volume and hyperscaler language specifically in earnings materials by June 2027. Qualification delays, guarded customer disclosure, and announcements describing sampling or initial revenue rather than volume shipments make this only modestly more likely than not." }, { "persona": "capital-markets-credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.47, "rationale": "The underlying 1.6T ramp is plausible within ten months, with Marvell, Credo, NVIDIA, Coherent and Lumentum offering several paths to qualification. However, resolution requires two vendors to use unusually specific earnings-language: shipping in production volume and serving a hyperscaler. Vendors often discuss design wins, sampling, qualification or initial revenue without confirming production volume or customer class. CPO reliability and deployment timing add risk, while the strong growth at Credo and Marvell modestly supports commercialization." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.34, "rationale": "Base rate: 0.318 for resolved networking-lens questions; the >90-day class is 0.400. I move modestly above the networking base rate because seven eligible vendors create multiple paths and revenue growth at Credo, Ciena, and Marvell supports strong networking demand. I then discount for the two-vendor conjunction and strict evidence test: each must explicitly state, in an earnings release or call, both production-volume shipment and a hyperscaler recipient. The ledger also shows midrange overconfidence." }, { "persona": "geopolitics-regulation-analyst", "model": "openai/gpt-5.6-sol", "p": 0.48, "rationale": "Ten months provides several earnings cycles, and strong growth at Credo, Marvell, and Ciena supports an accelerating networking transition. Multiple plausible suppliers could reach 1.6T volume during the Rubin-era ramp. However, resolution requires two vendors to make unusually explicit earnings-document disclosures covering production volume and a hyperscaler; design wins, sampling, qualification, or generic customer shipments do not suffice. Correlated qualification or CPO reliability delays could leave only one qualifying statement by June." } ], "publicSeries": [ { "id": "sec-xbrl-companyfacts", "latest": null, "asOf": "2026-09-06" } ], "dissent": null, "enginePrior": 0.55, "method": "council-log-odds-mean" } ], "engine": { "runId": "2026-09-06T22-42-59Z-51a6cc", "asOf": "2026-09-06", "engineVersion": "0.1.0", "dataSources": [ { "id": "fred-csv", "kind": "direct", "asOf": "2026-09-06", "records": 70420, "url": "https://fred.stlouisfed.org/graph/fredgraph.csv?id=", "status": "ok", "note": null }, { "id": "sec-xbrl-companyfacts", "kind": "direct", "asOf": "2026-09-06", "records": 1964, "url": "https://data.sec.gov/api/xbrl/companyfacts/CIK##########.json", "status": "partial", "note": "missing: TSM (no us-gaap quarterly frames (likely a 20-F filer))" }, { "id": "huggingface-hub-models", "kind": "direct", "asOf": "2026-09-06", "records": 4362, "url": "https://huggingface.co/api/models?author=&sort=createdAt&direction=-1&limit=1000&expand[]=createdAt&expand[]=downloads", "status": "ok", "note": null }, { "id": "lbnl-queued-up", "kind": "direct", "asOf": "2026-09-06", "records": 38201, "url": "https://eta-publications.lbl.gov/sites/default/files/2026-05/lbnl_ix_queue_data_file_thru2025.xlsx", "status": "ok", "note": "data through 2025" }, { "id": "magellan-mcp", "kind": "magellan", "asOf": "2026-09-06", "records": 0, "url": null, "status": "unavailable", "note": "MAGELLAN_API_KEY not set; no request made" } ], "shadow": { "note": "Backtest on resolved predictions, time-ordered folds by issue date; training uses only resolutions knowable before each fold's first issue date. Untrainable folds fall back to the stated probability and count as losses for the gate. 'walkForward' re-cuts the same rows by the date each resolution became known (diagnostic only).", "folds": 5, "resolvedN": 57, "scores": { "baseRate": { "brier": 0.1996, "brierPartialAsMiss": 0.3151, "logLoss": 0.7284, "n": 57 }, "calibrated": { "brier": 0.1996, "brierPartialAsMiss": 0.3151, "logLoss": 0.7284, "n": 57 }, "coinFlip": { "brier": 0.1842, "brierPartialAsMiss": 0.25, "logLoss": 0.6931, "n": 57 }, "gbm": { "brier": 0.1996, "brierPartialAsMiss": 0.3151, "logLoss": 0.7284, "n": 57 }, "model": { "brier": 0.1996, "brierPartialAsMiss": 0.3151, "logLoss": 0.7284, "n": 57 }, "platt": { "brier": 0.1996, "brierPartialAsMiss": 0.3151, "logLoss": 0.7284, "n": 57 }, "stated": { "brier": 0.1996, "brierPartialAsMiss": 0.3151, "logLoss": 0.7284, "n": 57 } }, "promotionGate": { "passed": false, "rule": "calibrated or model must beat stated Brier on >= 4 of 5 folds and n >= 80", "foldsWon": 0, "calibratedFoldsWon": 0, "minResolvedN": 80, "modelFoldsWon": 0, "resolvedN": 57, "trainableFolds": 0, "walkForwardFoldsWon": { "calibrated": 1, "model": 0 } }, "foldWins": { "baseRate": 0, "calibrated": 0, "coinFlip": 3, "gbm": 0, "model": 0, "platt": 0 }, "knownAtImputedIds": [ "p14-anthropic-round-close", "p17-samsung-walkout-shipments", "p21-ebo-msa-spec", "p23-anthropic-round-close-final", "p3-capex-revise", "p35-cpo-design-win", "p4-interconnect-outpaces", "p49-micron-hbm4-booked", "p52-agent-automation-governance", "p53-skhy-debut-validates-memory", "p55-terra-confirms-repricing-cycle", "p56-samsung-hbm4-to-nvidia", "p59-tsmc-q2-capex-raise", "p62-deepseek-v4-ga-jul31", "p88-nvidia-10q-guaranty-exhibits-oct31", "p90-glm-53-weights-sep15" ], "leaveOneOutBrier": { "calibrated": 0.1949, "gbm": 0.1867, "logistic": 0.1818, "stated": 0.1996 }, "perFold": [ { "cutoff": "2026-04-25", "fold": 0, "scores": { "baseRate": { "brier": 0.2345, "brierPartialAsMiss": 0.3778, "logLoss": 0.8384, "n": 15 }, "calibrated": { "brier": 0.2345, "brierPartialAsMiss": 0.3778, "logLoss": 0.8384, "n": 15 }, "coinFlip": { "brier": 0.1667, "brierPartialAsMiss": 0.25, "logLoss": 0.6931, "n": 15 }, "gbm": { "brier": 0.2345, "brierPartialAsMiss": 0.3778, "logLoss": 0.8384, "n": 15 }, "model": { "brier": 0.2345, "brierPartialAsMiss": 0.3778, "logLoss": 0.8384, "n": 15 }, "platt": { "brier": 0.2345, "brierPartialAsMiss": 0.3778, "logLoss": 0.8384, "n": 15 }, "stated": { "brier": 0.2345, "brierPartialAsMiss": 0.3778, "logLoss": 0.8384, "n": 15 } }, "testIds": [ "p1-2gw-customer-funded", "p3-capex-revise", "p4-interconnect-outpaces", "p6-custom-silicon-35", "p10-fabric-second-confirm", "p7-bedrock-gpt55-anchor", "p8-capex-revenue-ratio", "p12-voice-frontier-procurement", "p13-mrc-second-fabric", "p14-anthropic-round-close", "p16-nvidia-circular-flag", "p17-samsung-walkout-shipments", "p18-claude-code-churn", "p20-cisco-ai-orders-tracking", "p21-ebo-msa-spec" ], "testN": 15, "trainN": 0, "trainable": false }, { "cutoff": "2026-05-23", "fold": 1, "scores": { "baseRate": { "brier": 0.2518, "brierPartialAsMiss": 0.3967, "logLoss": 0.8592, "n": 10 }, "calibrated": { "brier": 0.2518, "brierPartialAsMiss": 0.3967, "logLoss": 0.8592, "n": 10 }, "coinFlip": { "brier": 0.175, "brierPartialAsMiss": 0.25, "logLoss": 0.6931, "n": 10 }, "gbm": { "brier": 0.2518, "brierPartialAsMiss": 0.3967, "logLoss": 0.8592, "n": 10 }, "model": { "brier": 0.2518, "brierPartialAsMiss": 0.3967, "logLoss": 0.8592, "n": 10 }, "platt": { "brier": 0.2518, "brierPartialAsMiss": 0.3967, "logLoss": 0.8592, "n": 10 }, "stated": { "brier": 0.2518, "brierPartialAsMiss": 0.3967, "logLoss": 0.8592, "n": 10 } }, "testIds": [ "p22-nvidia-q2-fy27", "p23-anthropic-round-close-final", "p24-agent-platform-lock-in", "p25-fabric-vendor-q2", "p26-gemini-3-5-pro", "p27-anthropic-s1-public", "p28-gemini-3-5-pro-june", "p29-vera-rubin-cadence", "p30-optics-design-wins", "p31-sovereign-power-followthrough" ], "testN": 10, "trainN": 0, "trainable": false }, { "cutoff": "2026-06-06", "fold": 2, "scores": { "baseRate": { "brier": 0.2017, "brierPartialAsMiss": 0.2979, "logLoss": 0.722, "n": 13 }, "calibrated": { "brier": 0.2017, "brierPartialAsMiss": 0.2979, "logLoss": 0.722, "n": 13 }, "coinFlip": { "brier": 0.1923, "brierPartialAsMiss": 0.25, "logLoss": 0.6931, "n": 13 }, "gbm": { "brier": 0.2017, "brierPartialAsMiss": 0.2979, "logLoss": 0.722, "n": 13 }, "model": { "brier": 0.2017, "brierPartialAsMiss": 0.2979, "logLoss": 0.722, "n": 13 }, "platt": { "brier": 0.2017, "brierPartialAsMiss": 0.2979, "logLoss": 0.722, "n": 13 }, "stated": { "brier": 0.2017, "brierPartialAsMiss": 0.2979, "logLoss": 0.722, "n": 13 } }, "testIds": [ "p32-gemini-3-5-pro-ga", "p34-hbm4-allocation-tightness", "p35-cpo-design-win", "p37-gemini-3-5-pro-below-fable5", "p38-fable5-access-restored", "p39-second-inference-asic-milestone", "p40-second-1-6t-design-win", "p41-no-public-s1", "p43-open-weight-top5", "p44-hbm4-allocation-2027", "p46-ferc-rto-compliance", "p47-hyperscaler-1gw-btm", "p48-closed-price-response" ], "testN": 13, "trainN": 0, "trainable": false }, { "cutoff": "2026-06-27", "fold": 3, "scores": { "baseRate": { "brier": 0.1428, "brierPartialAsMiss": 0.2218, "logLoss": 0.5724, "n": 10 }, "calibrated": { "brier": 0.1428, "brierPartialAsMiss": 0.2218, "logLoss": 0.5724, "n": 10 }, "coinFlip": { "brier": 0.2, "brierPartialAsMiss": 0.25, "logLoss": 0.6931, "n": 10 }, "gbm": { "brier": 0.1428, "brierPartialAsMiss": 0.2218, "logLoss": 0.5724, "n": 10 }, "model": { "brier": 0.1428, "brierPartialAsMiss": 0.2218, "logLoss": 0.5724, "n": 10 }, "platt": { "brier": 0.1428, "brierPartialAsMiss": 0.2218, "logLoss": 0.5724, "n": 10 }, "stated": { "brier": 0.1428, "brierPartialAsMiss": 0.2218, "logLoss": 0.5724, "n": 10 } }, "testIds": [ "p49-micron-hbm4-booked", "p51-cpo-partner-rack", "p52-agent-automation-governance", "p53-skhy-debut-validates-memory", "p54-pjm-2028-29-at-cap", "p55-terra-confirms-repricing-cycle", "p56-samsung-hbm4-to-nvidia", "p57-gemini-3-5-pro-ga-jul31", "p58-harness-cost-telemetry", "p59-tsmc-q2-capex-raise" ], "testN": 10, "trainN": 0, "trainable": false }, { "cutoff": "2026-07-18", "fold": 4, "scores": { "baseRate": { "brier": 0.1438, "brierPartialAsMiss": 0.2483, "logLoss": 0.5826, "n": 9 }, "calibrated": { "brier": 0.1438, "brierPartialAsMiss": 0.2483, "logLoss": 0.5826, "n": 9 }, "coinFlip": { "brier": 0.1944, "brierPartialAsMiss": 0.25, "logLoss": 0.6931, "n": 9 }, "gbm": { "brier": 0.1438, "brierPartialAsMiss": 0.2483, "logLoss": 0.5826, "n": 9 }, "model": { "brier": 0.1438, "brierPartialAsMiss": 0.2483, "logLoss": 0.5826, "n": 9 }, "platt": { "brier": 0.1438, "brierPartialAsMiss": 0.2483, "logLoss": 0.5826, "n": 9 }, "stated": { "brier": 0.1438, "brierPartialAsMiss": 0.2483, "logLoss": 0.5826, "n": 9 } }, "testIds": [ "p61-kimi-k3-weights-aug10", "p62-deepseek-v4-ga-jul31", "p63-hbm-soldout-2027", "p64-colo-interconnect-outpaces", "p66-no-cheap-floor-reset", "p67-opus5-aa-gap-aug15", "p70-flash-task-cost-aug31", "p88-nvidia-10q-guaranty-exhibits-oct31", "p90-glm-53-weights-sep15" ], "testN": 9, "trainN": 0, "trainable": false } ], "reliability": [ { "bin": "0.0-0.1", "n": 0 }, { "bin": "0.1-0.2", "n": 0 }, { "bin": "0.2-0.3", "n": 0 }, { "bin": "0.3-0.4", "n": 0 }, { "bin": "0.4-0.5", "hits": 1, "meanOutcome": 1, "meanStated": 0.43, "misses": 0, "n": 1, "partials": 0 }, { "bin": "0.5-0.6", "hits": 4, "meanOutcome": 0.3667, "meanStated": 0.586, "misses": 8, "n": 15, "partials": 3 }, { "bin": "0.6-0.7", "hits": 12, "meanOutcome": 0.5556, "meanStated": 0.6626, "misses": 9, "n": 27, "partials": 6 }, { "bin": "0.7-0.8", "hits": 5, "meanOutcome": 0.7, "meanStated": 0.736, "misses": 1, "n": 10, "partials": 4 }, { "bin": "0.8-0.9", "hits": 1, "meanOutcome": 0.5, "meanStated": 0.81, "misses": 1, "n": 4, "partials": 2 }, { "bin": "0.9-1.0", "n": 0 } ], "walkForward": { "foldWins": { "baseRate": 1, "calibrated": 1, "coinFlip": 2, "gbm": 1, "model": 0, "platt": 0 }, "folds": 4, "perFold": [ { "cutoff": "2026-07-18", "fold": 0, "scores": { "baseRate": { "brier": 0.1835, "brierPartialAsMiss": 0.3826, "logLoss": 0.793, "n": 22 }, "calibrated": { "brier": 0.1835, "brierPartialAsMiss": 0.3826, "logLoss": 0.793, "n": 22 }, "coinFlip": { "brier": 0.1364, "brierPartialAsMiss": 0.25, "logLoss": 0.6931, "n": 22 }, "gbm": { "brier": 0.1835, "brierPartialAsMiss": 0.3826, "logLoss": 0.793, "n": 22 }, "model": { "brier": 0.1835, "brierPartialAsMiss": 0.3826, "logLoss": 0.793, "n": 22 }, "platt": { "brier": 0.1835, "brierPartialAsMiss": 0.3826, "logLoss": 0.793, "n": 22 }, "stated": { "brier": 0.1835, "brierPartialAsMiss": 0.3826, "logLoss": 0.793, "n": 22 } }, "testIds": [ "p1-2gw-customer-funded", "p4-interconnect-outpaces", "p6-custom-silicon-35", "p7-bedrock-gpt55-anchor", "p12-voice-frontier-procurement", "p14-anthropic-round-close", "p16-nvidia-circular-flag", "p17-samsung-walkout-shipments", "p23-anthropic-round-close-final", "p26-gemini-3-5-pro", "p28-gemini-3-5-pro-june", "p29-vera-rubin-cadence", "p32-gemini-3-5-pro-ga", "p37-gemini-3-5-pro-below-fable5", "p38-fable5-access-restored", "p49-micron-hbm4-booked", "p53-skhy-debut-validates-memory", "p54-pjm-2028-29-at-cap", "p57-gemini-3-5-pro-ga-jul31", "p59-tsmc-q2-capex-raise", "p62-deepseek-v4-ga-jul31", "p63-hbm-soldout-2027" ], "testN": 22, "trainN": 0, "trainable": false }, { "cutoff": "2026-08-08", "fold": 1, "scores": { "baseRate": { "brier": 0.405, "brierPartialAsMiss": 0.405, "logLoss": 1.0116, "n": 1 }, "calibrated": { "brier": 0.4216, "brierPartialAsMiss": 0.4216, "logLoss": 1.0478, "n": 1 }, "coinFlip": { "brier": 0.25, "brierPartialAsMiss": 0.25, "logLoss": 0.6931, "n": 1 }, "gbm": { "brier": 0.527, "brierPartialAsMiss": 0.527, "logLoss": 1.2945, "n": 1 }, "model": { "brier": 0.6008, "brierPartialAsMiss": 0.6008, "logLoss": 1.4922, "n": 1 }, "platt": { "brier": 0.2734, "brierPartialAsMiss": 0.2734, "logLoss": 0.7399, "n": 1 }, "stated": { "brier": 0.0784, "brierPartialAsMiss": 0.0784, "logLoss": 0.3285, "n": 1 } }, "testIds": [ "p61-kimi-k3-weights-aug10" ], "testN": 1, "trainN": 22, "trainable": true }, { "cutoff": "2026-09-01", "fold": 2, "scores": { "baseRate": { "brier": 0.2541, "brierPartialAsMiss": 0.2723, "logLoss": 0.7672, "n": 31 }, "calibrated": { "brier": 0.2362, "brierPartialAsMiss": 0.2612, "logLoss": 0.7302, "n": 31 }, "coinFlip": { "brier": 0.2177, "brierPartialAsMiss": 0.25, "logLoss": 0.6931, "n": 31 }, "gbm": { "brier": 0.3335, "brierPartialAsMiss": 0.3428, "logLoss": 0.9584, "n": 31 }, "model": { "brier": 0.2789, "brierPartialAsMiss": 0.2975, "logLoss": 0.8304, "n": 31 }, "platt": { "brier": 0.2535, "brierPartialAsMiss": 0.2771, "logLoss": 0.7665, "n": 31 }, "stated": { "brier": 0.2184, "brierPartialAsMiss": 0.2716, "logLoss": 0.6986, "n": 31 } }, "testIds": [ "p10-fabric-second-confirm", "p8-capex-revenue-ratio", "p13-mrc-second-fabric", "p18-claude-code-churn", "p20-cisco-ai-orders-tracking", "p21-ebo-msa-spec", "p22-nvidia-q2-fy27", "p24-agent-platform-lock-in", "p25-fabric-vendor-q2", "p27-anthropic-s1-public", "p30-optics-design-wins", "p31-sovereign-power-followthrough", "p34-hbm4-allocation-tightness", "p35-cpo-design-win", "p39-second-inference-asic-milestone", "p40-second-1-6t-design-win", "p41-no-public-s1", "p43-open-weight-top5", "p44-hbm4-allocation-2027", "p46-ferc-rto-compliance", "p47-hyperscaler-1gw-btm", "p48-closed-price-response", "p51-cpo-partner-rack", "p52-agent-automation-governance", "p55-terra-confirms-repricing-cycle", "p56-samsung-hbm4-to-nvidia", "p58-harness-cost-telemetry", "p64-colo-interconnect-outpaces", "p66-no-cheap-floor-reset", "p67-opus5-aa-gap-aug15", "p70-flash-task-cost-aug31" ], "testN": 31, "trainN": 23, "trainable": true }, { "cutoff": "2026-09-06", "fold": 3, "scores": { "baseRate": { "brier": 0.1547, "brierPartialAsMiss": 0.2442, "logLoss": 0.6691, "n": 3 }, "calibrated": { "brier": 0.1546, "brierPartialAsMiss": 0.2656, "logLoss": 0.669, "n": 3 }, "coinFlip": { "brier": 0.1667, "brierPartialAsMiss": 0.25, "logLoss": 0.6931, "n": 3 }, "gbm": { "brier": 0.153, "brierPartialAsMiss": 0.2248, "logLoss": 0.6576, "n": 3 }, "model": { "brier": 0.1914, "brierPartialAsMiss": 0.2813, "logLoss": 0.7456, "n": 3 }, "platt": { "brier": 0.2102, "brierPartialAsMiss": 0.3449, "logLoss": 0.7849, "n": 3 }, "stated": { "brier": 0.1644, "brierPartialAsMiss": 0.3478, "logLoss": 0.6963, "n": 3 } }, "testIds": [ "p3-capex-revise", "p88-nvidia-10q-guaranty-exhibits-oct31", "p90-glm-53-weights-sep15" ], "testN": 3, "trainN": 54, "trainable": true } ], "scheme": "resolution", "scores": { "baseRate": { "brier": 0.2243, "brierPartialAsMiss": 0.3157, "logLoss": 0.7763, "n": 57 }, "calibrated": { "brier": 0.2148, "brierPartialAsMiss": 0.3111, "logLoss": 0.7568, "n": 57 }, "coinFlip": { "brier": 0.1842, "brierPartialAsMiss": 0.25, "logLoss": 0.6931, "n": 57 }, "gbm": { "brier": 0.2695, "brierPartialAsMiss": 0.3552, "logLoss": 0.8846, "n": 57 }, "model": { "brier": 0.2431, "brierPartialAsMiss": 0.3348, "logLoss": 0.8231, "n": 57 }, "platt": { "brier": 0.2246, "brierPartialAsMiss": 0.3213, "logLoss": 0.7772, "n": 57 }, "stated": { "brier": 0.1996, "brierPartialAsMiss": 0.3151, "logLoss": 0.7284, "n": 57 } }, "trainableFolds": 3 } }, "referenceClasses": [ { "className": "confidence:0.50-0.59", "n": 6, "baseRate": 0.2857142857142857, "ci80": [ 0.09259525891312874, 0.5103163065514917 ], "source": "ledger", "definition": "Resolved predictions with stated confidence in band 0.50-0.59.", "hitsEquivalent": 1.5, "wilsonCi80": [ 0.0959, 0.5115 ] }, { "className": "confidence:0.60-0.69", "n": 29, "baseRate": 0.5666666666666667, "ci80": [ 0.4504089864069636, 0.6809289061301997 ], "source": "ledger", "definition": "Resolved predictions with stated confidence in band 0.60-0.69.", "hitsEquivalent": 16.5, "wilsonCi80": [ 0.4506, 0.68 ] }, { "className": "confidence:<0.50", "n": 1, "baseRate": 0.75, "ci80": [ 0.35135719802562754, 0.9938187560615703 ], "source": "ledger", "definition": "Resolved predictions with stated confidence in band <0.50.", "hitsEquivalent": 1, "wilsonCi80": [ 0.3784, 1 ] }, { "className": "confidence:>=0.70", "n": 21, "baseRate": 0.5454545454545454, "ci80": [ 0.40953505337805407, 0.6794872388819284 ], "source": "ledger", "definition": "Resolved predictions with stated confidence in band >=0.70.", "hitsEquivalent": 11.5, "wilsonCi80": [ 0.4101, 0.6783 ] }, { "className": "conjunctive:multi-leg", "n": 14, "baseRate": 0.43333333333333335, "ci80": [ 0.27345071719282527, 0.5973933210906863 ], "source": "ledger", "definition": "Resolved predictions that are multi-leg: multi-leg means two or more conditions must all hold.", "hitsEquivalent": 6, "wilsonCi80": [ 0.2755, 0.5966 ] }, { "className": "conjunctive:single-leg", "n": 43, "baseRate": 0.5681818181818182, "ci80": [ 0.4721805176695916, 0.6628122146388403 ], "source": "ledger", "definition": "Resolved predictions that are single-leg: multi-leg means two or more conditions must all hold.", "hitsEquivalent": 24.5, "wilsonCi80": [ 0.4722, 0.6622 ] }, { "className": "horizon:31-90d", "n": 29, "baseRate": 0.6166666666666667, "ci80": [ 0.5019224732357823, 0.7279193416465807 ], "source": "ledger", "definition": "Resolved predictions with issue-to-deadline horizon 31-90d.", "hitsEquivalent": 18, "wilsonCi80": [ 0.5017, 0.7267 ] }, { "className": "horizon:<=30d", "n": 14, "baseRate": 0.5, "ci80": [ 0.3364493452891413, 0.6635506547108587 ], "source": "ledger", "definition": "Resolved predictions with issue-to-deadline horizon <=30d.", "hitsEquivalent": 7, "wilsonCi80": [ 0.338, 0.662 ] }, { "className": "horizon:>90d", "n": 14, "baseRate": 0.4, "ci80": [ 0.24315696941972811, 0.5631076944400868 ], "source": "ledger", "definition": "Resolved predictions with issue-to-deadline horizon >90d.", "hitsEquivalent": 5.5, "wilsonCi80": [ 0.2455, 0.5628 ] }, { "className": "lbnl-queue-completion:type-gas", "n": 3559, "baseRate": 0.2574438202247191, "ci80": [ 0.24808280602842456, 0.2668632141415802 ], "source": "public-series", "definition": "LBNL Queued Up: share of gas interconnection requests (all vintages) that reached commercial operation by the data end year.", "hitsEquivalent": 916, "wilsonCi80": [ 0.2481, 0.2669 ] }, { "className": "lbnl-queue-completion:type-solar", "n": 13343, "baseRate": 0.0895908273381295, "ci80": [ 0.08643590692106227, 0.09277209282395524 ], "source": "public-series", "definition": "LBNL Queued Up: share of solar interconnection requests (all vintages) that reached commercial operation by the data end year.", "hitsEquivalent": 1195, "wilsonCi80": [ 0.0864, 0.0928 ] }, { "className": "lbnl-queue-completion:type-storage", "n": 5721, "baseRate": 0.030496329954561342, "ci80": [ 0.027620679857636686, 0.0334422679265864 ], "source": "public-series", "definition": "LBNL Queued Up: share of storage interconnection requests (all vintages) that reached commercial operation by the data end year.", "hitsEquivalent": 174, "wilsonCi80": [ 0.0276, 0.0335 ] }, { "className": "lbnl-queue-completion:type-wind", "n": 6658, "baseRate": 0.16331280973119086, "ci80": [ 0.15752975637248293, 0.1691391776652143 ], "source": "public-series", "definition": "LBNL Queued Up: share of wind interconnection requests (all vintages) that reached commercial operation by the data end year.", "hitsEquivalent": 1087, "wilsonCi80": [ 0.1575, 0.1692 ] }, { "className": "lbnl-queue-completion:vintage-2015-2020", "n": 11622, "baseRate": 0.13924976340015485, "ci80": [ 0.13514793568450353, 0.14337817790246243 ], "source": "public-series", "definition": "LBNL Queued Up: share of transmission interconnection requests submitted 2015-2020 that reached commercial operation by the data end year (2025). Project count basis, all regions and types.", "hitsEquivalent": 1618, "wilsonCi80": [ 0.1352, 0.1434 ] }, { "className": "lbnl-queue-completion:vintage-2018", "n": 2044, "baseRate": 0.1332518337408313, "ci80": [ 0.12370168352678387, 0.14295568862016095 ], "source": "public-series", "definition": "LBNL Queued Up: share of requests submitted in 2018 that reached commercial operation by the data end year.", "hitsEquivalent": 272, "wilsonCi80": [ 0.1237, 0.143 ] }, { "className": "lbnl-queue-completion:vintage-2019", "n": 2329, "baseRate": 0.11351931330472104, "ci80": [ 0.10517192229872166, 0.12200885351243483 ], "source": "public-series", "definition": "LBNL Queued Up: share of requests submitted in 2019 that reached commercial operation by the data end year.", "hitsEquivalent": 264, "wilsonCi80": [ 0.1052, 0.122 ] }, { "className": "lbnl-queue-completion:vintage-2020", "n": 2569, "baseRate": 0.07062256809338521, "ci80": [ 0.0642223148891675, 0.07716598635897069 ], "source": "public-series", "definition": "LBNL Queued Up: share of requests submitted in 2020 that reached commercial operation by the data end year.", "hitsEquivalent": 181, "wilsonCi80": [ 0.0643, 0.0772 ] }, { "className": "lbnl-queue-completion:vintage-2021", "n": 3135, "baseRate": 0.02853954081632653, "ci80": [ 0.02479941460719688, 0.032408463192864154 ], "source": "public-series", "definition": "LBNL Queued Up: share of requests submitted in 2021 that reached commercial operation by the data end year.", "hitsEquivalent": 89, "wilsonCi80": [ 0.0248, 0.0324 ] }, { "className": "lens:capital", "n": 11, "baseRate": 0.7083333333333334, "ci80": [ 0.536077634919035, 0.8639495515276331 ], "source": "ledger", "definition": "Resolved predictions in the capital lens.", "hitsEquivalent": 8, "wilsonCi80": [ 0.5345, 0.861 ] }, { "className": "lens:hardware", "n": 12, "baseRate": 0.6153846153846154, "ci80": [ 0.44099667591511876, 0.7813188571402074 ], "source": "ledger", "definition": "Resolved predictions in the hardware lens.", "hitsEquivalent": 7.5, "wilsonCi80": [ 0.4413, 0.7786 ] }, { "className": "lens:networking", "n": 10, "baseRate": 0.3181818181818182, "ci80": [ 0.1505912258762015, 0.5017811181000882 ], "source": "ledger", "definition": "Resolved predictions in the networking lens.", "hitsEquivalent": 3, "wilsonCi80": [ 0.1538, 0.5026 ] }, { "className": "lens:power", "n": 4, "baseRate": 0.7, "ci80": [ 0.43150701531066554, 0.9247429264997867 ], "source": "ledger", "definition": "Resolved predictions in the power lens.", "hitsEquivalent": 3, "wilsonCi80": [ 0.4325, 0.9219 ] }, { "className": "lens:software", "n": 20, "baseRate": 0.4523809523809524, "ci80": [ 0.31538108569605494, 0.5914577632380597 ], "source": "ledger", "definition": "Resolved predictions in the software lens.", "hitsEquivalent": 9, "wilsonCi80": [ 0.3167, 0.5909 ] }, { "className": "overall", "n": 57, "baseRate": 0.5344827586206896, "ci80": [ 0.45052494459444165, 0.6179186478684153 ], "source": "ledger", "definition": "All resolved ledger predictions; hit=1, partial=0.5, miss=0.", "hitsEquivalent": 30.5, "wilsonCi80": [ 0.4506, 0.6176 ] }, { "className": "qtype:disclosure", "n": 29, "baseRate": 0.5, "ci80": [ 0.3836593372727645, 0.6163406627272354 ], "source": "ledger", "definition": "Resolved predictions of type disclosure: disclosure = requires a named primary document or named company; threshold = numeric bar without a named-disclosure requirement; timing = event-by-date without a numeric bar; other = none of the above.", "hitsEquivalent": 14.5, "wilsonCi80": [ 0.3842, 0.6158 ] }, { "className": "qtype:other", "n": 1, "baseRate": 0.5, "ci80": [ 0.09999999999999998, 0.9 ], "source": "ledger", "definition": "Resolved predictions of type other: disclosure = requires a named primary document or named company; threshold = numeric bar without a named-disclosure requirement; timing = event-by-date without a numeric bar; other = none of the above.", "hitsEquivalent": 0.5, "wilsonCi80": [ 0.1058, 0.8942 ] }, { "className": "qtype:threshold", "n": 19, "baseRate": 0.65, "ci80": [ 0.511435608199323, 0.7816762783590812 ], "source": "ledger", "definition": "Resolved predictions of type threshold: disclosure = requires a named primary document or named company; threshold = numeric bar without a named-disclosure requirement; timing = event-by-date without a numeric bar; other = none of the above.", "hitsEquivalent": 12.5, "wilsonCi80": [ 0.5109, 0.7797 ] }, { "className": "qtype:timing", "n": 8, "baseRate": 0.3888888888888889, "ci80": [ 0.191659204491743, 0.5984076459543821 ], "source": "ledger", "definition": "Resolved predictions of type timing: disclosure = requires a named primary document or named company; threshold = numeric bar without a named-disclosure requirement; timing = event-by-date without a numeric bar; other = none of the above.", "hitsEquivalent": 3, "wilsonCi80": [ 0.1954, 0.5972 ] } ] }, "chain": { "head": "b374483ca574cfcf9ebed767f3854797da57430110221442cf1c643ce3008359", "events": 291, "verified": true }, "worstMisses": [ { "id": "p46-ferc-rto-compliance", "text": "At least one RTO/ISO files a large-load interconnection compliance proposal answering FERC's Jun 18 show-cause orders by the August 17, 2026 deadline.", "confidencePct": 80, "notes": "All six RTOs/ISOs sought ~90-day abeyances instead of filing substantive large-load interconnection compliance proposals by the Aug 17 show-cause deadline; July 20 informational reports do not satisfy the trigger's compliance-proposal bar.", "evidenceUrl": "https://mgrid.org/2026/08/13/all-six-grid-operators-ask-ferc-for-90-more-days-on-large-load-rules-pushing-filings-to-november-16/" }, { "id": "p64-colo-interconnect-outpaces", "text": "The largest colocation operators' Q2 prints (starting July 29) show interconnect/fabric revenue growth again outpacing overall revenue growth, sustaining the Metcalfe read for a second consecutive quarter.", "confidencePct": 71, "notes": "Equinix Q2 FY26 — the largest operator's Jul 29 print — grew interconnection revenue 11.3% YoY versus total revenue up 16.4%; interconnect did not outpace overall growth, and no other major operator's Q2 filing showed fabric/interconnect beating total revenue either.", "evidenceUrl": "https://www.sec.gov/Archives/edgar/data/1101239/000110123926000145/a991eqix-q226xpr.htm" }, { "id": "p1-2gw-customer-funded", "text": "At least one frontier lab announces a customer-funded compute commitment greater than 2 GW.", "confidencePct": 70, "notes": "No frontier lab announced a single customer-funded commitment >2 GW by June 30. The period's biggest customer-funded deal — Anthropic paying SpaceX/xAI up to $45B for Colossus access (disclosed May 20 in the S-1) — covered ~300 MW; Meta-Crusoe (June 18) topped out at 1.6 GW. Stargate additions remained structural build-out, not a single >2 GW customer-funded announcement.", "evidenceUrl": "https://www.datacenterdynamics.com/en/news/anthropic-to-use-all-of-spacex-xais-colossus-1-data-center-compute/" } ], "evidenceSnapshots": [ { "predictionId": "p29-vera-rubin-cadence", "capturedAt": "2026-07-18", "citationUrl": "https://www.datacenterknowledge.com/data-center-chips/nvidia-says-vera-rubin-vera-cpu-on-track-launches-dsx-os-to-run-ai-factories", "notes": "Timing leg hit: Huang used the June 1 Computex keynote to declare Vera Rubin in full production with customer shipments in Q3 2026. Framing leg missed: he called compute demand 'the constraint' and (June 5) confirmed all three HBM4 suppliers qualified and in production — supply framed as ramping cleanly, not as the binding constraint.", "outcome": "partial" }, { "predictionId": "p23-anthropic-round-close-final", "capturedAt": "2026-06-13", "citationUrl": null, "notes": "Valuation condition HIT: the round closed May 28 as a $65B Series H at a $965B post-money, inside the $900B-$1.1T band. The named-lead-investor condition is not yet publicly confirmed in-window, so scored partial pending the full lead roster before the June 13 deadline.", "outcome": "partial" }, { "predictionId": "p14-anthropic-round-close", "capturedAt": "2026-06-15", "citationUrl": null, "notes": "Bloomberg May 12 confirms talks at >$900B with target close end-May; round still unsigned. On track for the June 15 deadline; lead investor not yet disclosed. Score moves to hit when the round prices and the lead is named publicly.", "outcome": "partial" }, { "predictionId": "p1-2gw-customer-funded", "capturedAt": "2026-07-18", "citationUrl": "https://www.datacenterdynamics.com/en/news/anthropic-to-use-all-of-spacex-xais-colossus-1-data-center-compute/", "notes": "No frontier lab announced a single customer-funded commitment >2 GW by June 30. The period's biggest customer-funded deal — Anthropic paying SpaceX/xAI up to $45B for Colossus access (disclosed May 20 in the S-1) — covered ~300 MW; Meta-Crusoe (June 18) topped out at 1.6 GW. Stargate additions remained structural build-out, not a single >2 GW customer-funded announcement.", "outcome": "miss" }, { "predictionId": "p28-gemini-3-5-pro-june", "capturedAt": "2026-07-18", "citationUrl": "https://www.searchenginejournal.com/gemini-3-5-pro-delayed-over-coding-bloomberg-reports/582660/", "notes": "GA premise failed: Gemini 3.5 Pro had not reached general availability by June 30 (third slip; Bloomberg reported July 16 it was months behind schedule on coding quality), so the AA Intelligence Index >= 61 leg was never tested.", "outcome": "miss" }, { "predictionId": "p32-gemini-3-5-pro-ga", "capturedAt": "2026-07-18", "citationUrl": "https://ai.google.dev/gemini-api/docs/models", "notes": "GA premise failed: Gemini 3.5 Pro had not shipped by June 30 — the public Gemini API still listed only gemini-3.5-flash and gemini-3.1-pro as of mid-July — so the SWE-Bench Pro comparison leg was never tested.", "outcome": "miss" }, { "predictionId": "p37-gemini-3-5-pro-below-fable5", "capturedAt": "2026-07-18", "citationUrl": "https://www.searchenginejournal.com/gemini-3-5-pro-delayed-over-coding-bloomberg-reports/582660/", "notes": "GA premise failed: Gemini 3.5 Pro had not reached public GA by June 30 (its third slip), so its debut position on the Artificial Analysis Intelligence Index was never tested.", "outcome": "miss" }, { "predictionId": "p54-pjm-2028-29-at-cap", "capturedAt": "2026-07-18", "citationUrl": "https://www.pjm.com/-/media/DotCom/markets-ops/rpm/rpm-auction-info/2028-2029/2028-2029-bra-results-report.pdf", "notes": "The PJM 2028/29 base residual auction cleared exactly at the $325.00/MW-day cap RTO-wide, posted July 14 as predicted — squarely within the 5% band.", "outcome": "hit" }, { "predictionId": "p26-gemini-3-5-pro", "capturedAt": "2026-07-18", "citationUrl": "https://www.bloomberg.com/news/articles/2026-07-16/google-gemini-launch-delayed-as-tech-falls-short-of-internal-goals", "notes": "Gemini 3.5 Pro had not launched by July 15 — it missed the leaked July 17 target too, its third slip, with Bloomberg reporting July 16 the model was months behind schedule — so the AA Index >= 60 leg was never tested.", "outcome": "miss" }, { "predictionId": "p59-tsmc-q2-capex-raise", "capturedAt": "2026-07-16", "citationUrl": null, "notes": "The capex leg hit decisively — TSMC raised the entire FY26 range to $60-64B (from $52-56B), well beyond reiterating the top end. The revenue leg fell just short: HPC at 66% of a $40.2B quarter implies ~47% YoY platform growth versus the >50% bar. The thesis behind the prediction (packaging-constrained ramp) was confirmed on the record by the CEO's packaging warning (quoted in the hardware lens).", "outcome": "partial" }, { "predictionId": "p63-hbm-soldout-2027", "capturedAt": "2026-08-01", "citationUrl": "https://news.skhynix.com/en/q2-2026-business-results/", "notes": "LTA leg hit: SK hynix's Jul 29 Q2 release and call confirmed Long-Term Agreements with around 10 key customers and HBM4 mass shipments underway. Sold-out leg missed: management said 2027 HBM volumes and prices were still under negotiation — no primary disclosure that 2027 capacity is substantially sold out or committed.", "outcome": "partial" }, { "predictionId": "p4-interconnect-outpaces", "capturedAt": "2026-07-31", "citationUrl": null, "notes": "Equinix Q1 Apr 29: Fabric revenue +26% YoY, Fabric bookings +70% YoY, raised guidance — first of two quarters needed. Score moves to hit if Q2 confirms the same outpacing, expected early August.", "outcome": "partial" }, { "predictionId": "p6-custom-silicon-35", "capturedAt": "2026-08-01", "citationUrl": "https://www.trendforce.com/presscenter/news/20260120-12887.html", "notes": "No TrendForce / SemiAnalysis quarterly mix breakdown showed custom silicon at 35% of incremental AI compute share by Jul 31. The best public TrendForce figure kept ASIC-based AI servers at a projected 27.8% of 2026 shipments (GPUs 69.7%) — ~7pp short even on stock share, with no verified incremental-share print crossing 35%.", "outcome": "miss" }, { "predictionId": "p7-bedrock-gpt55-anchor", "capturedAt": "2026-08-01", "citationUrl": "https://aws.amazon.com/blogs/machine-learning/openai-models-and-codex-on-amazon-bedrock-are-now-generally-available/", "notes": "Bedrock GA for GPT-5.5 landed (AWS Jun 2026 blog; Amgen and Autodesk named as evaluating), but no customer-disclosed Bedrock-resident GPT-5.5 multi-year contract greater than $500M annualized with a named customer appeared by Jul 31.", "outcome": "miss" }, { "predictionId": "p12-voice-frontier-procurement", "capturedAt": "2026-08-01", "citationUrl": "https://openai.com/index/advancing-voice-intelligence-with-new-models-in-the-api/", "notes": "Named-trigger customers (Zillow, Priceline, Deutsche Telekom) were cited building on GPT-Realtime-2, but none disclosed replacement of an existing voice/IVR platform on a contract greater than $25M annualized by Jul 31 — no qualifying Fortune 500 procurement disclosure surfaced.", "outcome": "miss" }, { "predictionId": "p16-nvidia-circular-flag", "capturedAt": "2026-08-01", "citationUrl": "https://247wallst.com/investing/2026/05/11/jpmorgan-hikes-iren-price-target-to-46-but-stays-underweight-on-circular-nvidia-deal-concerns/", "notes": "JPMorgan equity research on IREN (May 11) kept an Underweight rating while flagging the NVIDIA partnership's \"circular nature\" and undefined GPU access as tempering the deal — a major sell-side written report explicitly framing an NVIDIA 2026 customer-equity counterparty as circular-financing risk, which is what the trigger asked for. Recorded with a caveat: the JPMorgan note itself is subscriber-only, so the citation is contemporaneous coverage quoting it rather than the primary document. Holding out for the PDF would make sell-side predictions effectively unresolvable, but the weaker evidence chain is noted.", "outcome": "hit" }, { "predictionId": "p17-samsung-walkout-shipments", "capturedAt": "2026-07-31", "citationUrl": null, "notes": "Samsung-union tentative agreement May 20 averted the 18-day walkout; ratification vote May 27-28. HBM4 shipments likely NOT to drop >25%; this prediction is on track to MISS. Confidence revised down. Score moves to miss if the ratification passes; score moves to hit only if the ratification fails and the strike actually proceeds.", "outcome": "partial" }, { "predictionId": "p38-fable5-access-restored", "capturedAt": "2026-08-01", "citationUrl": "https://www.anthropic.com/news/redeploying-fable-5", "notes": "Anthropic's Jun 30 / Jul 1 redeployment post confirmed public Fable 5 access restored globally across the Claude Platform, Claude.ai, Claude Code, and Claude Cowork after export controls lifted, well inside the Jul 31 window. Scored partial rather than hit because the prediction specified restoration \"with geo-gating or KYC controls\" and the actual restoration was global with safety classifiers and no geographic or identity gating. The timing and direction were right; the predicted access-control mechanism did not materialize.", "outcome": "partial" }, { "predictionId": "p49-micron-hbm4-booked", "capturedAt": "2026-07-31", "citationUrl": null, "notes": "Hit. Samsung's CFO stated at the Apr 30 call that 2026 HBM4 output is fully sold out, and in-window reporting says SK hynix removed price caps from long-term agreements — explicit 2027 price-up posture. Caveat: the in-window items are supply-chain reporting; SK hynix's late-July earnings call would make it unambiguous.", "outcome": "hit" }, { "predictionId": "p53-skhy-debut-validates-memory", "capturedAt": "2026-07-31", "citationUrl": null, "notes": "Partial. The pricing clause missed — the offering priced at $149/ADS, below the ~$166 indication — but the market clause is on track: demand was ~7x available shares and day one closed up ~13% at ~$168, well above offer. The memory thesis was validated by the demand, not the price; regular trading (SKHY) begins Jul 13.", "outcome": "partial" }, { "predictionId": "p57-gemini-3-5-pro-ga-jul31", "capturedAt": "2026-08-01", "citationUrl": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-6-flash-3-5-flash-lite-3-5-flash-cyber/", "notes": "Google's Jul 21 post still described Gemini 3.5 Pro as \"testing with partners\" while shipping 3.6 Flash / 3.5 Flash-Lite / Flash Cyber instead — no public GA gemini-3.5-pro model ID with published pricing appeared through Jul 31.", "outcome": "miss" }, { "predictionId": "p62-deepseek-v4-ga-jul31", "capturedAt": "2026-07-31", "citationUrl": null, "notes": "The alias-retirement leg hit at 15:59 UTC Jul 24 with hard failure and no redirect. Explicit V4 Pro/Flash IDs and thinking parameter support the migration thesis, but the full GA/pricing trigger is not yet documented strongly enough for a hit.", "outcome": "partial" }, { "predictionId": "p61-kimi-k3-weights-aug10", "capturedAt": "2026-08-08", "citationUrl": "https://huggingface.co/moonshotai/Kimi-K3", "notes": "Moonshot published Kimi K3 weights on Hugging Face on Jul 27 with license text, and the model was reclassified from closed to open weights on that drop. Scored a hit rather than a partial: the Kimi K3 License is MIT-like for internal commercial use, which is what \"commercial self-hosting\" asked for, and the revenue gate it does impose applies to MaaS operators reselling the model above $20M trailing-twelve-month revenue rather than to self-hosters. Resolved two days ahead of the Aug 10 deadline because the trigger fired unambiguously on Jul 27 and no later evidence could reverse it.", "outcome": "hit" }, { "predictionId": "p21-ebo-msa-spec", "capturedAt": "2026-08-12", "citationUrl": null, "notes": "EBO MSA membership expanded 17 to 23 vendors May 18 (HPE marquee addition, Bellwether, JPC Connectivity, Mixx, TIME, TFC). v1.0 spec not yet published. Member growth is positive signal but spec + in-production deployment still pending. On track.", "outcome": "partial" }, { "predictionId": "p8-capex-revenue-ratio", "capturedAt": "2026-09-01", "citationUrl": "https://io-fund.com/ai-stocks/big-tech-ai-revenue-surges-suppliers-win-bigger", "notes": "Q2 2026 prints put every top-4 hyperscaler capex/AI-revenue ratio below 5.0x on published run-rate math: Meta ~1.66x, Microsoft Azure ~3.64x, Google Cloud ~3.84x, AWS ~4.25x — confirming the sub-5.0 directional flip the trigger asked for.", "outcome": "hit" }, { "predictionId": "p18-claude-code-churn", "capturedAt": "2026-09-01", "citationUrl": "https://thenewstack.io/anthropic-pauses-claude-agent-sdk-subscription-change/", "notes": "Anthropic paused the planned June 15 Agent SDK credit split on the scheduled date, so the pricing cutover never took effect and no >20% Claude Code third-party user drop was disclosed or surveyed within 30 days of a change that did not happen.", "outcome": "miss" }, { "predictionId": "p64-colo-interconnect-outpaces", "capturedAt": "2026-09-01", "citationUrl": "https://www.sec.gov/Archives/edgar/data/1101239/000110123926000145/a991eqix-q226xpr.htm", "notes": "Equinix Q2 FY26 — the largest operator's Jul 29 print — grew interconnection revenue 11.3% YoY versus total revenue up 16.4%; interconnect did not outpace overall growth, and no other major operator's Q2 filing showed fabric/interconnect beating total revenue either.", "outcome": "miss" }, { "predictionId": "p67-opus5-aa-gap-aug15", "capturedAt": "2026-09-01", "citationUrl": "https://artificialanalysis.ai/models/comparisons/claude-opus-5-xhigh-vs-claude-fable-5", "notes": "Artificial Analysis v4.1.1 (Aug 6) scored Claude Opus 5 (max) at 63 and Claude Fable 5 at 62 — a one-point gap, inside the three-point bar — before the Aug 15 deadline.", "outcome": "hit" }, { "predictionId": "p46-ferc-rto-compliance", "capturedAt": "2026-09-01", "citationUrl": "https://mgrid.org/2026/08/13/all-six-grid-operators-ask-ferc-for-90-more-days-on-large-load-rules-pushing-filings-to-november-16/", "notes": "All six RTOs/ISOs sought ~90-day abeyances instead of filing substantive large-load interconnection compliance proposals by the Aug 17 show-cause deadline; July 20 informational reports do not satisfy the trigger's compliance-proposal bar.", "outcome": "miss" }, { "predictionId": "p20-cisco-ai-orders-tracking", "capturedAt": "2026-09-01", "citationUrl": "https://investor.cisco.com/news/news-details/2026/CISCO-REPORTS-FOURTH-QUARTER-AND-FISCAL-YEAR-2026-EARNINGS/default.aspx", "notes": "Cisco's Aug 13 Q4 FY26 release reported $4.0B of hyperscaler AI infrastructure orders in the quarter — above the $2.5B Q4 bar — and $9.3B for the full fiscal year, validating the ~$9B annual run-rate thesis.", "outcome": "hit" }, { "predictionId": "p10-fabric-second-confirm", "capturedAt": "2026-09-01", "citationUrl": "https://www.sec.gov/Archives/edgar/data/2105398/000210539826000002/csqr-ex99_1.htm", "notes": "No major colocation operator other than Equinix reported Q2 2026 interconnect revenue growth above 25% YoY: CoreSite grew total revenue 7.7% without a >25% interconnect line, and Csquare's interconnect revenue fell 10.3% YoY to $24.7M.", "outcome": "miss" }, { "predictionId": "p13-mrc-second-fabric", "capturedAt": "2026-09-01", "citationUrl": "https://openai.com/index/mrc-supercomputer-networking/", "notes": "Public MRC production deployments at >50k-GPU scale are attributed to Microsoft/OpenAI sites (Fairwater, Abilene); no Google, Amazon, or Meta hyperscaler other than Microsoft disclosed a production AI fabric running MRC above the 50,000-GPU threshold by Aug 31.", "outcome": "miss" }, { "predictionId": "p22-nvidia-q2-fy27", "capturedAt": "2026-09-01", "citationUrl": "https://www.sec.gov/Archives/edgar/data/1045810/000104581026000073/q2fy27cfocommentary.htm", "notes": "NVIDIA's Aug 26 Q2 FY27 release reported Data Center revenue of $89.0B — $1.0B below the $90B-plus trigger despite beating the $91B guide midpoint on consolidated revenue.", "outcome": "miss" }, { "predictionId": "p24-agent-platform-lock-in", "capturedAt": "2026-09-01", "citationUrl": "https://www.anthropic.com/news/claude-partner-network", "notes": "No Fortune 500 enterprise disclosed standardizing on a single agent-platform runtime (Claude Code / Codex / Antigravity 2.0 / Grok Build) with a named >$50M annualized commitment; Anthropic's $100M partner-network spend is vendor-side, not a customer lock-in disclosure.", "outcome": "miss" }, { "predictionId": "p25-fabric-vendor-q2", "capturedAt": "2026-09-01", "citationUrl": "https://investors.arista.com/Communications/Press-Releases-and-Events/Press-Release-Detail/2026/Arista-Networks-Inc--Reports-Second-Quarter-2026-Financial-Results/default.aspx", "notes": "None of the four vendors disclosed a standalone networking-segment print above 40% YoY in Q2: Arista total revenue grew 37.7%, Marvell reports data-center (not networking-only) growth, Broadcom cites AI networking as ~40% of AI semiconductors without a >40% networking-segment line, and NVIDIA does not break out networking YoY separately.", "outcome": "miss" }, { "predictionId": "p27-anthropic-s1-public", "capturedAt": "2026-09-01", "citationUrl": "https://decodethefuture.org/en/anthropic-s1-ipo-filing-explained/", "notes": "Neither Anthropic nor OpenAI converted its confidential draft to a publicly visible S-1 on SEC EDGAR before Aug 31, 2026 — only the Jun 1 confidential submission and DRS-stage filings remained non-public through the deadline.", "outcome": "hit" }, { "predictionId": "p30-optics-design-wins", "capturedAt": "2026-09-01", "citationUrl": "https://www.fool.com/earnings/call-transcripts/2026/08/31/marvell-mrvl-q2-2027-earnings-call-transcript/", "notes": "Marvell's Aug 31 Q2 FY27 call cited secured 1.6T optical-DSP design wins and a rapidly ramping 1.6T business (one leg). Credo said on its Q2 call it sees no significant customer movement toward co-packaged optics, and Broadcom's Q2 print discussed CPO programs without a second qualifying 1.6T/CPO design-win citation — one of two required vendors.", "outcome": "partial" }, { "predictionId": "p31-sovereign-power-followthrough", "capturedAt": "2026-09-01", "citationUrl": "https://www.datacenterknowledge.com/energy-power-supply/amazon-explores-grid-connection-for-8-000-acre-ai-campus", "notes": "Amazon confirmed in August that it acquired the GW Ranch site in Pecos County, Texas, for an AI campus powered by Pacifico Energy's planned 7.65GW behind-the-meter gas-and-storage system — a named hyperscaler >1GW BTM generation deal inside the Aug 31 window.", "outcome": "hit" }, { "predictionId": "p34-hbm4-allocation-tightness", "capturedAt": "2026-09-01", "citationUrl": "https://en.sedaily.com/finance/2026/08/31/spot-hbm-prices-hit-5-times-contract-levels", "notes": "Before Aug 31, supply-chain and trade press documented HBM4 allocation tightness despite three-supplier qualification: Seoul Economic Daily (Aug 31) reported spot HBM prices at 4–5x LTA levels with customers not receiving full contracted volumes even under LTAs.", "outcome": "hit" }, { "predictionId": "p35-cpo-design-win", "capturedAt": "2026-08-31", "citationUrl": null, "notes": "Arista's 1.6T 7060XE7 portfolio on Broadcom's Tomahawk 6 (Jun 9) is a fresh Broadcom 1.6T production design win, satisfying the 1.6T leg; no co-packaged-optics production win or vendor revenue-guide uplift yet. Tracking to a full hit by deadline.", "outcome": "partial" }, { "predictionId": "p39-second-inference-asic-milestone", "capturedAt": "2026-09-01", "citationUrl": "https://investor.nvidia.com/news/press-release-details/2026/NVIDIA-Groq-3-LPX-Now-in-Full-Production-With-World-Class-Speed-for-Agentic-AI/default.aspx", "notes": "NVIDIA announced Aug 24 that Groq 3 LPX — the merchant inference ASIC line acquired from Groq — is in full production, with Nebius planning production deployment on Nebius Token Factory later in 2026, satisfying the second-vendor volume-production milestone bar.", "outcome": "hit" }, { "predictionId": "p40-second-1-6t-design-win", "capturedAt": "2026-09-01", "citationUrl": "https://www.fool.com/earnings/call-transcripts/2026/08/31/marvell-mrvl-q2-2027-earnings-call-transcript/", "notes": "Marvell's Aug 31 Q2 FY27 earnings call cited secured design wins and a rapidly ramping 1.6T optical-DSP business, joining Broadcom's Tomahawk 6 as the second named vendor with a production-scale 1.6T win disclosure before Aug 31.", "outcome": "hit" }, { "predictionId": "p41-no-public-s1", "capturedAt": "2026-09-01", "citationUrl": "https://decodethefuture.org/en/anthropic-s1-ipo-filing-explained/", "notes": "SEC EDGAR had no publicly visible OpenAI or Anthropic S-1 through Aug 31, 2026; both remained at the confidential-DRS stage after their June confidential submissions.", "outcome": "hit" }, { "predictionId": "p43-open-weight-top5", "capturedAt": "2026-09-01", "citationUrl": "https://artificialanalysis.ai/models", "notes": "AA Intelligence Index v4.1.1's overall top five through Aug 31 were all Anthropic closed variants (Fable 5.1 and Opus 5 scores 62–66); the highest MIT/Apache-licensed open-weight model, Kimi K3 at 60, sat outside the overall top five, and GLM-5.3 uses a bespoke license rather than MIT/Apache.", "outcome": "miss" }, { "predictionId": "p44-hbm4-allocation-2027", "capturedAt": "2026-09-01", "citationUrl": "https://news.skhynix.com/en/q2-2026-business-results/", "notes": "Samsung's Jul 30 call and industry reporting described 2027 memory/HBM capacity largely committed with rising contract prices, but SK hynix's Jul 29 Q2 release still said 2027 HBM volumes and prices were under negotiation — not all three primary makers confirmed fully allocated 2027 HBM on their own disclosures.", "outcome": "partial" }, { "predictionId": "p47-hyperscaler-1gw-btm", "capturedAt": "2026-09-01", "citationUrl": "https://www.datacenterknowledge.com/energy-power-supply/amazon-explores-grid-connection-for-8-000-acre-ai-campus", "notes": "Amazon confirmed Aug 7+ that it acquired GW Ranch in Texas and will buy power from Pacifico Energy's planned 7.65GW behind-the-meter generation complex for an on-site AI data center — a named hyperscaler >1GW off-grid/BTM deal by Aug 31.", "outcome": "hit" }, { "predictionId": "p48-closed-price-response", "capturedAt": "2026-09-01", "citationUrl": "https://www.reuters.com/technology/openai-cuts-developer-pricing-frontier-gpt-56-sol-model-by-more-than-20-2026-08-21/", "notes": "OpenAI cut flagship GPT-5.6 Sol API list prices >20% on Aug 21 (to $4/$20 per MTok from $5/$30), explicitly framing the move against Anthropic and Chinese-model competition — a major closed-lab flagship API price cut inside the deadline.", "outcome": "hit" }, { "predictionId": "p51-cpo-partner-rack", "capturedAt": "2026-09-01", "citationUrl": "https://investor.nvidia.com/news/press-release-details/2026/NVIDIA-Vera-Rubin-Ramps-Into-Full-Production-to-Power-Agentic-AI-Factories-Worldwide/default.aspx", "notes": "Supermicro announced CPO-capable Vera Rubin rack blueprints in June and NVIDIA named Spectrum-X Photonics adopters, but no named Vera Rubin partner publicly announced a CPO/Spectrum-X Ethernet Photonics rack or cluster design win — product/integration announcements fall short of the trigger's design-win bar.", "outcome": "miss" }, { "predictionId": "p52-agent-automation-governance", "capturedAt": "2026-08-31", "citationUrl": null, "notes": "Hit. GitHub shipped Copilot agent session streaming to public preview (Jul 2) — SIEM/Purview streaming of all agent sessions — on top of its agent control plane, and GitHub also added AI-credit session limits covering background agents (Jul 1, per Agent Techniques coverage).", "outcome": "hit" }, { "predictionId": "p55-terra-confirms-repricing-cycle", "capturedAt": "2026-08-31", "citationUrl": null, "notes": "Hit, seven weeks early. GPT-5.6 went GA Jul 9 with Terra at exactly $2.50/$15 per MTok. Grok 4.5's $2/$6 launch the day before makes it a three-vendor repricing cycle (Sonnet 5, Terra, Grok 4.5), not a one-off.", "outcome": "hit" }, { "predictionId": "p56-samsung-hbm4-to-nvidia", "capturedAt": "2026-08-31", "citationUrl": null, "notes": "Hit on the multi-source-reporting trigger: Korean press (Seoul Economic Daily, Korea Herald) reported alongside Samsung's record Q2 guidance that HBM4 — in mass production since February for NVIDIA's Vera Rubin — reached $1B in sales within four months. Caveat: Samsung's Jul 30 divisional results would make it unambiguous from the company itself.", "outcome": "hit" }, { "predictionId": "p58-harness-cost-telemetry", "capturedAt": "2026-09-01", "citationUrl": "https://cursor.com/blog/router", "notes": "Cursor shipped Cursor Router in July 2026 with Auto Balance/Intelligence routing controls and published measured cost-per-commit figures ($4.63–$6.76) from live traffic — product-level harness routing and cost telemetry beyond session budget caps.", "outcome": "hit" }, { "predictionId": "p66-no-cheap-floor-reset", "capturedAt": "2026-09-01", "citationUrl": "https://api-docs.deepseek.com/quick_start/pricing/", "notes": "DeepSeek's official API pricing page kept GA deepseek-v4-pro off-peak output at $1.98/MTok (~¥14+) through Aug 31 — well above the ¥6 (~$0.85)/MTok ultra-cheap floor the trigger set as the kill condition.", "outcome": "hit" }, { "predictionId": "p70-flash-task-cost-aug31", "capturedAt": "2026-09-01", "citationUrl": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-6-flash-3-5-flash-lite-3-5-flash-cyber/", "notes": "Artificial Analysis measured Gemini 3.6 Flash at $0.50 average cost per completed agentic task versus $0.59 for 3.5 Flash — a 15% reduction, above the 12% cheaper-per-task bar — before Aug 31.", "outcome": "hit" }, { "predictionId": "p90-glm-53-weights-sep15", "capturedAt": "2026-09-15", "citationUrl": null, "notes": "Z.ai published the full 753B-parameter GLM-5.3 weights to Hugging Face at zai-org/GLM-5.3 on August 27–28, 2026 — in-window and inside the trigger's September 15 window, distinct from GLM-5.2 — after GLM-5.3-Flash MIT weights landed Aug 26. The material nuance is licensing, not availability: GLM-5.3 ships under a bespoke GLM-5.3 license rather than MIT, requiring Z.AI security review before commercial use by any Model-as-a-Service operator whose group revenue exceeds $10B over any 12 consecutive months.", "outcome": "hit" }, { "predictionId": "p3-capex-revise", "capturedAt": "2026-10-31", "citationUrl": null, "notes": "Q1 prints (MSFT $190B, GOOG $180-190B, META $125-145B, AMZN $200B reaffirmed) take 2026 aggregate to $695-725B (+77% YoY) vs the $700B W17 baseline. At/near baseline; +10% revision (~$770B) plausible by Q2 print. Score moves to hit if Q2 takes aggregate above $770B.", "outcome": "partial" }, { "predictionId": "p88-nvidia-10q-guaranty-exhibits-oct31", "capturedAt": "2026-10-31", "citationUrl": null, "notes": "NVIDIA filed the Form 10-Q for the quarter ended July 26, 2026 on August 26, 2026 — inside the window. It satisfies all three trigger elements: guarantees 'capped at a total of $105 billion' with an exposure table of $3.5B AI-cloud guarantees plus $105.0B SB Energy for $108.5B total; effectiveness conditioned on SB Energy satisfying applicable ready-for-service conditions as each of nine phases is placed in service from fiscal 2029; and the tenant identified as 'an affiliate of OpenAI Group PBC' at the PORTS Technology Campus in Pike County, Ohio. Exhibit 10.1 is the Form of Residual Value Guaranty.", "outcome": "hit" } ], "entries": [ { "id": "p29-vera-rubin-cadence", "text": "At GTC Taipei / Computex (June 1), NVIDIA reaffirms Vera Rubin production starting in 2H 2026 and frames HBM4 + CoWoS as the binding supply constraint rather than demand.", "confidencePct": 75, "deadline": "By June 7, 2026", "deadlineDate": "2026-06-07", "lens": "hardware", "firstIssue": "2026-W22", "outcome": "partial", "overdue": false, "notes": "Timing leg hit: Huang used the June 1 Computex keynote to declare Vera Rubin in full production with customer shipments in Q3 2026. Framing leg missed: he called compute demand 'the constraint' and (June 5) confirmed all three HBM4 suppliers qualified and in production — supply framed as ramping cleanly, not as the binding constraint.", "evidenceUrl": "https://www.datacenterknowledge.com/data-center-chips/nvidia-says-vera-rubin-vera-cpu-on-track-launches-dsx-os-to-run-ai-factories", "resolvedAt": "2026-07-18" }, { "id": "p23-anthropic-round-close-final", "text": "Anthropic round closes at a final post-money valuation between $900B and $1.1T with at least three of (Sequoia, Dragoneer, Altimeter, Greenoaks) named as lead investors publicly.", "confidencePct": 80, "deadline": "By June 13, 2026", "deadlineDate": "2026-06-13", "lens": "capital", "firstIssue": "2026-W21", "outcome": "partial", "overdue": false, "notes": "Valuation condition HIT: the round closed May 28 as a $65B Series H at a $965B post-money, inside the $900B-$1.1T band. The named-lead-investor condition is not yet publicly confirmed in-window, so scored partial pending the full lead roster before the June 13 deadline." }, { "id": "p14-anthropic-round-close", "text": "Anthropic's pending raise closes at a final post-money valuation between $850B and $1T, with the lead investor and check size publicly disclosed.", "confidencePct": 75, "deadline": "By June 15, 2026", "deadlineDate": "2026-06-15", "lens": "capital", "firstIssue": "2026-W19", "outcome": "partial", "overdue": false, "notes": "Bloomberg May 12 confirms talks at >$900B with target close end-May; round still unsigned. On track for the June 15 deadline; lead investor not yet disclosed. Score moves to hit when the round prices and the lead is named publicly." }, { "id": "p1-2gw-customer-funded", "text": "At least one frontier lab announces a customer-funded compute commitment greater than 2 GW.", "confidencePct": 70, "deadline": "By June 30, 2026", "deadlineDate": "2026-06-30", "lens": "capital", "firstIssue": "2026-W17", "outcome": "miss", "overdue": false, "notes": "No frontier lab announced a single customer-funded commitment >2 GW by June 30. The period's biggest customer-funded deal — Anthropic paying SpaceX/xAI up to $45B for Colossus access (disclosed May 20 in the S-1) — covered ~300 MW; Meta-Crusoe (June 18) topped out at 1.6 GW. Stargate additions remained structural build-out, not a single >2 GW customer-funded announcement.", "evidenceUrl": "https://www.datacenterdynamics.com/en/news/anthropic-to-use-all-of-spacex-xais-colossus-1-data-center-compute/", "resolvedAt": "2026-07-18" }, { "id": "p28-gemini-3-5-pro-june", "text": "Gemini 3.5 Pro reaches general availability by June 30, 2026 and scores AA Intelligence Index >= 61, contesting Claude Opus 4.8's fresh lead.", "confidencePct": 60, "deadline": "By June 30, 2026", "deadlineDate": "2026-06-30", "lens": "software", "firstIssue": "2026-W22", "outcome": "miss", "overdue": false, "notes": "GA premise failed: Gemini 3.5 Pro had not reached general availability by June 30 (third slip; Bloomberg reported July 16 it was months behind schedule on coding quality), so the AA Intelligence Index >= 61 leg was never tested.", "evidenceUrl": "https://www.searchenginejournal.com/gemini-3-5-pro-delayed-over-coding-bloomberg-reports/582660/", "resolvedAt": "2026-07-18" }, { "id": "p32-gemini-3-5-pro-ga", "text": "Gemini 3.5 Pro reaches public GA by June 30, 2026, but does not exceed Claude Opus 4.8 on SWE-Bench Pro in its first independent Artificial Analysis run.", "confidencePct": 60, "deadline": "By June 30, 2026", "deadlineDate": "2026-06-30", "lens": "software", "firstIssue": "2026-W23", "outcome": "miss", "overdue": false, "notes": "GA premise failed: Gemini 3.5 Pro had not shipped by June 30 — the public Gemini API still listed only gemini-3.5-flash and gemini-3.1-pro as of mid-July — so the SWE-Bench Pro comparison leg was never tested.", "evidenceUrl": "https://ai.google.dev/gemini-api/docs/models", "resolvedAt": "2026-07-18" }, { "id": "p37-gemini-3-5-pro-below-fable5", "text": "Gemini 3.5 Pro reaches public GA by June 30, 2026 but debuts below Claude Fable 5 on the Artificial Analysis Intelligence Index in its first independent pass.", "confidencePct": 58, "deadline": "By June 30, 2026", "deadlineDate": "2026-06-30", "lens": "software", "firstIssue": "2026-W24", "outcome": "miss", "overdue": false, "notes": "GA premise failed: Gemini 3.5 Pro had not reached public GA by June 30 (its third slip), so its debut position on the Artificial Analysis Intelligence Index was never tested.", "evidenceUrl": "https://www.searchenginejournal.com/gemini-3-5-pro-delayed-over-coding-bloomberg-reports/582660/", "resolvedAt": "2026-07-18" }, { "id": "p54-pjm-2028-29-at-cap", "text": "The PJM 2028/29 base residual auction clears within 5% of the ~$325/MW-day cap when results post on July 14, 2026.", "confidencePct": 74, "deadline": "By July 14, 2026", "deadlineDate": "2026-07-14", "lens": "power", "firstIssue": "2026-W27", "outcome": "hit", "overdue": false, "notes": "The PJM 2028/29 base residual auction cleared exactly at the $325.00/MW-day cap RTO-wide, posted July 14 as predicted — squarely within the 5% band.", "evidenceUrl": "https://www.pjm.com/-/media/DotCom/markets-ops/rpm/rpm-auction-info/2028-2029/2028-2029-bra-results-report.pdf", "resolvedAt": "2026-07-18" }, { "id": "p26-gemini-3-5-pro", "text": "Gemini 3.5 Pro launches by July 15, 2026, scoring AA Intelligence Index greater than or equal to 60, taking the closed-frontier intelligence lead from GPT-5.5 (currently 60.2).", "confidencePct": 60, "deadline": "By July 15, 2026", "deadlineDate": "2026-07-15", "lens": "software", "firstIssue": "2026-W21", "outcome": "miss", "overdue": false, "notes": "Gemini 3.5 Pro had not launched by July 15 — it missed the leaked July 17 target too, its third slip, with Bloomberg reporting July 16 the model was months behind schedule — so the AA Index >= 60 leg was never tested.", "evidenceUrl": "https://www.bloomberg.com/news/articles/2026-07-16/google-gemini-launch-delayed-as-tech-falls-short-of-internal-goals", "resolvedAt": "2026-07-18" }, { "id": "p59-tsmc-q2-capex-raise", "text": "TSMC's July 16 Q2 earnings raise or reiterate the top end of full-year 2026 capex guidance and report HPC/AI platform revenue up more than 50% year over year, confirming the packaging-constrained AI capex ramp.", "confidencePct": 62, "deadline": "By July 16, 2026", "deadlineDate": "2026-07-16", "lens": "hardware", "firstIssue": "2026-W28", "outcome": "partial", "overdue": false, "notes": "The capex leg hit decisively — TSMC raised the entire FY26 range to $60-64B (from $52-56B), well beyond reiterating the top end. The revenue leg fell just short: HPC at 66% of a $40.2B quarter implies ~47% YoY platform growth versus the >50% bar. The thesis behind the prediction (packaging-constrained ramp) was confirmed on the record by the CEO's packaging warning (quoted in the hardware lens)." }, { "id": "p63-hbm-soldout-2027", "text": "SK hynix's July 29 Q2 earnings call discloses that 2027 HBM capacity is substantially sold out or committed under long-term agreements, extending the memory-scarcity trade into a second year.", "confidencePct": 68, "deadline": "By July 29, 2026", "deadlineDate": "2026-07-29", "lens": "hardware", "firstIssue": "2026-W29", "outcome": "partial", "overdue": false, "notes": "LTA leg hit: SK hynix's Jul 29 Q2 release and call confirmed Long-Term Agreements with around 10 key customers and HBM4 mass shipments underway. Sold-out leg missed: management said 2027 HBM volumes and prices were still under negotiation — no primary disclosure that 2027 capacity is substantially sold out or committed.", "evidenceUrl": "https://news.skhynix.com/en/q2-2026-business-results/", "resolvedAt": "2026-08-01" }, { "id": "p4-interconnect-outpaces", "text": "At least one major colocation or interconnect operator reports cross-connect or interconnect revenue growth outpacing compute capacity revenue growth for two consecutive quarters.", "confidencePct": 70, "deadline": "By July 31, 2026", "deadlineDate": "2026-07-31", "lens": "networking", "firstIssue": "2026-W17", "outcome": "partial", "overdue": false, "notes": "Equinix Q1 Apr 29: Fabric revenue +26% YoY, Fabric bookings +70% YoY, raised guidance — first of two quarters needed. Score moves to hit if Q2 confirms the same outpacing, expected early August." }, { "id": "p6-custom-silicon-35", "text": "Custom silicon (TPU + Trainium + Maia + MTIA + Granite Rapids AI) reaches 35% of incremental AI compute share, up from ~30% today.", "confidencePct": 55, "deadline": "By July 31, 2026", "deadlineDate": "2026-07-31", "lens": "hardware", "firstIssue": "2026-W17", "outcome": "miss", "overdue": false, "notes": "No TrendForce / SemiAnalysis quarterly mix breakdown showed custom silicon at 35% of incremental AI compute share by Jul 31. The best public TrendForce figure kept ASIC-based AI servers at a projected 27.8% of 2026 shipments (GPUs 69.7%) — ~7pp short even on stock share, with no verified incremental-share print crossing 35%.", "evidenceUrl": "https://www.trendforce.com/presscenter/news/20260120-12887.html", "resolvedAt": "2026-08-01" }, { "id": "p7-bedrock-gpt55-anchor", "text": "AWS reports a customer-disclosed Bedrock-resident GPT-5.5 multi-year contract greater than $500M annualized, with the customer named publicly.", "confidencePct": 65, "deadline": "By July 31, 2026", "deadlineDate": "2026-07-31", "lens": "software", "firstIssue": "2026-W18", "outcome": "miss", "overdue": false, "notes": "Bedrock GA for GPT-5.5 landed (AWS Jun 2026 blog; Amgen and Autodesk named as evaluating), but no customer-disclosed Bedrock-resident GPT-5.5 multi-year contract greater than $500M annualized with a named customer appeared by Jul 31.", "evidenceUrl": "https://aws.amazon.com/blogs/machine-learning/openai-models-and-codex-on-amazon-bedrock-are-now-generally-available/", "resolvedAt": "2026-08-01" }, { "id": "p12-voice-frontier-procurement", "text": "At least one Fortune 500 enterprise discloses replacement of an existing voice / IVR platform with a frontier voice model (GPT-Realtime-2, Inworld TTS-2, or competitor) on a contract greater than $25M annualized.", "confidencePct": 65, "deadline": "By July 31, 2026", "deadlineDate": "2026-07-31", "lens": "software", "firstIssue": "2026-W19", "outcome": "miss", "overdue": false, "notes": "Named-trigger customers (Zillow, Priceline, Deutsche Telekom) were cited building on GPT-Realtime-2, but none disclosed replacement of an existing voice/IVR platform on a contract greater than $25M annualized by Jul 31 — no qualifying Fortune 500 procurement disclosure surfaced.", "evidenceUrl": "https://openai.com/index/advancing-voice-intelligence-with-new-models-in-the-api/", "resolvedAt": "2026-08-01" }, { "id": "p16-nvidia-circular-flag", "text": "At least one major equity research firm or audit-grade publication explicitly frames NVIDIA's 2026 customer-equity stakes (CoreWeave + IREN, plus any subsequent counterparty) as a circular-financing risk in a written report or audit qualifier.", "confidencePct": 70, "deadline": "By July 31, 2026", "deadlineDate": "2026-07-31", "lens": "capital", "firstIssue": "2026-W19", "outcome": "hit", "overdue": false, "notes": "JPMorgan equity research on IREN (May 11) kept an Underweight rating while flagging the NVIDIA partnership's \"circular nature\" and undefined GPU access as tempering the deal — a major sell-side written report explicitly framing an NVIDIA 2026 customer-equity counterparty as circular-financing risk, which is what the trigger asked for. Recorded with a caveat: the JPMorgan note itself is subscriber-only, so the citation is contemporaneous coverage quoting it rather than the primary document. Holding out for the PDF would make sell-side predictions effectively unresolvable, but the weaker evidence chain is noted.", "evidenceUrl": "https://247wallst.com/investing/2026/05/11/jpmorgan-hikes-iren-price-target-to-46-but-stays-underweight-on-circular-nvidia-deal-concerns/", "resolvedAt": "2026-08-01" }, { "id": "p17-samsung-walkout-shipments", "text": "Samsung HBM4 shipments to NVIDIA Vera Rubin drop by more than 25% in May vs April due to the May 21-June 7 walkout, with at least one downstream hyperscaler publicly delaying a Q3 capacity ramp.", "confidencePct": 55, "deadline": "By July 31, 2026", "deadlineDate": "2026-07-31", "lens": "hardware", "firstIssue": "2026-W20", "outcome": "partial", "overdue": false, "notes": "Samsung-union tentative agreement May 20 averted the 18-day walkout; ratification vote May 27-28. HBM4 shipments likely NOT to drop >25%; this prediction is on track to MISS. Confidence revised down. Score moves to miss if the ratification passes; score moves to hit only if the ratification fails and the strike actually proceeds." }, { "id": "p38-fable5-access-restored", "text": "Anthropic restores public (non-government) access to a Fable-5-class model — with geo-gating or KYC controls — by July 31, 2026.", "confidencePct": 60, "deadline": "By July 31, 2026", "deadlineDate": "2026-07-31", "lens": "software", "firstIssue": "2026-W24", "outcome": "partial", "overdue": false, "notes": "Anthropic's Jun 30 / Jul 1 redeployment post confirmed public Fable 5 access restored globally across the Claude Platform, Claude.ai, Claude Code, and Claude Cowork after export controls lifted, well inside the Jul 31 window. Scored partial rather than hit because the prediction specified restoration \"with geo-gating or KYC controls\" and the actual restoration was global with safety classifiers and no geographic or identity gating. The timing and direction were right; the predicted access-control mechanism did not materialize.", "evidenceUrl": "https://www.anthropic.com/news/redeploying-fable-5", "resolvedAt": "2026-08-01" }, { "id": "p49-micron-hbm4-booked", "text": "By July 31, 2026, at least one additional memory supplier besides Micron publicly confirms 2026 HBM4 supply is fully allocated or materially price-up for 2027.", "confidencePct": 72, "deadline": "By July 31, 2026", "deadlineDate": "2026-07-31", "lens": "hardware", "firstIssue": "2026-W26", "outcome": "hit", "overdue": false, "notes": "Hit. Samsung's CFO stated at the Apr 30 call that 2026 HBM4 output is fully sold out, and in-window reporting says SK hynix removed price caps from long-term agreements — explicit 2027 price-up posture. Caveat: the in-window items are supply-chain reporting; SK hynix's late-July earnings call would make it unambiguous." }, { "id": "p53-skhy-debut-validates-memory", "text": "SK hynix's Nasdaq ADS offering prices at or above its indicated ~$166/ADS level and closes its first trading week above the offer price, by July 31, 2026.", "confidencePct": 67, "deadline": "By July 31, 2026", "deadlineDate": "2026-07-31", "lens": "capital", "firstIssue": "2026-W27", "outcome": "partial", "overdue": false, "notes": "Partial. The pricing clause missed — the offering priced at $149/ADS, below the ~$166 indication — but the market clause is on track: demand was ~7x available shares and day one closed up ~13% at ~$168, well above offer. The memory thesis was validated by the demand, not the price; regular trading (SKHY) begins Jul 13." }, { "id": "p57-gemini-3-5-pro-ga-jul31", "text": "Gemini 3.5 Pro reaches public general availability — a callable API model ID with published pricing — by July 31, 2026, after slipping past its June window and the reported July 17 target.", "confidencePct": 58, "deadline": "By July 31, 2026", "deadlineDate": "2026-07-31", "lens": "software", "firstIssue": "2026-W28", "outcome": "miss", "overdue": false, "notes": "Google's Jul 21 post still described Gemini 3.5 Pro as \"testing with partners\" while shipping 3.6 Flash / 3.5 Flash-Lite / Flash Cyber instead — no public GA gemini-3.5-pro model ID with published pricing appeared through Jul 31.", "evidenceUrl": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-6-flash-3-5-flash-lite-3-5-flash-cyber/", "resolvedAt": "2026-08-01" }, { "id": "p62-deepseek-v4-ga-jul31", "text": "DeepSeek retires its legacy deepseek-chat and deepseek-reasoner aliases on July 24 as scheduled and ships DeepSeek V4 to official GA by July 31, 2026, with peak-hour surge pricing in effect.", "confidencePct": 76, "deadline": "By July 31, 2026", "deadlineDate": "2026-07-31", "lens": "software", "firstIssue": "2026-W29", "outcome": "partial", "overdue": false, "notes": "The alias-retirement leg hit at 15:59 UTC Jul 24 with hard failure and no redirect. Explicit V4 Pro/Flash IDs and thinking parameter support the migration thesis, but the full GA/pricing trigger is not yet documented strongly enough for a hit." }, { "id": "p61-kimi-k3-weights-aug10", "text": "Moonshot publishes Kimi K3 open weights on Hugging Face with a license permitting commercial self-hosting by August 10, 2026 (vendor-committed 'by July 27', with slippage buffer).", "confidencePct": 72, "deadline": "By August 10, 2026", "deadlineDate": "2026-08-10", "lens": "software", "firstIssue": "2026-W29", "outcome": "hit", "overdue": false, "notes": "Moonshot published Kimi K3 weights on Hugging Face on Jul 27 with license text, and the model was reclassified from closed to open weights on that drop. Scored a hit rather than a partial: the Kimi K3 License is MIT-like for internal commercial use, which is what \"commercial self-hosting\" asked for, and the revenue gate it does impose applies to MaaS operators reselling the model above $20M trailing-twelve-month revenue rather than to self-hosters. Resolved two days ahead of the Aug 10 deadline because the trigger fired unambiguously on Jul 27 and no later evidence could reverse it.", "evidenceUrl": "https://huggingface.co/moonshotai/Kimi-K3", "resolvedAt": "2026-08-08" }, { "id": "p21-ebo-msa-spec", "text": "Expanded Beam Optical MSA publishes a v1.0 spec within 90 days of launch (May 12), with at least one in-production deployment announced by a hyperscaler member (AMD, Cisco, Meta, Oracle).", "confidencePct": 60, "deadline": "By August 12, 2026", "deadlineDate": "2026-08-12", "lens": "networking", "firstIssue": "2026-W20", "outcome": "partial", "overdue": false, "notes": "EBO MSA membership expanded 17 to 23 vendors May 18 (HPE marquee addition, Bellwether, JPC Connectivity, Mixx, TIME, TFC). v1.0 spec not yet published. Member growth is positive signal but spec + in-production deployment still pending. On track." }, { "id": "p8-capex-revenue-ratio", "text": "The top-4 hyperscaler capex / AI revenue ratio drops below 5.0 for Q2 2026, confirming the W18 directional flip.", "confidencePct": 60, "deadline": "By August 15, 2026", "deadlineDate": "2026-08-15", "lens": "capital", "firstIssue": "2026-W18", "outcome": "hit", "overdue": false, "notes": "Q2 2026 prints put every top-4 hyperscaler capex/AI-revenue ratio below 5.0x on published run-rate math: Meta ~1.66x, Microsoft Azure ~3.64x, Google Cloud ~3.84x, AWS ~4.25x — confirming the sub-5.0 directional flip the trigger asked for.", "evidenceUrl": "https://io-fund.com/ai-stocks/big-tech-ai-revenue-surges-suppliers-win-bigger", "resolvedAt": "2026-09-01" }, { "id": "p18-claude-code-churn", "text": "Anthropic discloses or third-party developer surveys document a greater than 20% drop in active Claude Code third-party agent users within 30 days of the June 15 pricing cutover.", "confidencePct": 65, "deadline": "By August 15, 2026", "deadlineDate": "2026-08-15", "lens": "software", "firstIssue": "2026-W20", "outcome": "miss", "overdue": false, "notes": "Anthropic paused the planned June 15 Agent SDK credit split on the scheduled date, so the pricing cutover never took effect and no >20% Claude Code third-party user drop was disclosed or surveyed within 30 days of a change that did not happen.", "evidenceUrl": "https://thenewstack.io/anthropic-pauses-claude-agent-sdk-subscription-change/", "resolvedAt": "2026-09-01" }, { "id": "p64-colo-interconnect-outpaces", "text": "The largest colocation operators' Q2 prints (starting July 29) show interconnect/fabric revenue growth again outpacing overall revenue growth, sustaining the Metcalfe read for a second consecutive quarter.", "confidencePct": 71, "deadline": "By August 15, 2026", "deadlineDate": "2026-08-15", "lens": "networking", "firstIssue": "2026-W29", "outcome": "miss", "overdue": false, "notes": "Equinix Q2 FY26 — the largest operator's Jul 29 print — grew interconnection revenue 11.3% YoY versus total revenue up 16.4%; interconnect did not outpace overall growth, and no other major operator's Q2 filing showed fabric/interconnect beating total revenue either.", "evidenceUrl": "https://www.sec.gov/Archives/edgar/data/1101239/000110123926000145/a991eqix-q226xpr.htm", "resolvedAt": "2026-09-01" }, { "id": "p67-opus5-aa-gap-aug15", "text": "Artificial Analysis publishes an Opus 5 Intelligence Index result within three points of Claude Fable 5 by August 15, 2026.", "confidencePct": 74, "deadline": "By August 15, 2026", "deadlineDate": "2026-08-15", "lens": "software", "firstIssue": "2026-W30", "outcome": "hit", "overdue": false, "notes": "Artificial Analysis v4.1.1 (Aug 6) scored Claude Opus 5 (max) at 63 and Claude Fable 5 at 62 — a one-point gap, inside the three-point bar — before the Aug 15 deadline.", "evidenceUrl": "https://artificialanalysis.ai/models/comparisons/claude-opus-5-xhigh-vs-claude-fable-5", "resolvedAt": "2026-09-01" }, { "id": "p46-ferc-rto-compliance", "text": "At least one RTO/ISO files a large-load interconnection compliance proposal answering FERC's Jun 18 show-cause orders by the August 17, 2026 deadline.", "confidencePct": 80, "deadline": "By August 17, 2026", "deadlineDate": "2026-08-17", "lens": "power", "firstIssue": "2026-W25", "outcome": "miss", "overdue": false, "notes": "All six RTOs/ISOs sought ~90-day abeyances instead of filing substantive large-load interconnection compliance proposals by the Aug 17 show-cause deadline; July 20 informational reports do not satisfy the trigger's compliance-proposal bar.", "evidenceUrl": "https://mgrid.org/2026/08/13/all-six-grid-operators-ask-ferc-for-90-more-days-on-large-load-rules-pushing-filings-to-november-16/", "resolvedAt": "2026-09-01" }, { "id": "p20-cisco-ai-orders-tracking", "text": "Cisco's Q4 FY26 AI orders confirm the $9B annual run rate (Q4 AI orders at least $2.5B), validating the enterprise AI fabric demand-pull narrative.", "confidencePct": 70, "deadline": "By August 20, 2026", "deadlineDate": "2026-08-20", "lens": "capital", "firstIssue": "2026-W20", "outcome": "hit", "overdue": false, "notes": "Cisco's Aug 13 Q4 FY26 release reported $4.0B of hyperscaler AI infrastructure orders in the quarter — above the $2.5B Q4 bar — and $9.3B for the full fiscal year, validating the ~$9B annual run-rate thesis.", "evidenceUrl": "https://investor.cisco.com/news/news-details/2026/CISCO-REPORTS-FOURTH-QUARTER-AND-FISCAL-YEAR-2026-EARNINGS/default.aspx", "resolvedAt": "2026-09-01" }, { "id": "p10-fabric-second-confirm", "text": "At least one major colocation operator other than Equinix reports Q2 2026 interconnect / fabric revenue growth greater than 25% YoY.", "confidencePct": 65, "deadline": "By August 31, 2026", "deadlineDate": "2026-08-31", "lens": "networking", "firstIssue": "2026-W18", "outcome": "miss", "overdue": false, "notes": "No major colocation operator other than Equinix reported Q2 2026 interconnect revenue growth above 25% YoY: CoreSite grew total revenue 7.7% without a >25% interconnect line, and Csquare's interconnect revenue fell 10.3% YoY to $24.7M.", "evidenceUrl": "https://www.sec.gov/Archives/edgar/data/2105398/000210539826000002/csqr-ex99_1.htm", "resolvedAt": "2026-09-01" }, { "id": "p13-mrc-second-fabric", "text": "At least one major hyperscaler other than Microsoft publicly discloses a production AI fabric running MRC at greater than 50,000-GPU scale.", "confidencePct": 70, "deadline": "By August 31, 2026", "deadlineDate": "2026-08-31", "lens": "networking", "firstIssue": "2026-W19", "outcome": "miss", "overdue": false, "notes": "Public MRC production deployments at >50k-GPU scale are attributed to Microsoft/OpenAI sites (Fairwater, Abilene); no Google, Amazon, or Meta hyperscaler other than Microsoft disclosed a production AI fabric running MRC above the 50,000-GPU threshold by Aug 31.", "evidenceUrl": "https://openai.com/index/mrc-supercomputer-networking/", "resolvedAt": "2026-09-01" }, { "id": "p22-nvidia-q2-fy27", "text": "NVIDIA Q2 FY27 (August 2026) prints Data Center revenue at $90B-plus, confirming the $91B Q2 guide is conservative and supply-constraint thesis holds.", "confidencePct": 70, "deadline": "By August 31, 2026", "deadlineDate": "2026-08-31", "lens": "hardware", "firstIssue": "2026-W21", "outcome": "miss", "overdue": false, "notes": "NVIDIA's Aug 26 Q2 FY27 release reported Data Center revenue of $89.0B — $1.0B below the $90B-plus trigger despite beating the $91B guide midpoint on consolidated revenue.", "evidenceUrl": "https://www.sec.gov/Archives/edgar/data/1045810/000104581026000073/q2fy27cfocommentary.htm", "resolvedAt": "2026-09-01" }, { "id": "p24-agent-platform-lock-in", "text": "At least one Fortune 500 enterprise discloses standardization on a single agent-platform runtime (Claude Code / Codex / Antigravity 2.0 / Grok Build) for greater than $50M annualized commitment.", "confidencePct": 60, "deadline": "By August 31, 2026", "deadlineDate": "2026-08-31", "lens": "software", "firstIssue": "2026-W21", "outcome": "miss", "overdue": false, "notes": "No Fortune 500 enterprise disclosed standardizing on a single agent-platform runtime (Claude Code / Codex / Antigravity 2.0 / Grok Build) with a named >$50M annualized commitment; Anthropic's $100M partner-network spend is vendor-side, not a customer lock-in disclosure.", "evidenceUrl": "https://www.anthropic.com/news/claude-partner-network", "resolvedAt": "2026-09-01" }, { "id": "p25-fabric-vendor-q2", "text": "At least three of (Marvell, Broadcom, NVIDIA networking, Arista) post Q2 networking-segment revenue growth greater than 40% YoY, validating EBO MSA momentum into Q2 prints.", "confidencePct": 70, "deadline": "By August 31, 2026", "deadlineDate": "2026-08-31", "lens": "networking", "firstIssue": "2026-W21", "outcome": "miss", "overdue": false, "notes": "None of the four vendors disclosed a standalone networking-segment print above 40% YoY in Q2: Arista total revenue grew 37.7%, Marvell reports data-center (not networking-only) growth, Broadcom cites AI networking as ~40% of AI semiconductors without a >40% networking-segment line, and NVIDIA does not break out networking YoY separately.", "evidenceUrl": "https://investors.arista.com/Communications/Press-Releases-and-Events/Press-Release-Detail/2026/Arista-Networks-Inc--Reports-Second-Quarter-2026-Financial-Results/default.aspx", "resolvedAt": "2026-09-01" }, { "id": "p27-anthropic-s1-public", "text": "No frontier lab (Anthropic or OpenAI) files a publicly visible S-1 on SEC EDGAR before August 31, 2026, keeping the IPO race at the confidential-DRS stage.", "confidencePct": 65, "deadline": "By August 31, 2026", "deadlineDate": "2026-08-31", "lens": "capital", "firstIssue": "2026-W22", "outcome": "hit", "overdue": false, "notes": "Neither Anthropic nor OpenAI converted its confidential draft to a publicly visible S-1 on SEC EDGAR before Aug 31, 2026 — only the Jun 1 confidential submission and DRS-stage filings remained non-public through the deadline.", "evidenceUrl": "https://decodethefuture.org/en/anthropic-s1-ipo-filing-explained/", "resolvedAt": "2026-09-01" }, { "id": "p30-optics-design-wins", "text": "At least two of (Credo, Marvell, Broadcom) cite co-packaged-optics or 1.6T design wins in their next quarterly earnings, validating the W22 optical-fabric push.", "confidencePct": 65, "deadline": "By August 31, 2026", "deadlineDate": "2026-08-31", "lens": "networking", "firstIssue": "2026-W22", "outcome": "partial", "overdue": false, "notes": "Marvell's Aug 31 Q2 FY27 call cited secured 1.6T optical-DSP design wins and a rapidly ramping 1.6T business (one leg). Credo said on its Q2 call it sees no significant customer movement toward co-packaged optics, and Broadcom's Q2 print discussed CPO programs without a second qualifying 1.6T/CPO design-win citation — one of two required vendors.", "evidenceUrl": "https://www.fool.com/earnings/call-transcripts/2026/08/31/marvell-mrvl-q2-2027-earnings-call-transcript/", "resolvedAt": "2026-09-01" }, { "id": "p31-sovereign-power-followthrough", "text": "A major hyperscaler or sovereign program announces a new behind-the-meter or >1GW power-procurement deal (SMR, gas, or grid) by August 31, 2026, as time-to-power stays the binding US constraint.", "confidencePct": 60, "deadline": "By August 31, 2026", "deadlineDate": "2026-08-31", "lens": "power", "firstIssue": "2026-W22", "outcome": "hit", "overdue": false, "notes": "Amazon confirmed in August that it acquired the GW Ranch site in Pecos County, Texas, for an AI campus powered by Pacifico Energy's planned 7.65GW behind-the-meter gas-and-storage system — a named hyperscaler >1GW BTM generation deal inside the Aug 31 window.", "evidenceUrl": "https://www.datacenterknowledge.com/energy-power-supply/amazon-explores-grid-connection-for-8-000-acre-ai-campus", "resolvedAt": "2026-09-01" }, { "id": "p34-hbm4-allocation-tightness", "text": "Before August 31, 2026, at least one memory supplier or supply-chain analyst reports HBM4 allocation tightness despite three-supplier qualification.", "confidencePct": 65, "deadline": "By August 31, 2026", "deadlineDate": "2026-08-31", "lens": "hardware", "firstIssue": "2026-W23", "outcome": "hit", "overdue": false, "notes": "Before Aug 31, supply-chain and trade press documented HBM4 allocation tightness despite three-supplier qualification: Seoul Economic Daily (Aug 31) reported spot HBM prices at 4–5x LTA levels with customers not receiving full contracted volumes even under LTAs.", "evidenceUrl": "https://en.sedaily.com/finance/2026/08/31/spot-hbm-prices-hit-5-times-contract-levels", "resolvedAt": "2026-09-01" }, { "id": "p35-cpo-design-win", "text": "Broadcom, Marvell, or NVIDIA announces a new CPO/1.6T production design win or revenue guide uplift tied to AI networking before August 31, 2026.", "confidencePct": 65, "deadline": "By August 31, 2026", "deadlineDate": "2026-08-31", "lens": "networking", "firstIssue": "2026-W23", "outcome": "partial", "overdue": false, "notes": "Arista's 1.6T 7060XE7 portfolio on Broadcom's Tomahawk 6 (Jun 9) is a fresh Broadcom 1.6T production design win, satisfying the 1.6T leg; no co-packaged-optics production win or vendor revenue-guide uplift yet. Tracking to a full hit by deadline." }, { "id": "p39-second-inference-asic-milestone", "text": "At least one more merchant inference-ASIC vendor (Groq, Cerebras, Tenstorrent, or d-Matrix) announces a volume-production or named-hyperscaler milestone before August 31, 2026.", "confidencePct": 65, "deadline": "By August 31, 2026", "deadlineDate": "2026-08-31", "lens": "hardware", "firstIssue": "2026-W24", "outcome": "hit", "overdue": false, "notes": "NVIDIA announced Aug 24 that Groq 3 LPX — the merchant inference ASIC line acquired from Groq — is in full production, with Nebius planning production deployment on Nebius Token Factory later in 2026, satisfying the second-vendor volume-production milestone bar.", "evidenceUrl": "https://investor.nvidia.com/news/press-release-details/2026/NVIDIA-Groq-3-LPX-Now-in-Full-Production-With-World-Class-Speed-for-Agentic-AI/default.aspx", "resolvedAt": "2026-09-01" }, { "id": "p40-second-1-6t-design-win", "text": "A second of (Marvell, Credo) cites a 1.6T or co-packaged-optics production design win by August 31, 2026, joining Broadcom's Tomahawk 6.", "confidencePct": 60, "deadline": "By August 31, 2026", "deadlineDate": "2026-08-31", "lens": "networking", "firstIssue": "2026-W24", "outcome": "hit", "overdue": false, "notes": "Marvell's Aug 31 Q2 FY27 earnings call cited secured design wins and a rapidly ramping 1.6T optical-DSP business, joining Broadcom's Tomahawk 6 as the second named vendor with a production-scale 1.6T win disclosure before Aug 31.", "evidenceUrl": "https://www.fool.com/earnings/call-transcripts/2026/08/31/marvell-mrvl-q2-2027-earnings-call-transcript/", "resolvedAt": "2026-09-01" }, { "id": "p41-no-public-s1", "text": "Neither OpenAI nor Anthropic converts its confidential draft to a publicly visible S-1 on SEC EDGAR before August 31, 2026.", "confidencePct": 62, "deadline": "By August 31, 2026", "deadlineDate": "2026-08-31", "lens": "capital", "firstIssue": "2026-W24", "outcome": "hit", "overdue": false, "notes": "SEC EDGAR had no publicly visible OpenAI or Anthropic S-1 through Aug 31, 2026; both remained at the confidential-DRS stage after their June confidential submissions.", "evidenceUrl": "https://decodethefuture.org/en/anthropic-s1-ipo-filing-explained/", "resolvedAt": "2026-09-01" }, { "id": "p43-open-weight-top5", "text": "An MIT- or Apache-licensed open-weight model (e.g., GLM-5.2) enters the overall top 5 of the Artificial Analysis Intelligence Index v4.1 — not just the open-weight subset — by August 31, 2026.", "confidencePct": 58, "deadline": "By August 31, 2026", "deadlineDate": "2026-08-31", "lens": "software", "firstIssue": "2026-W25", "outcome": "miss", "overdue": false, "notes": "AA Intelligence Index v4.1.1's overall top five through Aug 31 were all Anthropic closed variants (Fable 5.1 and Opus 5 scores 62–66); the highest MIT/Apache-licensed open-weight model, Kimi K3 at 60, sat outside the overall top five, and GLM-5.3 uses a bespoke license rather than MIT/Apache.", "evidenceUrl": "https://artificialanalysis.ai/models", "resolvedAt": "2026-09-01" }, { "id": "p44-hbm4-allocation-2027", "text": "By August 31, 2026, all three HBM makers (SK hynix, Samsung, Micron) confirm HBM fully allocated for 2026 and/or 2027 price increases.", "confidencePct": 75, "deadline": "By August 31, 2026", "deadlineDate": "2026-08-31", "lens": "hardware", "firstIssue": "2026-W25", "outcome": "partial", "overdue": false, "notes": "Samsung's Jul 30 call and industry reporting described 2027 memory/HBM capacity largely committed with rising contract prices, but SK hynix's Jul 29 Q2 release still said 2027 HBM volumes and prices were under negotiation — not all three primary makers confirmed fully allocated 2027 HBM on their own disclosures.", "evidenceUrl": "https://news.skhynix.com/en/q2-2026-business-results/", "resolvedAt": "2026-09-01" }, { "id": "p47-hyperscaler-1gw-btm", "text": "A named hyperscaler announces a >1GW behind-the-meter or off-grid generation deal for AI data centers by August 31, 2026.", "confidencePct": 68, "deadline": "By August 31, 2026", "deadlineDate": "2026-08-31", "lens": "power", "firstIssue": "2026-W25", "outcome": "hit", "overdue": false, "notes": "Amazon confirmed Aug 7+ that it acquired GW Ranch in Texas and will buy power from Pacifico Energy's planned 7.65GW behind-the-meter generation complex for an on-site AI data center — a named hyperscaler >1GW off-grid/BTM deal by Aug 31.", "evidenceUrl": "https://www.datacenterknowledge.com/energy-power-supply/amazon-explores-grid-connection-for-8-000-acre-ai-campus", "resolvedAt": "2026-09-01" }, { "id": "p48-closed-price-response", "text": "At least one major closed lab cuts flagship API prices or ships a cheaper tier by August 31, 2026, in response to open-weight cost pressure.", "confidencePct": 55, "deadline": "By August 31, 2026", "deadlineDate": "2026-08-31", "lens": "capital", "firstIssue": "2026-W25", "outcome": "hit", "overdue": false, "notes": "OpenAI cut flagship GPT-5.6 Sol API list prices >20% on Aug 21 (to $4/$20 per MTok from $5/$30), explicitly framing the move against Anthropic and Chinese-model competition — a major closed-lab flagship API price cut inside the deadline.", "evidenceUrl": "https://www.reuters.com/technology/openai-cuts-developer-pricing-frontier-gpt-56-sol-model-by-more-than-20-2026-08-21/", "resolvedAt": "2026-09-01" }, { "id": "p51-cpo-partner-rack", "text": "A named Vera Rubin partner announces a CPO/Spectrum-X Ethernet Photonics rack or cluster design win by August 31, 2026.", "confidencePct": 63, "deadline": "By August 31, 2026", "deadlineDate": "2026-08-31", "lens": "networking", "firstIssue": "2026-W26", "outcome": "miss", "overdue": false, "notes": "Supermicro announced CPO-capable Vera Rubin rack blueprints in June and NVIDIA named Spectrum-X Photonics adopters, but no named Vera Rubin partner publicly announced a CPO/Spectrum-X Ethernet Photonics rack or cluster design win — product/integration announcements fall short of the trigger's design-win bar.", "evidenceUrl": "https://investor.nvidia.com/news/press-release-details/2026/NVIDIA-Vera-Rubin-Ramps-Into-Full-Production-to-Power-Agentic-AI-Factories-Worldwide/default.aspx", "resolvedAt": "2026-09-01" }, { "id": "p52-agent-automation-governance", "text": "At least one major enterprise platform ships an admin control specifically for scheduled/background coding or app-building agents by August 31, 2026.", "confidencePct": 66, "deadline": "By August 31, 2026", "deadlineDate": "2026-08-31", "lens": "software", "firstIssue": "2026-W26", "outcome": "hit", "overdue": false, "notes": "Hit. GitHub shipped Copilot agent session streaming to public preview (Jul 2) — SIEM/Purview streaming of all agent sessions — on top of its agent control plane, and GitHub also added AI-credit session limits covering background agents (Jul 1, per Agent Techniques coverage)." }, { "id": "p55-terra-confirms-repricing-cycle", "text": "GPT-5.6 reaches broad GA with the Terra tier priced at or below $2.50/$15 per MTok — half of GPT-5.5's rate — confirming a closed-lab repricing cycle rather than a one-off Sonnet 5 cut, by August 31, 2026.", "confidencePct": 62, "deadline": "By August 31, 2026", "deadlineDate": "2026-08-31", "lens": "software", "firstIssue": "2026-W27", "outcome": "hit", "overdue": false, "notes": "Hit, seven weeks early. GPT-5.6 went GA Jul 9 with Terra at exactly $2.50/$15 per MTok. Grok 4.5's $2/$6 launch the day before makes it a three-vendor repricing cycle (Sonnet 5, Terra, Grok 4.5), not a one-off." }, { "id": "p56-samsung-hbm4-to-nvidia", "text": "Samsung's HBM4 supply to NVIDIA is publicly confirmed — via earnings call, company statement, or multi-source supply-chain reporting — by August 31, 2026.", "confidencePct": 66, "deadline": "By August 31, 2026", "deadlineDate": "2026-08-31", "lens": "hardware", "firstIssue": "2026-W27", "outcome": "hit", "overdue": false, "notes": "Hit on the multi-source-reporting trigger: Korean press (Seoul Economic Daily, Korea Herald) reported alongside Samsung's record Q2 guidance that HBM4 — in mass production since February for NVIDIA's Vera Rubin — reached $1B in sales within four months. Caveat: Samsung's Jul 30 divisional results would make it unambiguous from the company itself." }, { "id": "p58-harness-cost-telemetry", "text": "At least one major agent platform (OpenAI, Anthropic, GitHub, or Cursor) ships product-level per-task or per-harness cost telemetry or routing controls — beyond session budget caps — by August 31, 2026.", "confidencePct": 64, "deadline": "By August 31, 2026", "deadlineDate": "2026-08-31", "lens": "software", "firstIssue": "2026-W28", "outcome": "hit", "overdue": false, "notes": "Cursor shipped Cursor Router in July 2026 with Auto Balance/Intelligence routing controls and published measured cost-per-commit figures ($4.63–$6.76) from live traffic — product-level harness routing and cost telemetry beyond session budget caps.", "evidenceUrl": "https://cursor.com/blog/router", "resolvedAt": "2026-09-01" }, { "id": "p66-no-cheap-floor-reset", "text": "DeepSeek V4's official GA pricing does not reset the ultra-cheap floor: off-peak deepseek-v4-pro output pricing stays at or above ¥6 (~$0.85) per MTok through August 31, 2026 — the kill-condition test for this issue's price-band-convergence claim.", "confidencePct": 84, "deadline": "By August 31, 2026", "deadlineDate": "2026-08-31", "lens": "software", "firstIssue": "2026-W29", "outcome": "hit", "overdue": false, "notes": "DeepSeek's official API pricing page kept GA deepseek-v4-pro off-peak output at $1.98/MTok (~¥14+) through Aug 31 — well above the ¥6 (~$0.85)/MTok ultra-cheap floor the trigger set as the kill condition.", "evidenceUrl": "https://api-docs.deepseek.com/quick_start/pricing/", "resolvedAt": "2026-09-01" }, { "id": "p70-flash-task-cost-aug31", "text": "An independent benchmark finds Gemini 3.6 Flash at least 12% cheaper per completed agentic task than Gemini 3.5 Flash by August 31, 2026.", "confidencePct": 66, "deadline": "By August 31, 2026", "deadlineDate": "2026-08-31", "lens": "software", "firstIssue": "2026-W30", "outcome": "hit", "overdue": false, "notes": "Artificial Analysis measured Gemini 3.6 Flash at $0.50 average cost per completed agentic task versus $0.59 for 3.5 Flash — a 15% reduction, above the 12% cheaper-per-task bar — before Aug 31.", "evidenceUrl": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-6-flash-3-5-flash-lite-3-5-flash-cyber/", "resolvedAt": "2026-09-01" }, { "id": "p90-glm-53-weights-sep15", "text": "Z.ai publishes GLM-5.3 weights to Hugging Face by September 15, 2026, closing the two-week window promised at the model's August 14 announcement.", "confidencePct": 43, "deadline": "By September 15, 2026", "deadlineDate": "2026-09-15", "lens": "software", "firstIssue": "2026-W34", "outcome": "hit", "overdue": false, "notes": "Z.ai published the full 753B-parameter GLM-5.3 weights to Hugging Face at zai-org/GLM-5.3 on August 27–28, 2026 — in-window and inside the trigger's September 15 window, distinct from GLM-5.2 — after GLM-5.3-Flash MIT weights landed Aug 26. The material nuance is licensing, not availability: GLM-5.3 ships under a bespoke GLM-5.3 license rather than MIT, requiring Z.AI security review before commercial use by any Model-as-a-Service operator whose group revenue exceeds $10B over any 12 consecutive months." }, { "id": "p2-f500-on-prem", "text": "At least one Fortune 500 enterprise discloses an on-prem AI workload greater than $100M annual using open-weight models.", "confidencePct": 60, "deadline": "By September 30, 2026", "deadlineDate": "2026-09-30", "lens": "software", "firstIssue": "2026-W17", "outcome": "pending", "overdue": false }, { "id": "p5-neocloud-anchor-loss", "text": "At least one neocloud loses an anchor tenant or sees backlog growth turn negative quarter-over-quarter.", "confidencePct": 35, "deadline": "By September 30, 2026", "deadlineDate": "2026-09-30", "lens": "capital", "firstIssue": "2026-W17", "outcome": "pending", "overdue": false }, { "id": "p9-hbm4-second-source", "text": "Samsung HBM4 reaches greater than 25% share of NVIDIA Vera Rubin BOM by Q3 2026 supply data, ending the SK Hynix monopoly on the platform.", "confidencePct": 60, "deadline": "By September 30, 2026", "deadlineDate": "2026-09-30", "lens": "hardware", "firstIssue": "2026-W18", "outcome": "pending", "overdue": false }, { "id": "p11-gated-cyber-revenue", "text": "By Q3 2026, at least two frontier labs publicly disclose a separately-priced 'gated cyber' or 'gated security' SKU with revenue commentary, formalizing capability-gating as a product line.", "confidencePct": 60, "deadline": "By September 30, 2026", "deadlineDate": "2026-09-30", "lens": "software", "firstIssue": "2026-W18", "outcome": "pending", "overdue": false }, { "id": "p15-amd-second-frontier-train", "text": "A second public open-weights frontier-class model trained end-to-end on AMD silicon (MI300X or MI400) is announced, confirming AMD-as-training-substrate is structural rather than a one-off proof.", "confidencePct": 55, "deadline": "By September 30, 2026", "deadlineDate": "2026-09-30", "lens": "hardware", "firstIssue": "2026-W19", "outcome": "pending", "overdue": false }, { "id": "p19-specialist-multimodal-procurement", "text": "At least one Fortune 500 enterprise discloses a production deployment greater than $10M annualized of a specialist video / embodied multimodal model (Perceptron Mk1, MolmoAct 2 successors, or analogous) as a primary tier rather than a general LMM.", "confidencePct": 60, "deadline": "By September 30, 2026", "deadlineDate": "2026-09-30", "lens": "software", "firstIssue": "2026-W20", "outcome": "pending", "overdue": false }, { "id": "p33-vera-rubin-first-shipments", "text": "At least one major OEM announces customer shipment or formal order availability for Vera Rubin NVL72-class systems before September 30, 2026.", "confidencePct": 70, "deadline": "By September 30, 2026", "deadlineDate": "2026-09-30", "lens": "hardware", "firstIssue": "2026-W23", "outcome": "pending", "overdue": false }, { "id": "p36-power-first-followthrough", "text": "A hyperscaler announces another >500MW power-first AI campus or behind-the-meter generation deal by September 30, 2026.", "confidencePct": 60, "deadline": "By September 30, 2026", "deadlineDate": "2026-09-30", "lens": "power", "firstIssue": "2026-W23", "outcome": "pending", "overdue": false }, { "id": "p42-hyperscaler-power-first", "text": "A hyperscaler announces a >500MW power-first AI campus or behind-the-meter generation deal by September 30, 2026.", "confidencePct": 60, "deadline": "By September 30, 2026", "deadlineDate": "2026-09-30", "lens": "power", "firstIssue": "2026-W24", "outcome": "pending", "overdue": false }, { "id": "p45-second-1-6t-design-win", "text": "A second non-Broadcom vendor (Marvell or Credo) cites a 1.6T or co-packaged-optics production design win by September 30, 2026.", "confidencePct": 55, "deadline": "By September 30, 2026", "deadlineDate": "2026-09-30", "lens": "networking", "firstIssue": "2026-W25", "outcome": "pending", "overdue": false }, { "id": "p50-groq-enterprise-customer", "text": "Groq announces at least one named Fortune 500 or hyperscaler inference-cloud customer by September 30, 2026.", "confidencePct": 57, "deadline": "By September 30, 2026", "deadlineDate": "2026-09-30", "lens": "capital", "firstIssue": "2026-W26", "outcome": "pending", "overdue": false }, { "id": "p60-scale-across-follow-on", "text": "A second named vendor or operator announces a commercial cross-data-center scale-across AI fabric deployment or product launch — following DriveNets/WhiteFiber — by September 30, 2026.", "confidencePct": 61, "deadline": "By September 30, 2026", "deadlineDate": "2026-09-30", "lens": "networking", "firstIssue": "2026-W28", "outcome": "pending", "overdue": false }, { "id": "p71-luna-task-cost-sep30", "text": "An independent evaluator publishes completed-task cost showing GPT-5.6 Luna at least 60% cheaper per completed agentic task than GPT-5.6 Terra by September 30, 2026.", "confidencePct": 62, "deadline": "By September 30, 2026", "deadlineDate": "2026-09-30", "lens": "software", "firstIssue": "2026-W31", "outcome": "pending", "overdue": false }, { "id": "p83-deepseek-0813-weights-sep30", "text": "DeepSeek publishes the V4-Pro-0813 build weights to Hugging Face by September 30, 2026.", "confidencePct": 64, "deadline": "By September 30, 2026", "deadlineDate": "2026-09-30", "lens": "software", "firstIssue": "2026-W33", "outcome": "pending", "overdue": false }, { "id": "p96-openrouter-volume-sep30", "text": "The highest single ISO week of OpenRouter aggregate token volume in September 2026 exceeds the Ox Alpha stealth-week peak (week of August 20–26, 2026) by at least 15%, by September 30, 2026.", "confidencePct": 40, "deadline": "By September 30, 2026", "deadlineDate": "2026-09-30", "lens": "software", "firstIssue": "2026-W35", "outcome": "pending", "overdue": false }, { "id": "p3-capex-revise", "text": "Aggregate 2026 hyperscaler capex revises upward by 10% or more from the $700B baseline.", "confidencePct": 80, "deadline": "By October 31, 2026", "deadlineDate": "2026-10-31", "lens": "capital", "firstIssue": "2026-W17", "outcome": "partial", "overdue": false, "notes": "Q1 prints (MSFT $190B, GOOG $180-190B, META $125-145B, AMZN $200B reaffirmed) take 2026 aggregate to $695-725B (+77% YoY) vs the $700B W17 baseline. At/near baseline; +10% revision (~$770B) plausible by Q2 print. Score moves to hit if Q2 takes aggregate above $770B." }, { "id": "p65-state-moratorium-copycat", "text": "At least one additional US state announces a moratorium, discretionary-permit pause, or equivalent statewide restriction on large data-center development by October 31, 2026, following New York's EO 62.", "confidencePct": 57, "deadline": "By October 31, 2026", "deadlineDate": "2026-10-31", "lens": "power", "firstIssue": "2026-W29", "outcome": "pending", "overdue": false }, { "id": "p72-harness-memory-reproduction-oct31", "text": "An independent party reproduces at least a 15-point ARC-AGI-3 improvement from harness memory and compaction settings alone, holding model weights fixed, by October 31, 2026.", "confidencePct": 44, "deadline": "By October 31, 2026", "deadlineDate": "2026-10-31", "lens": "software", "firstIssue": "2026-W31", "outcome": "pending", "overdue": false }, { "id": "p73-hbm-2027-committed-oct31", "text": "SK hynix or Samsung states in a primary release or earnings transcript that 2027 HBM capacity is substantially committed or sold out by October 31, 2026.", "confidencePct": 81, "deadline": "By October 31, 2026", "deadlineDate": "2026-10-31", "lens": "hardware", "firstIssue": "2026-W31", "outcome": "pending", "overdue": false }, { "id": "p77-endpoint-index-expansion-oct31", "text": "Artificial Analysis publishes Endpoint Accuracy Index results covering at least two models beyond the initial GLM-5.2, gpt-oss-120b and DeepSeek V4 Pro set by October 31, 2026.", "confidencePct": 83, "deadline": "By October 31, 2026", "deadlineDate": "2026-10-31", "lens": "software", "firstIssue": "2026-W32", "outcome": "pending", "overdue": false }, { "id": "p88-nvidia-10q-guaranty-exhibits-oct31", "text": "NVIDIA files exhibits with the 10-Q for the quarter ended July 26, 2026 that translate the SB Energy PORTS-Pike residual-value guaranty into a per-quarter contingent-obligation disclosure and identify the OpenAI affiliate as tenant, by October 31, 2026.", "confidencePct": 72, "deadline": "By October 31, 2026", "deadlineDate": "2026-10-31", "lens": "hardware", "firstIssue": "2026-W34", "outcome": "hit", "overdue": false, "notes": "NVIDIA filed the Form 10-Q for the quarter ended July 26, 2026 on August 26, 2026 — inside the window. It satisfies all three trigger elements: guarantees 'capped at a total of $105 billion' with an exposure table of $3.5B AI-cloud guarantees plus $105.0B SB Energy for $108.5B total; effectiveness conditioned on SB Energy satisfying applicable ready-for-service conditions as each of nine phases is placed in service from fiscal 2029; and the tenant identified as 'an affiliate of OpenAI Group PBC' at the PORTS Technology Campus in Pike County, Ohio. Exhibit 10.1 is the Form of Residual Value Guaranty." }, { "id": "p91-anthropic-skills-case-study-oct31", "text": "Anthropic or a named enterprise customer publishes an independently attributed Skills API production case study with a stated manual baseline and a measured post-Skills outcome (time, cost, or completion), by October 31, 2026.", "confidencePct": 58, "deadline": "By October 31, 2026", "deadlineDate": "2026-10-31", "lens": "software", "firstIssue": "2026-W34", "outcome": "pending", "overdue": false }, { "id": "p95-jalapeno-agentx-oct31", "text": "SemiAnalysis publishes AgentX v3 multi-turn benchmark results for OpenAI Jalapeño on production-representative agentic traces, with methodology comparable to Vera Rubin NVL72 AgentX runs cited by NVIDIA, by October 31, 2026.", "confidencePct": 36, "deadline": "By October 31, 2026", "deadlineDate": "2026-10-31", "lens": "hardware", "firstIssue": "2026-W35", "outcome": "pending", "overdue": false }, { "id": "p99-metaroce-ocp-spec-oct31", "text": "Meta contributes MetaRoCE specification through OCP at the October 2026 Global Summit with documented production deployment targets beyond the 64-node AMD proof-of-concept, by October 31, 2026.", "confidencePct": 44, "deadline": "By October 31, 2026", "deadlineDate": "2026-10-31", "lens": "networking", "firstIssue": "2026-W35", "outcome": "pending", "overdue": false }, { "id": "p74-interconnect-revenue-q3-nov15", "text": "No publicly listed global colocation operator reports Q3 2026 interconnection revenue growing faster than total revenue on a normalized basis by November 15, 2026.", "confidencePct": 46, "deadline": "By November 15, 2026", "deadlineDate": "2026-11-15", "lens": "networking", "firstIssue": "2026-W31", "outcome": "pending", "overdue": false }, { "id": "p89-marvell-google-first-tranche-nov30", "text": "Marvell discloses the first attributable Custom Products revenue tranche under the Google warrant agreement in a filed earnings release or subsequent 10-Q, by November 30, 2026.", "confidencePct": 34, "deadline": "By November 30, 2026", "deadlineDate": "2026-11-30", "lens": "hardware", "firstIssue": "2026-W34", "outcome": "pending", "overdue": false }, { "id": "p97-bis-remote-gpu-nprm-nov30", "text": "Commerce BIS publishes a Federal Register notice of proposed rulemaking on remote access to advanced US AI compute by Chinese end users, by November 30, 2026.", "confidencePct": 27, "deadline": "By November 30, 2026", "deadlineDate": "2026-11-30", "lens": "power", "firstIssue": "2026-W35", "outcome": "pending", "overdue": false }, { "id": "p98-nvidia-rubin-mix-q3-earnings", "text": "NVIDIA Q3 FY2027 earnings disclosure states Vera Rubin contributed more than 25% of datacenter revenue for the quarter ended October 26, 2026.", "confidencePct": 74, "deadline": "By NVIDIA Q3 FY2027 earnings release (expected November 2026)", "deadlineDate": "2026-11-30", "lens": "hardware", "firstIssue": "2026-W35", "outcome": "pending", "overdue": false }, { "id": "p100-astra-neutral-memory-dec15", "text": "ARC Prize or another provider-neutral evaluator publishes an Astra run without provider-private reasoning state that closes at least half of the 35.9-point matched-effort Standard-to-Provider-Adapter gap by December 15, 2026.", "confidencePct": 32, "deadline": "By December 15, 2026", "deadlineDate": "2026-12-15", "lens": "software", "firstIssue": "2026-W36", "outcome": "pending", "overdue": false }, { "id": "p75-vendor-backstop-documented-dec31", "text": "A definitive agreement of at least $100B in vendor-guaranteed AI data-center financing is publicly documented in a filing or company release by December 31, 2026.", "confidencePct": 34, "deadline": "By December 31, 2026", "deadlineDate": "2026-12-31", "lens": "capital", "firstIssue": "2026-W31", "outcome": "pending", "overdue": false }, { "id": "p76-paducah-psc-filing-dec31", "text": "A power service agreement for the Paducah AI campus is filed with the Kentucky Public Service Commission by December 31, 2026.", "confidencePct": 63, "deadline": "By December 31, 2026", "deadlineDate": "2026-12-31", "lens": "power", "firstIssue": "2026-W31", "outcome": "pending", "overdue": false }, { "id": "p78-astra-critical-final-dec31", "text": "OpenAI publicly assigns its Astra model a final Preparedness Framework cybersecurity rating of Critical by December 31, 2026.", "confidencePct": 24, "deadline": "By December 31, 2026", "deadlineDate": "2026-12-31", "lens": "software", "firstIssue": "2026-W32", "outcome": "pending", "overdue": false }, { "id": "p82-discovery-loop-amount-dec31", "text": "Alphabet discloses the size of its investment in Discovery Loop in an SEC filing or official release by December 31, 2026.", "confidencePct": 27, "deadline": "By December 31, 2026", "deadlineDate": "2026-12-31", "lens": "power", "firstIssue": "2026-W32", "outcome": "pending", "overdue": false }, { "id": "p84-second-neocloud-tenor-gap-dec31", "text": "A second publicly traded neocloud discloses, in an SEC or equivalent filing, GPU-backed debt whose maturity extends beyond the stated weighted-average or characteristic tenor of the customer contracts securing it, by December 31, 2026.", "confidencePct": 69, "deadline": "By December 31, 2026", "deadlineDate": "2026-12-31", "lens": "capital", "firstIssue": "2026-W33", "outcome": "pending", "overdue": false }, { "id": "p86-ocp-silicon-photonics-spec-dec31", "text": "The OCP Open Silicon Photonics for AI Systems workstream submits its first specification by December 31, 2026, meeting the Q4 2026 target stated at launch.", "confidencePct": 61, "deadline": "By December 31, 2026", "deadlineDate": "2026-12-31", "lens": "networking", "firstIssue": "2026-W33", "outcome": "pending", "overdue": false }, { "id": "p92-second-state-dc-grid-rule-dec31", "text": "A second US state with major hyperscale exposure (Virginia, Texas, Georgia, Oregon, or Illinois) publishes a permitting or interconnection rule that adds a binding compliance stage above a stated MW threshold, comparable in kind to Pennsylvania Executive Order 2026-05, by December 31, 2026.", "confidencePct": 39, "deadline": "By December 31, 2026", "deadlineDate": "2026-12-31", "lens": "power", "firstIssue": "2026-W34", "outcome": "pending", "overdue": false }, { "id": "p102-second-queue-contract-dec31", "text": "A second AI infrastructure contract above $500M discloses both a future service date and an option, delayed-draw, earnout, or guarantee allocating schedule risk by December 31, 2026.", "confidencePct": 68, "deadline": "By December 31, 2026", "deadlineDate": "2026-12-31", "lens": "capital", "firstIssue": "2026-W36", "outcome": "pending", "overdue": false }, { "id": "p103-fabric-one-control-schema-dec31", "text": "Equinix publishes Fabric One beta documentation that exposes approval, rollback, or auditable intent-history controls before December 31, 2026.", "confidencePct": 57, "deadline": "By December 31, 2026", "deadlineDate": "2026-12-31", "lens": "networking", "firstIssue": "2026-W36", "outcome": "pending", "overdue": false }, { "id": "p104-broadcom-ai-split-dec31", "text": "Broadcom discloses separate quarterly revenue figures for custom AI accelerators and AI networking by December 31, 2026.", "confidencePct": 24, "deadline": "By December 31, 2026", "deadlineDate": "2026-12-31", "lens": "hardware", "firstIssue": "2026-W36", "outcome": "pending", "overdue": false }, { "id": "p69-camellia-curtailment-template-jan31", "text": "A second US multi-gigawatt AI campus publicly commits to at least 250 MW of utility-directed peak curtailment by January 31, 2027.", "confidencePct": 49, "deadline": "By January 31, 2027", "deadlineDate": "2027-01-31", "lens": "power", "firstIssue": "2026-W30", "outcome": "pending", "overdue": false }, { "id": "p79-gateway-endpoint-fidelity-jan31", "text": "A major model-serving platform or AI gateway publishes per-endpoint accuracy, precision, or output-token-limit disclosures for the open-weight models it serves by January 31, 2027.", "confidencePct": 31, "deadline": "By January 31, 2027", "deadlineDate": "2027-01-31", "lens": "networking", "firstIssue": "2026-W32", "outcome": "pending", "overdue": false }, { "id": "p81-eval-vendor-isolation-policy-jan31", "text": "At least two frontier labs publish network isolation or containment requirements for third-party cyber evaluation partners by January 31, 2027.", "confidencePct": 44, "deadline": "By January 31, 2027", "deadlineDate": "2027-01-31", "lens": "capital", "firstIssue": "2026-W32", "outcome": "pending", "overdue": false }, { "id": "p87-turn-count-in-model-card-jan31", "text": "A frontier lab publishes turn count or task-completion cost as a headline metric alongside benchmark scores in an official model card or launch post by January 31, 2027.", "confidencePct": 26, "deadline": "By January 31, 2027", "deadlineDate": "2027-01-31", "lens": "software", "firstIssue": "2026-W33", "outcome": "pending", "overdue": false }, { "id": "p85-nvidia-rubin-ultra-memory-confirm-mar31", "text": "NVIDIA publicly confirms a Rubin Ultra memory configuration at or below 512GB, or an 8-high HBM4E stack option, in official specifications or an earnings disclosure by March 31, 2027.", "confidencePct": 37, "deadline": "By March 31, 2027", "deadlineDate": "2027-03-31", "lens": "hardware", "firstIssue": "2026-W33", "outcome": "pending", "overdue": false }, { "id": "p94-second-vendor-cpo-shipping-mar31", "text": "A second major networking or accelerator vendor (Broadcom, Marvell, Cisco, or Arista) publicly discloses production shipments of co-packaged-optics Ethernet or scale-up switches to a named hyperscaler by March 31, 2027.", "confidencePct": 47, "deadline": "By March 31, 2027", "deadlineDate": "2027-03-31", "lens": "networking", "firstIssue": "2026-W34", "outcome": "pending", "overdue": false }, { "id": "p68-helios-production-rack-q2-2027", "text": "At least one named customer reports receiving a production AMD Helios rack for workload qualification by June 30, 2027.", "confidencePct": 68, "deadline": "By June 30, 2027", "deadlineDate": "2027-06-30", "lens": "hardware", "firstIssue": "2026-W30", "outcome": "pending", "overdue": false }, { "id": "p80-amd-taalas-roadmap-jun30", "text": "AMD publicly names a Taalas-derived product or roadmap item tied to a specific model or model class by June 30, 2027.", "confidencePct": 46, "deadline": "By June 30, 2027", "deadlineDate": "2027-06-30", "lens": "hardware", "firstIssue": "2026-W32", "outcome": "pending", "overdue": false }, { "id": "p93-anthropic-in-house-silicon-tapeout-jun30", "text": "Anthropic publicly confirms a tape-out or first-silicon milestone on the in-house accelerator program being led by Amir Salek, by June 30, 2027.", "confidencePct": 22, "deadline": "By June 30, 2027", "deadlineDate": "2027-06-30", "lens": "hardware", "firstIssue": "2026-W34", "outcome": "pending", "overdue": false }, { "id": "p101-fervo-expansion-firm-jun30", "text": "Google accepts at least 500 MW of Fervo's conditional expansion and the parties execute a definitive agreement by June 30, 2027.", "confidencePct": 41, "deadline": "By June 30, 2027", "deadlineDate": "2027-06-30", "lens": "power", "firstIssue": "2026-W36", "outcome": "pending", "overdue": false } ] }