{ "schema": "brianletort-radar-v1", "asOf": "2026-09-07", "audience": "Operators, allocators, and anyone building in AI.", "disclaimer": "The Radar publishes multi-year theses on AI infrastructure and economics from public data only. It is commentary and a scored forecasting exercise, not investment, legal, or engineering advice. Theses about the author's employer, its named competitors, and its customer relationships are excluded. Named companies appear on observable fundamentals and financing events; any thesis that references a security price carries the market-call notice below.", "scoreboardRule": "The Radar keeps its own scoreboard. It publishes no aggregate Brier score until 30 tier-1 or tier-2 items have resolved, and it never averages with the Ledger.", "counts": { "total": 33, "visible": 30, "drafts": 3, "registered": 30, "resolved": 0, "gatedHidden": 3, "marketCalls": 1, "byTier": { "T1": 10, "T2": 14, "T3": 6 }, "byTag": { "CALL": 11, "WARNING": 12, "THESIS": 7 } }, "arenas": [ { "id": "compute-supply", "title": "Compute supply and custom silicon", "gated": false, "open": true, "theses": [ "RT-01", "RT-04", "RT-03", "RT-02" ] }, { "id": "power-and-sites", "title": "Power and sites", "gated": true, "open": false, "theses": [] }, { "id": "networking-fabric", "title": "Networking fabric", "gated": false, "open": true, "theses": [ "RT-10", "RT-08", "RT-09" ] }, { "id": "model-frontier", "title": "Model frontier and open weights", "gated": false, "open": true, "theses": [ "RT-12", "RT-11", "RT-13" ] }, { "id": "agi-capabilities", "title": "AGI and capabilities", "gated": false, "open": true, "theses": [ "RT-16", "RT-15", "RT-14", "RT-33", "RT-17" ] }, { "id": "enterprise-absorption", "title": "Enterprise absorption", "gated": false, "open": true, "theses": [ "RT-21", "RT-20", "RT-19", "RT-18" ] }, { "id": "consumer-commercial", "title": "Consumer and commercial use", "gated": false, "open": true, "theses": [ "RT-23", "RT-22", "RT-24" ] }, { "id": "capital-credit", "title": "Capital and credit", "gated": false, "open": true, "theses": [ "RT-27", "RT-26", "RT-28", "RT-25" ] }, { "id": "geopolitics-policy", "title": "Geopolitics and policy", "gated": false, "open": true, "theses": [ "RT-30", "RT-32", "RT-31", "RT-29" ] } ], "scoreboard": { "t1": { "resolved": 0, "pending": 11, "brier": null, "logLoss": null, "hitRate": null }, "t2": { "rungsResolved": 0, "rungsPending": 63, "rungBrier": null, "rungHitRate": null }, "t3": { "theses": 7, "falsified": 0, "survivalRate": 1 }, "aggregateGate": { "required": 30, "resolved": 0, "open": false }, "humanVsEngine": { "meanAbsDelta": 0.1673, "engineCoolerCount": 23, "engineHotterCount": 10, "overrideAccuracy": null, "n": 33 }, "unamendedSurvivalRate": null }, "engine": { "runId": "radar-2026-09-07T23-49-04Z-de8abf", "asOf": "2026-09-07", "engineVersion": "0.1.0", "dataSources": [ { "id": "artificial-analysis", "kind": "artificial-analysis-api", "asOf": "2026-09-07", "records": 0, "url": "https://artificialanalysis.ai/leaderboards/models", "status": "unavailable", "note": "ARTIFICIAL_ANALYSIS_API_KEY not set; the data API answers 401 and the leaderboard is JS-rendered. Enter readings under 'artificial-analysis:' in data/manual-series.json with a public source; ARTIFICIAL_ANALYSIS_API_KEY not set; API is key-only, site is JS-rendered" }, { "id": "bls", "kind": "bls-v1", "asOf": "2026-09-07", "records": 75, "url": "https://api.bls.gov/publicAPI/v1/timeseries/data/", "status": "ok", "note": null }, { "id": "eia", "kind": "eia-v2", "asOf": "2026-09-07", "records": 30, "url": "https://api.eia.gov/v2/", "status": "partial", "note": "EIA_API_KEY not set; only retail-sales (EPM Table 5.1, US) and steo/ work without a key; no EIA_API_KEY; open EPM/STEO workbooks only" }, { "id": "epoch", "kind": "epoch-csv", "asOf": "2026-09-07", "records": 10, "url": "https://epoch.ai/data/notable_ai_models.csv, https://epoch.ai/data/large_scale_ai_models.csv, https://epoch.ai/data/ml_hardware.csv, https://epoch.ai/data/gpu_clusters.csv, https://epoch.ai/data/benchmarks.csv", "status": "partial", "note": "1 seriesId(s) unresolved (see unresolvedSeries)" }, { "id": "fred", "kind": "fred-csv", "asOf": "2026-09-07", "records": 4498, "url": "https://fred.stlouisfed.org/graph/fredgraph.csv?id=", "status": "ok", "note": null }, { "id": "hf-hub", "kind": "huggingface-hub", "asOf": "2026-09-07", "records": 2, "url": "https://huggingface.co/api/models?author=&sort=createdAt&direction=-1&limit=1000&expand[]=createdAt&expand[]=downloads", "status": "ok", "note": null }, { "id": "manual", "kind": "manual-file", "asOf": "2026-09-07", "records": 0, "url": null, "status": "unavailable", "note": "33 seriesId(s) unresolved (see unresolvedSeries)" }, { "id": "metaculus", "kind": "metaculus-api", "asOf": "2026-09-07", "records": 0, "url": "https://www.metaculus.com/api/posts//", "status": "unavailable", "note": "METACULUS_TOKEN not set; the Metaculus API rejects anonymous requests (HTTP 403); METACULUS_TOKEN not set; API is authenticated-only" }, { "id": "sec-xbrl", "kind": "sec-xbrl-companyconcept", "asOf": "2026-09-07", "records": 313, "url": "https://data.sec.gov/api/xbrl/companyconcept/CIK##########/us-gaap/.json", "status": "ok", "note": null } ] }, "theses": [ { "id": "RT-01", "title": "Broadcom AI silicon reaches a quarter of NVIDIA datacenter", "arena": "compute-supply", "tier": "T1", "tag": "CALL", "direction": "up", "status": "registered", "statement": "Broadcom's reported quarterly AI semiconductor revenue reaches at least 25% of NVIDIA's Data Center segment revenue for the nearest-ending NVIDIA fiscal quarter in at least one quarter reported on or before 2027-12-31.", "resolutionRule": "For each Broadcom fiscal quarter with an earnings release dated on or before 2027-12-31, take the \"AI semiconductor revenue\" figure as stated in Broadcom's earnings press release or CFO commentary (USD, GAAP). Pair it with the NVIDIA fiscal quarter whose period-end date is closest to the Broadcom period-end date (Broadcom quarters end early Feb/May/Aug/Nov; NVIDIA quarters end late Jan/Apr/Jul/Oct) and take NVIDIA's \"Data Center\" revenue from its earnings press release. Resolves HIT if Broadcom AI / NVIDIA Data Center >= 0.25 in any such pair. If either company stops reporting the named line item before a qualifying pair exists, resolves MISS. Baseline as of 2026-09-04: Broadcom Q3 FY2026 AI revenue $16.7B vs NVIDIA Q2 FY2027 Data Center $89.0B = 18.8%.", "resolutionSource": { "name": "Broadcom and NVIDIA quarterly earnings press releases (investor relations)", "url": "https://investors.broadcom.com/" }, "ambiguity": "resolves-against-forecaster", "horizon": "2027-12-31", "p": 0.5, "ci80": [ 0.35, 0.65 ], "consensus": { "value": "Company guidance implies roughly 19-22%: Broadcom guided FY2027 AI revenue of ~$115B (~$29B/quarter) while NVIDIA guided fiscal 2028 revenue growth of ~70% (Data Center ~$150B+/quarter by late 2027).", "impliedP": 0.2, "impliedBy": "Broadcom guidance of ~$29B/quarter against NVIDIA Data Center of ~$130-150B/quarter through 2027 puts the ratio at 19-22% in every quarter. Reaching 25% in any single quarter needs Broadcom to beat its own path by ~20-30% while NVIDIA merely meets guidance; the guidance path gives that roughly one chance in five.", "source": "Broadcom Q3 FY2026 earnings call (2026-09-04) and NVIDIA Q2 FY2027 earnings call (2026-08-26)", "url": "https://www.fool.com/earnings/call-transcripts/2026/08/26/nvidia-nvda-q2-2027-earnings-call-transcript/", "asOf": "2026-09-04", "note": "No prediction-market question exists on this ratio. My 0.50 rests on Broadcom beating its AI guide in six consecutive quarters and NVIDIA calling itself supply-constrained; the Q4 FY2026 guide already implies ~22%." }, "leadingIndicators": [ { "id": "li-1", "label": "Broadcom AI semiconductor revenue, quarterly", "connector": "manual", "seriesId": "avgo-ai-semiconductor-revenue-quarterly", "unit": "USD billions", "onTrack": { "op": ">=", "value": 24 }, "offTrack": { "op": "<", "value": 20 }, "url": "https://investors.broadcom.com/", "note": "Dollar figure labelled \"AI semiconductor revenue\" in Broadcom's quarterly earnings press release. Thresholds apply to the quarter reported nearest 2027-03-31 (Q1 FY2027). Q3 FY2026 actual: $16.7B; Q4 FY2026 guide: $21.7B." }, { "id": "li-2", "label": "NVIDIA Data Center revenue, quarterly", "connector": "sec-xbrl", "seriesId": "0001045810:Revenues:USD", "unit": "USD billions", "onTrack": { "op": "<=", "value": 110 }, "offTrack": { "op": ">", "value": 130 }, "url": "https://data.sec.gov/api/xbrl/companyconcept/CIK0001045810/us-gaap/Revenues.json", "note": "Total quarterly revenue from XBRL company-concept endpoint as a proxy for the denominator (Data Center is ~93% of total). Thresholds refer to the quarter ending 2027-01. Faster NVIDIA growth makes the ratio harder to reach." } ], "milestones": [ { "id": "m1", "date": "2026-11-30", "statement": "Marvell discloses the first attributable Custom Products revenue tranche under the Google relationship.", "p": 0.55, "liveQuestionId": "lq-p89" }, { "id": "m2", "date": "2027-06-30", "statement": "Broadcom reports a quarter with AI semiconductor revenue of at least $28B.", "p": 0.55 } ], "falsifier": "Broadcom AI revenue stays below 22% of NVIDIA Data Center revenue in every paired quarter through the Q3 FY2027 pair reported in September 2027.", "whyItMatters": "The custom-silicon share of accelerator spend decides how much of the AI capex cycle accrues to one vendor. If Broadcom closes to a quarter of NVIDIA's data center line, hyperscaler in-house chips are a second supply chain, not a hedge. Power and cooling design in the buildings then splits across two architectures.", "whatWouldRaise": [ "Broadcom raises FY2027 AI guidance above $130B or names a fourth XPU customer at volume.", "NVIDIA Data Center revenue grows below 10% quarter over quarter for two consecutive quarters.", "Google or Meta disclose TPU or MTIA capex as a majority of accelerator spend." ], "whatWouldCut": [ "NVIDIA Data Center revenue exceeds $120B in a quarter before mid-2027.", "Broadcom cuts or defers FY2027 AI revenue guidance.", "A hyperscaler cancels or delays a named custom ASIC generation." ], "namedEntities": [ "Broadcom", "NVIDIA" ], "marketCall": false, "recusalCheck": { "passed": true, "keysChecked": [ "employer-leasing", "named-competitor-bookings", "customer-pipeline", "ticker" ] }, "signedBy": "Brian Letort", "registeredAt": "2026-09-08", "amendments": [], "ruleHash": "3a47df941561867636614bae53815467b73c6c6f22dec6d256d1b1b8d74a6501", "distance": 1.3862943611198906, "distanceLabel": "clearly above consensus", "engine": { "id": "RT-01", "prior": { "p": 0.02, "method": "sec-xbrl-loglinear-bootstrap:li-2", "note": "log-linear growth +0.597/yr on the last 16 points, extrapolated from 2026-07-26; P(li-2 satisfies onTrack <= 110.0 at 2027-12-31); proxy for the thesis, not its rule" }, "council": { "p": 0.3325264590832613, "votes": [ { "persona": "grid-and-power-planner", "model": "openai/gpt-5.6-sol", "p": 0.31, "rationale": "The supplied guidance puts the central ratio near 19–22%, with the closest expected quarter around 22%. Reaching 25% therefore requires roughly 14% relative upside versus that pair—Broadcom materially outperforming while NVIDIA does not. Broadcom’s repeated AI-guide beats and potentially lumpy custom-silicon ramps create a credible transient-hit path, but NVIDIA’s rapid, supply-constrained growth raises the denominator and both firms’ demand surprises are correlated. I use five major-ramp quarterly forecast-error opportunities as the reference class, discounted heavily for serial correlation rather than treated independently. The 0.02 structural prior is poorly specified because it extrapolates only NVIDIA revenue and lacks the numerator." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.38, "rationale": "Five eligible pairs remain. The current ratio is 0.188, and supplied guidance points to roughly 0.19–0.22, so a hit needs at least a 14% relative overshoot even from the top of that range. My reference class is major custom-accelerator ramps: customer qualification and advanced-packaging or HBM constraints commonly shift volume by one or more quarters, while guidance beats are correlated rather than five independent chances. Broadcom’s repeated beats and lumpy hyperscaler ramps create a meaningful upside tail; an NVIDIA supply pause could also compress the denominator. Conversely, NVIDIA’s denominator is co-growing rapidly, and line-item discontinuation adds a small miss risk." }, { "persona": "enterprise-cio", "model": "openai/gpt-5.6-sol", "p": 0.34, "rationale": "Using supplier revenue-share thresholds during overlapping hyperscaler ramps as the reference class, I apply a roughly 0.27 annual crossing hazard over the remaining 1.32 years, implying about 0.34 after compounding. The ratio starts at 0.188 and supplied guidance centers subsequent quarters near 0.19–0.22, so a hit requires Broadcom to outperform NVIDIA by at least 14% relative even from 0.22. Five quarterly opportunities are highly correlated, not independent trials; NVIDIA’s supplied series also shows rapid denominator growth. Broadcom’s repeated guide beats and potentially lumpy ramps provide upside, but a $28B quarter alone may not clear the ratio. Reporting-continuity and exact pairing add downside." }, { "persona": "credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.37, "rationale": "The central guidance path remains a MISS: the ratio is 0.188 currently and roughly 0.19–0.22 through FY2027. Even from 0.22, qualifying requires about 0.14 relative upside from Broadcom outperformance, slower NVIDIA growth, or favorable quarterly lumpiness. The approximately five remaining observations are highly correlated, so they should not be treated as independent chances. My reference class is rapidly growing semiconductor businesses versus near-term guidance: a greater-than-0.10 relative beat can occur, but sustained denominator growth makes a one-quarter crossover less than even. Six prior Broadcom beats and custom-silicon lumpiness support meaningful upside. The structural prior is poorly specified because it extrapolates NVIDIA revenue alone, not the ratio. Reporting-line discontinuation adds modest downside." }, { "persona": "geopolitics-policy-analyst", "model": "openai/gpt-5.6-sol", "p": 0.29, "rationale": "The baseline ratio is 0.188, so the threshold requires a 33% relative increase. The guidance path—about $29B versus at least $150B by late 2027—remains near 0.19, while Q4 implies roughly 0.22. A hit therefore needs a sizable relative Broadcom beat, NVIDIA slowdown, or quarterly lumpiness. I use a five-quarter reference class of relative-revenue races between high-growth semiconductor segments, discounting repeated opportunities because their demand trends are correlated. Broadcom’s beat history provides upside, but NVIDIA’s supply constraints support denominator growth. I give little weight to the 0.02 structural prior because it models NVIDIA alone. Reporting-line discontinuation adds slight downside." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.31, "rationale": "Base rate: I use 0.25 for a high-growth semiconductor supplier starting near a 0.19 revenue ratio to reach 0.25 within five quarterly comparisons. The current ratio is 0.188, while supplied FY2027 guidance centers near 0.19–0.22, so a HIT requires a material Broadcom beat, NVIDIA shortfall, or favorable quarterly lumpiness—not merely rapid growth by both. Broadcom’s six guide beats and Q4 indication near 0.22 raise the estimate, and multiple quarters provide several chances. NVIDIA’s indicated growth keeps the denominator difficult. I largely discount the 0.02 structural prior because it extrapolates NVIDIA revenue without Broadcom’s numerator. Finally, loss of either named disclosure automatically causes MISS." } ], "aggregation": "log-odds-mean", "extremization": 1, "dissent": null, "spread": 0.09, "warnings": [ "all 6 votes came from one vendor (openai); extremization disabled, persona diversity only" ] }, "final": { "p": 0.09159610843952803, "ci80": [ 0.0780385621113882, 0.1890561048620667 ], "method": "stacked-fixed-weights (radar: prior 0.5, council 0.5; no resolved radar items to learn from)", "weights": { "council": 0.5, "prior": 0.5 } }, "stated": 0.5, "history": [ { "asOf": "2026-09-07", "p": 0.4282860601584378, "source": "radar-engine", "runId": "radar-2026-09-07T23-12-09Z-06a58f" }, { "asOf": "2026-09-07", "p": 0.09159610843952803, "source": "radar-engine", "runId": "radar-2026-09-07T23-49-04Z-de8abf" } ], "indicatorStatus": "on-track", "milestones": [ { "id": "m1", "enginePrior": 0.23476783233214954, "councilP": 0.02290460225354313 }, { "id": "m2", "enginePrior": 0.5, "councilP": 0.5151840155877333 } ], "arena": "compute-supply", "status": "draft", "tier": "T1" }, "indicators": [ { "thesisId": "RT-01", "indicatorId": "li-1", "connector": "manual", "seriesId": "avgo-ai-semiconductor-revenue-quarterly", "unit": "USD billions", "asOf": "2026-09-07", "current": null, "status": "no-data", "history": [], "url": "https://investors.broadcom.com/", "note": "manual: 'avgo-ai-semiconductor-revenue-quarterly' not in manual-series.json" }, { "thesisId": "RT-01", "indicatorId": "li-2", "connector": "sec-xbrl", "seriesId": "0001045810:Revenues:USD", "unit": "USD billions", "asOf": "2026-07-26", "current": 96.221, "status": "on-track", "history": [ { "asOf": "2009-10-25", "value": 0.9032060000000001 }, { "asOf": "2010-01-31", "value": 0.982488 }, { "asOf": "2010-05-02", "value": 1.001813 }, { "asOf": "2010-07-31", "value": 0.811208 }, { "asOf": "2010-10-31", "value": 0.8439120000000001 }, { "asOf": "2011-01-30", "value": 0.886376 }, { "asOf": "2011-05-01", "value": 0.9620390000000001 }, { "asOf": "2011-07-31", "value": 1.0165170000000001 }, { "asOf": "2011-10-30", "value": 1.0661800000000001 }, { "asOf": "2012-01-29", "value": 0.9531940000000001 }, { "asOf": "2012-04-29", "value": 0.9248770000000001 }, { "asOf": "2012-07-29", "value": 1.04427 }, { "asOf": "2012-10-28", "value": 1.20411 }, { "asOf": "2013-01-27", "value": 1.106902 }, { "asOf": "2013-04-28", "value": 0.954739 }, { "asOf": "2013-07-28", "value": 0.977238 }, { "asOf": "2013-10-27", "value": 1.053967 }, { "asOf": "2014-01-26", "value": 1.1442180000000002 }, { "asOf": "2014-04-27", "value": 1.103 }, { "asOf": "2014-07-27", "value": 1.103 }, { "asOf": "2014-10-26", "value": 1.225 }, { "asOf": "2015-01-25", "value": 1.2510000000000001 }, { "asOf": "2015-04-26", "value": 1.151 }, { "asOf": "2015-07-26", "value": 1.153 }, { "asOf": "2015-10-25", "value": 1.3050000000000002 }, { "asOf": "2016-01-31", "value": 1.401 }, { "asOf": "2016-05-01", "value": 1.3050000000000002 }, { "asOf": "2016-07-31", "value": 1.4280000000000002 }, { "asOf": "2016-10-30", "value": 2.004 }, { "asOf": "2017-01-29", "value": 2.173 }, { "asOf": "2017-04-30", "value": 1.937 }, { "asOf": "2017-07-30", "value": 2.23 }, { "asOf": "2017-10-29", "value": 2.636 }, { "asOf": "2018-01-28", "value": 2.911 }, { "asOf": "2018-04-29", "value": 3.2070000000000003 }, { "asOf": "2018-07-29", "value": 3.123 }, { "asOf": "2018-10-28", "value": 3.181 }, { "asOf": "2019-04-28", "value": 2.22 }, { "asOf": "2019-07-28", "value": 2.579 }, { "asOf": "2019-10-27", "value": 3.0140000000000002 }, { "asOf": "2020-04-26", "value": 3.08 }, { "asOf": "2020-07-26", "value": 3.866 }, { "asOf": "2020-10-25", "value": 4.726 }, { "asOf": "2021-05-02", "value": 5.6610000000000005 }, { "asOf": "2021-08-01", "value": 6.507000000000001 }, { "asOf": "2021-10-31", "value": 7.103000000000001 }, { "asOf": "2022-05-01", "value": 8.288 }, { "asOf": "2022-07-31", "value": 6.704000000000001 }, { "asOf": "2022-10-30", "value": 5.931 }, { "asOf": "2023-04-30", "value": 7.192 }, { "asOf": "2023-07-30", "value": 13.507000000000001 }, { "asOf": "2023-10-29", "value": 18.12 }, { "asOf": "2024-04-28", "value": 26.044 }, { "asOf": "2024-07-28", "value": 30.040000000000003 }, { "asOf": "2024-10-27", "value": 35.082 }, { "asOf": "2025-04-27", "value": 44.062000000000005 }, { "asOf": "2025-07-27", "value": 46.743 }, { "asOf": "2025-10-26", "value": 57.006 }, { "asOf": "2026-04-26", "value": 81.61500000000001 }, { "asOf": "2026-07-26", "value": 96.221 } ], "url": "https://data.sec.gov/api/xbrl/companyconcept/CIK0001045810/us-gaap/Revenues.json", "note": "quarterly frames; raw USD facts scaled by 1e-09 to match unit 'USD billions'" } ], "indicatorStatus": "on-track", "chainEvents": [ { "seq": 198, "ts": "2026-09-07", "kind": "repriced", "hash": "187555a3001005013822e10e870331a25aa05dee7e99426764ce4a8bf7db8e6e" }, { "seq": 199, "ts": "2026-09-07", "kind": "repriced", "hash": "c88bed10ea3e154fd54effdc9079befec4ae81eb7d803abec94a001e6a325712" }, { "seq": 258, "ts": "2026-09-08", "kind": "registered", "hash": "41e5e5bfcf96dcb2be0d4a9f687975b82781df82138f56c0139cca6b08334465" } ], "url": "/predictions/compute-supply/RT-01" }, { "id": "RT-02", "title": "NVIDIA posts a down datacenter quarter before 2028", "arena": "compute-supply", "tier": "T1", "tag": "WARNING", "direction": "down", "status": "registered", "statement": "NVIDIA reports at least one fiscal quarter, in an earnings release dated on or before 2027-12-31, in which Data Center segment revenue is lower than the immediately preceding fiscal quarter.", "resolutionRule": "Use the \"Data Center\" revenue figure (USD, GAAP) as stated in each NVIDIA quarterly earnings press release from Q3 FY2027 (reported November 2026) through the last release dated on or before 2027-12-31 (expected Q3 FY2028, November 2027). Resolves HIT if any quarter's Data Center revenue is strictly below the prior quarter's as-reported figure (use the figure as originally reported for the prior quarter, not later restatements). Resolves MISS if every quarter grows sequentially or if NVIDIA stops reporting a Data Center segment. Baseline: Q2 FY2027 Data Center $89.0B, up 18% quarter over quarter.", "resolutionSource": { "name": "NVIDIA quarterly earnings press releases (investor relations)", "url": "https://investor.nvidia.com/financial-info/quarterly-results/default.aspx" }, "ambiguity": "resolves-against-forecaster", "horizon": "2027-12-31", "p": 0.25, "ci80": [ 0.15, 0.4 ], "consensus": { "value": "NVIDIA guided Q3 FY2027 revenue of $108B (+12% q/q) and a supply-constrained fiscal 2028 growth target of about 70%; management stated demand exceeds supply through fiscal 2028, implying no sequential decline.", "impliedP": 0.1, "impliedBy": "A 70% fiscal-2028 growth target under stated supply constraint implies every quarter grows sequentially; a down quarter needs either a guide miss of more than ~12% or a product-transition air pocket. NVIDIA has had one such quarter (Q1 FY2023 gaming, not Data Center) in the last five years; five chances at roughly 2% each gives about one in ten.", "source": "NVIDIA Q2 FY2027 earnings call, 2026-08-26", "url": "https://www.fool.com/earnings/call-transcripts/2026/08/26/nvidia-nvda-q2-2027-earnings-call-transcript/", "asOf": "2026-08-26", "note": "No public market prices a sequential decline; Kalshi contracts cover GPU rental prices, not NVIDIA revenue. My 0.25 weights the Rubin transition and hyperscaler digestion higher than the guidance path does." }, "leadingIndicators": [ { "id": "li-1", "label": "NVIDIA total quarterly revenue (XBRL)", "connector": "sec-xbrl", "seriesId": "0001045810:Revenues:USD", "unit": "USD billions", "onTrack": { "op": "<", "value": 105 }, "offTrack": { "op": ">=", "value": 115 }, "url": "https://data.sec.gov/api/xbrl/companyconcept/CIK0001045810/us-gaap/Revenues.json", "note": "Quarterly (10-Q/10-K duration facts) revenue; thresholds are for the quarter ending 2026-10 against the $108B guide. A miss versus guide is the first sign of the digestion this thesis expects." }, { "id": "li-2", "label": "NVIDIA accounts receivable and inventory growth", "connector": "sec-xbrl", "seriesId": "0001045810:InventoryNet:USD", "unit": "USD billions", "onTrack": { "op": ">=", "value": 30 }, "offTrack": { "op": "<", "value": 20 }, "url": "https://data.sec.gov/api/xbrl/companyconcept/CIK0001045810/us-gaap/InventoryNet.json", "note": "Balance-sheet inventory from XBRL. Inventory building faster than revenue is the classic precursor of a sequential dip in a supply-constrained cycle." } ], "milestones": [ { "id": "m1", "date": "2026-11-30", "statement": "NVIDIA Q3 FY2027 disclosure states Vera Rubin contributed more than 25% of datacenter revenue.", "p": 0.35, "liveQuestionId": "lq-p98" } ], "falsifier": "Every NVIDIA quarter through the November 2027 release shows Data Center revenue above the prior quarter.", "whyItMatters": "A single down quarter would be the first public evidence that hyperscaler digestion, product transitions, or power-limited deployment can interrupt the cycle. Financing terms for GPU-backed debt and neocloud capacity are priced as if it cannot. The buildings get planned three years ahead of the chips, so the timing of the first pause matters more than its depth.", "whatWouldRaise": [ "NVIDIA misses its own quarterly revenue guide by more than 3%.", "Two or more hyperscalers guide capex flat or down for calendar 2027.", "A Rubin-generation transition slips a quarter, as Blackwell did in 2024." ], "whatWouldCut": [ "Fiscal 2028 guidance is raised above 70% growth.", "Sovereign and neocloud orders exceed 40% of Data Center revenue, diversifying demand.", "Memory and CoWoS supply expands faster than NVIDIA's stated constraint." ], "namedEntities": [ "NVIDIA" ], "marketCall": false, "recusalCheck": { "passed": true, "keysChecked": [ "employer-leasing", "named-competitor-bookings", "customer-pipeline", "ticker" ] }, "signedBy": "Brian Letort", "registeredAt": "2026-09-08", "amendments": [], "ruleHash": "dab52e675074af64f928270a292538acb0831fc35858a18cb591141c5dd35fb3", "distance": 1.0986122886681093, "distanceLabel": "clearly above consensus", "engine": { "id": "RT-02", "prior": { "p": 0.02, "method": "sec-xbrl-loglinear-bootstrap:li-1", "note": "log-linear growth +0.597/yr on the last 16 points, extrapolated from 2026-07-26; P(li-1 satisfies onTrack < 105.0 at 2027-12-31); proxy for the thesis, not its rule" }, "council": { "p": 0.19727869459800232, "votes": [ { "persona": "grid-and-power-planner", "model": "openai/gpt-5.6-sol", "p": 0.19, "rationale": "I use NVIDIA’s recent AI-cycle quarters as the reference class, adjusted for five sequential opportunities and the unusually strong supplied guidance. A roughly 0.04–0.05 conditional quarterly decline hazard compounds to about 0.19, allowing for correlated demand and supply conditions. The $89.0B baseline, 18% sequential growth, $108B company guidance, claimed 70% fiscal-2028 growth, and rising inventory all favor continued expansion. However, one-quarter shipment timing, export restrictions, component bottlenecks, or the Rubin transition could create a lower Data Center print despite strong annual growth. I am below 0.25 because supply constraint and backlog reduce near-term digestion risk; the 0.02 trend prior is too insensitive to transition shocks." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.24, "rationale": "I use NVIDIA’s post-2016 Data Center sequence as the reference class, conditioned on an unusually supply-constrained expansion: an annual decline hazard near 0.18 compounds to roughly 0.22 across the five comparisons. I raise that modestly because inventory is growing faster than revenue, increasing transition, shipment-timing, and digestion risk. Q3 FY2027 likely clears the $89B baseline given the $108B company guide, leaving most risk in FY2028 around Rubin, HBM, and packaging synchronization. Conversely, the cited roughly 70% FY2028 growth path and demand exceeding supply make an outright decline substantially less likely than mere deceleration." }, { "persona": "enterprise-cio", "model": "openai/gpt-5.6-sol", "p": 0.22, "rationale": "My reference class is five-quarter runs for a hypergrowth semiconductor segment during a supply-constrained product ramp. The $108B Q3 total-revenue guide makes the first comparison very unlikely to decline, while the stated fiscal-2028 growth path supports continued expansion. However, five reporting opportunities compound transition, shipment-timing, export-control, and customer-digestion risks; even a modest one-quarter interruption qualifies. Inventory rose 22% sequentially versus 18% for revenue, increasing execution and digestion risk but also supporting future shipments. I give limited weight to the 0.02 structural prior because it extrapolates consolidated revenue, not Data Center’s quarter-to-quarter path. Reporting discontinuation resolving MISS slightly lowers the estimate." }, { "persona": "credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.17, "rationale": "Reference class: high-growth, supply-constrained semiconductor ramps, where sequential declines are uncommon but transition or execution shocks create a low-single-digit quarterly hazard. Five observations compound that risk, although strong Q3 guidance makes the first test unlikely to fail. The stated fiscal-2028 growth target and demand exceeding supply materially favor monotonic growth. Offsetting this, inventory rose from $21.4B to $31.6B in two quarters, faster than revenue, increasing transition, mix, and digestion risk around Rubin. I place more weight than the 0.02 trend prior on discrete shocks, but less than the forecaster’s 0.25 because a decline must appear in the named GAAP line by November 2027." }, { "persona": "geopolitics-policy-analyst", "model": "openai/gpt-5.6-sol", "p": 0.19, "rationale": "I use a roughly 0.12 annual hazard for a sequential decline during a supply-constrained semiconductor ramp growing above 50%; compounded across the five covered releases, that gives about 0.15. Q3 guidance and the stated fiscal-2028 growth path make the first declines unlikely. I raise the estimate for later quarters because Rubin transitions, customer digestion, export restrictions, and inventory rising from $21.4B to $31.6B can produce shipment timing or correction risk. The revenue indicator remains strongly upward, however, and measures total rather than Data Center revenue. Discontinued segment reporting counts as MISS, modestly reducing the probability." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.18, "rationale": "I use a 0.40 reference-class base rate for at least one sequential decline across five quarters among rapidly growing, cyclical semiconductor businesses. NVIDIA’s Q3 guidance, reported supply constraints, and targeted fiscal-2028 growth near 0.70 materially reduce that prior; the first observation especially appears unlikely to decline from $89.0B. Rising inventory and the Rubin transition preserve execution, mix, and customer-digestion risk in later quarters. The 0.02 structural prior overweights smooth extrapolation and omits transition shocks. Conversely, stopping Data Center reporting explicitly resolves MISS, slightly lowering the estimate. My estimate is below the forecaster’s 0.25 because sustained supply-constrained growth is stronger evidence than a generic transition-risk narrative." } ], "aggregation": "log-odds-mean", "extremization": 1, "dissent": null, "spread": 0.07, "warnings": [ "all 6 votes came from one vendor (openai); extremization disabled, persona diversity only" ] }, "final": { "p": 0.06613676326296278, "ci80": [ 0.056114785428577806, 0.14070340550246302 ], "method": "stacked-fixed-weights (radar: prior 0.5, council 0.5; no resolved radar items to learn from)", "weights": { "council": 0.5, "prior": 0.5 } }, "stated": 0.25, "history": [ { "asOf": "2026-09-07", "p": 0.817052365623855, "source": "radar-engine", "runId": "radar-2026-09-07T23-12-09Z-06a58f" }, { "asOf": "2026-09-07", "p": 0.06613676326296278, "source": "radar-engine", "runId": "radar-2026-09-07T23-49-04Z-de8abf" } ], "indicatorStatus": "on-track", "milestones": [ { "id": "m1", "enginePrior": 0.44378638815176197, "councilP": 0.18008751377183058 } ], "arena": "compute-supply", "status": "draft", "tier": "T1" }, "indicators": [ { "thesisId": "RT-02", "indicatorId": "li-1", "connector": "sec-xbrl", "seriesId": "0001045810:Revenues:USD", "unit": "USD billions", "asOf": "2026-07-26", "current": 96.221, "status": "on-track", "history": [ { "asOf": "2009-10-25", "value": 0.9032060000000001 }, { "asOf": "2010-01-31", "value": 0.982488 }, { "asOf": "2010-05-02", "value": 1.001813 }, { "asOf": "2010-07-31", "value": 0.811208 }, { "asOf": "2010-10-31", "value": 0.8439120000000001 }, { "asOf": "2011-01-30", "value": 0.886376 }, { "asOf": "2011-05-01", "value": 0.9620390000000001 }, { "asOf": "2011-07-31", "value": 1.0165170000000001 }, { "asOf": "2011-10-30", "value": 1.0661800000000001 }, { "asOf": "2012-01-29", "value": 0.9531940000000001 }, { "asOf": "2012-04-29", "value": 0.9248770000000001 }, { "asOf": "2012-07-29", "value": 1.04427 }, { "asOf": "2012-10-28", "value": 1.20411 }, { "asOf": "2013-01-27", "value": 1.106902 }, { "asOf": "2013-04-28", "value": 0.954739 }, { "asOf": "2013-07-28", "value": 0.977238 }, { "asOf": "2013-10-27", "value": 1.053967 }, { "asOf": "2014-01-26", "value": 1.1442180000000002 }, { "asOf": "2014-04-27", "value": 1.103 }, { "asOf": "2014-07-27", "value": 1.103 }, { "asOf": "2014-10-26", "value": 1.225 }, { "asOf": "2015-01-25", "value": 1.2510000000000001 }, { "asOf": "2015-04-26", "value": 1.151 }, { "asOf": "2015-07-26", "value": 1.153 }, { "asOf": "2015-10-25", "value": 1.3050000000000002 }, { "asOf": "2016-01-31", "value": 1.401 }, { "asOf": "2016-05-01", "value": 1.3050000000000002 }, { "asOf": "2016-07-31", "value": 1.4280000000000002 }, { "asOf": "2016-10-30", "value": 2.004 }, { "asOf": "2017-01-29", "value": 2.173 }, { "asOf": "2017-04-30", "value": 1.937 }, { "asOf": "2017-07-30", "value": 2.23 }, { "asOf": "2017-10-29", "value": 2.636 }, { "asOf": "2018-01-28", "value": 2.911 }, { "asOf": "2018-04-29", "value": 3.2070000000000003 }, { "asOf": "2018-07-29", "value": 3.123 }, { "asOf": "2018-10-28", "value": 3.181 }, { "asOf": "2019-04-28", "value": 2.22 }, { "asOf": "2019-07-28", "value": 2.579 }, { "asOf": "2019-10-27", "value": 3.0140000000000002 }, { "asOf": "2020-04-26", "value": 3.08 }, { "asOf": "2020-07-26", "value": 3.866 }, { "asOf": "2020-10-25", "value": 4.726 }, { "asOf": "2021-05-02", "value": 5.6610000000000005 }, { "asOf": "2021-08-01", "value": 6.507000000000001 }, { "asOf": "2021-10-31", "value": 7.103000000000001 }, { "asOf": "2022-05-01", "value": 8.288 }, { "asOf": "2022-07-31", "value": 6.704000000000001 }, { "asOf": "2022-10-30", "value": 5.931 }, { "asOf": "2023-04-30", "value": 7.192 }, { "asOf": "2023-07-30", "value": 13.507000000000001 }, { "asOf": "2023-10-29", "value": 18.12 }, { "asOf": "2024-04-28", "value": 26.044 }, { "asOf": "2024-07-28", "value": 30.040000000000003 }, { "asOf": "2024-10-27", "value": 35.082 }, { "asOf": "2025-04-27", "value": 44.062000000000005 }, { "asOf": "2025-07-27", "value": 46.743 }, { "asOf": "2025-10-26", "value": 57.006 }, { "asOf": "2026-04-26", "value": 81.61500000000001 }, { "asOf": "2026-07-26", "value": 96.221 } ], "url": "https://data.sec.gov/api/xbrl/companyconcept/CIK0001045810/us-gaap/Revenues.json", "note": "quarterly frames; raw USD facts scaled by 1e-09 to match unit 'USD billions'" }, { "thesisId": "RT-02", "indicatorId": "li-2", "connector": "sec-xbrl", "seriesId": "0001045810:InventoryNet:USD", "unit": "USD billions", "asOf": "2026-07-26", "current": 31.575000000000003, "status": "on-track", "history": [ { "asOf": "2011-10-30", "value": 0.319602 }, { "asOf": "2012-01-29", "value": 0.340297 }, { "asOf": "2012-04-29", "value": 0.34270700000000004 }, { "asOf": "2012-07-29", "value": 0.387185 }, { "asOf": "2012-10-28", "value": 0.428983 }, { "asOf": "2013-01-27", "value": 0.41246700000000003 }, { "asOf": "2013-04-28", "value": 0.377597 }, { "asOf": "2013-07-28", "value": 0.37828 }, { "asOf": "2013-10-27", "value": 0.380319 }, { "asOf": "2014-01-26", "value": 0.387765 }, { "asOf": "2014-04-27", "value": 0.39328 }, { "asOf": "2014-07-27", "value": 0.387434 }, { "asOf": "2014-10-26", "value": 0.408081 }, { "asOf": "2015-01-25", "value": 0.48300000000000004 }, { "asOf": "2015-04-26", "value": 0.438 }, { "asOf": "2015-07-26", "value": 0.441 }, { "asOf": "2015-10-25", "value": 0.42500000000000004 }, { "asOf": "2016-01-31", "value": 0.41800000000000004 }, { "asOf": "2016-05-01", "value": 0.394 }, { "asOf": "2016-07-31", "value": 0.521 }, { "asOf": "2016-10-30", "value": 0.679 }, { "asOf": "2017-01-29", "value": 0.794 }, { "asOf": "2017-04-30", "value": 0.8210000000000001 }, { "asOf": "2017-07-30", "value": 0.8550000000000001 }, { "asOf": "2017-10-29", "value": 0.8570000000000001 }, { "asOf": "2018-01-28", "value": 0.796 }, { "asOf": "2018-04-29", "value": 0.797 }, { "asOf": "2018-07-29", "value": 1.09 }, { "asOf": "2018-10-28", "value": 1.417 }, { "asOf": "2019-01-27", "value": 1.5750000000000002 }, { "asOf": "2019-04-28", "value": 1.4260000000000002 }, { "asOf": "2019-07-28", "value": 1.2040000000000002 }, { "asOf": "2019-10-27", "value": 1.0470000000000002 }, { "asOf": "2020-01-26", "value": 0.9790000000000001 }, { "asOf": "2020-04-26", "value": 1.1280000000000001 }, { "asOf": "2020-07-26", "value": 1.401 }, { "asOf": "2020-10-25", "value": 1.495 }, { "asOf": "2021-01-31", "value": 1.826 }, { "asOf": "2021-05-02", "value": 1.9920000000000002 }, { "asOf": "2021-08-01", "value": 2.1140000000000003 }, { "asOf": "2021-10-31", "value": 2.233 }, { "asOf": "2022-01-30", "value": 2.605 }, { "asOf": "2022-05-01", "value": 3.1630000000000003 }, { "asOf": "2022-07-31", "value": 3.8890000000000002 }, { "asOf": "2022-10-30", "value": 4.454000000000001 }, { "asOf": "2023-01-29", "value": 5.159000000000001 }, { "asOf": "2023-04-30", "value": 4.611000000000001 }, { "asOf": "2023-07-30", "value": 4.319 }, { "asOf": "2023-10-29", "value": 4.779 }, { "asOf": "2024-01-28", "value": 5.282 }, { "asOf": "2024-04-28", "value": 5.864000000000001 }, { "asOf": "2024-07-28", "value": 6.675000000000001 }, { "asOf": "2024-10-27", "value": 7.654000000000001 }, { "asOf": "2025-01-26", "value": 10.08 }, { "asOf": "2025-04-27", "value": 11.333 }, { "asOf": "2025-07-27", "value": 14.962000000000002 }, { "asOf": "2025-10-26", "value": 19.784000000000002 }, { "asOf": "2026-01-25", "value": 21.403000000000002 }, { "asOf": "2026-04-26", "value": 25.797 }, { "asOf": "2026-07-26", "value": 31.575000000000003 } ], "url": "https://data.sec.gov/api/xbrl/companyconcept/CIK0001045810/us-gaap/InventoryNet.json", "note": "quarterly frames; raw USD facts scaled by 1e-09 to match unit 'USD billions'" } ], "indicatorStatus": "on-track", "chainEvents": [ { "seq": 200, "ts": "2026-09-07", "kind": "repriced", "hash": "e265b16bf493eea15c7096ef212f4d3d8bbe1f1dc2ad7d807daf39a207bc5f7e" }, { "seq": 201, "ts": "2026-09-07", "kind": "repriced", "hash": "7f40efa96e1392f292de50eaf630ef76871369c483a84defdcb31c1a244c7f36" }, { "seq": 259, "ts": "2026-09-08", "kind": "registered", "hash": "6e2c7d869f28260552647156def866f27bc65ca9d237289f3c33b0a1bcf1b2dc" } ], "url": "/predictions/compute-supply/RT-02" }, { "id": "RT-03", "title": "A 10-zettaFLOP/s cluster is operational by 2028", "arena": "compute-supply", "tier": "T2", "tag": "CALL", "direction": "up", "status": "registered", "statement": "Epoch AI's GPU Clusters dataset lists at least one single-site AI cluster with estimated 16-bit performance of 1e22 FLOP/s or more, with status operational and an operational date on or before 2028-12-31.", "resolutionRule": "Open the Epoch AI \"GPU Clusters\" data page (epoch.ai/data/gpu-clusters) on or after 2029-01-01 and no later than 2029-03-31. Resolves HIT if any row with status \"Operational\" (or Epoch's equivalent existing-status label) has an estimated performance of >= 1e22 FLOP/s at 16-bit precision (Epoch's headline performance column) and a first-operational date on or before 2028-12-31. Distributed multi-site training systems count only if Epoch lists them as a single cluster entry. If Epoch discontinues the dataset or removes the performance column before resolution, resolves MISS. Epoch's 2026 extrapolation places the leading cluster near 3e21 FLOP/s by mid-2028.", "resolutionSource": { "name": "Epoch AI GPU Clusters dataset", "url": "https://epoch.ai/data/gpu-clusters" }, "ambiguity": "resolves-against-forecaster", "horizon": "2029-03-31", "p": 0.5, "ci80": [ 0.35, 0.65 ], "consensus": { "value": "Epoch AI's trend extrapolation (Pilz et al., \"Trends in AI Supercomputers\") projects the leading cluster at roughly 3e21 FLOP/s by June 2028 and 8e21 by June 2029, reaching 1e22 only around 2029-2030.", "impliedP": 0.25, "impliedBy": "Epoch's 2.5x/year extrapolation reaches ~5e21 FLOP/s by end-2028, half the threshold; 1e22 by that date requires the leading cluster to run about one standard deviation (roughly a year) ahead of trend, which the historical scatter around Epoch's line gives about one chance in four.", "source": "Epoch AI, Trends in AI Supercomputers / GPU Clusters data page", "url": "https://epoch.ai/data/gpu-clusters", "asOf": "2026-09-07", "note": "No Metaculus question targets 1e22 FLOP/s at a dated threshold. My 0.50 reflects announced 2027-2028 multi-gigawatt single-site builds (xAI, OpenAI-Oracle, Anthropic-Amazon) that are already above Epoch's trend line." }, "leadingIndicators": [ { "id": "li-1", "label": "Largest operational cluster, 16-bit FLOP/s (Epoch)", "connector": "epoch", "seriesId": "gpu_clusters:performance_16bit_flop_s:max", "unit": "FLOP/s", "onTrack": { "op": ">=", "value": 2e+21 }, "offTrack": { "op": "<", "value": 1.2e+21 }, "url": "https://epoch.ai/data/gpu-clusters", "note": "Maximum of the 16-bit performance column over rows with status Operational. Thresholds are for the check on 2027-06-30." }, { "id": "li-2", "label": "NVIDIA quarterly revenue (proxy for accelerator shipments)", "connector": "sec-xbrl", "seriesId": "0001045810:Revenues:USD", "unit": "USD billions", "onTrack": { "op": ">=", "value": 120 }, "offTrack": { "op": "<", "value": 95 }, "url": "https://data.sec.gov/api/xbrl/companyconcept/CIK0001045810/us-gaap/Revenues.json", "note": "Quarterly revenue for the quarter ending 2027-04. A cluster of this size needs roughly one to two million Rubin-class accelerators, so shipment pace bounds it." } ], "milestones": [ { "id": "m1", "date": "2027-06-30", "statement": "Epoch lists an operational cluster at or above 2e21 FLOP/s (16-bit).", "p": 0.7 }, { "id": "m2", "date": "2028-06-30", "statement": "Epoch lists an operational cluster at or above 5e21 FLOP/s (16-bit).", "p": 0.55 }, { "id": "m3", "date": "2028-12-31", "statement": "Epoch lists an operational cluster at or above 1e22 FLOP/s (16-bit).", "p": 0.5 } ], "falsifier": "On 2028-06-30 the largest operational cluster in Epoch's dataset is below 4e21 FLOP/s, which leaves no plausible path to 1e22 within six months.", "whyItMatters": "A 1e22 FLOP/s cluster at Rubin-class efficiency draws on the order of five gigawatts at one site. Whether that arrives in 2028 or 2030 determines when single-campus power demand outgrows what any single utility interconnection queue can deliver, and whether training stays centralized or is forced to distribute across regions.", "whatWouldRaise": [ "xAI, OpenAI, or Anthropic publicly state a single-site deployment above 1.5 million GB300/Rubin-class GPUs.", "Epoch adds a planned cluster entry at or above 1e22 with a 2028 operational date.", "A utility or ISO approves a single-customer interconnection above 3 GW with a 2028 energization date." ], "whatWouldCut": [ "Frontier labs shift to multi-site distributed training and stop building single-site systems above 2 GW.", "Epoch's mid-2027 leading cluster is below 1.5e21 FLOP/s.", "Turbine or transformer delivery delays push announced 2028 campuses into 2029." ], "namedEntities": [ "Epoch AI" ], "marketCall": false, "recusalCheck": { "passed": true, "keysChecked": [ "employer-leasing", "named-competitor-bookings", "customer-pipeline", "ticker" ] }, "signedBy": "Brian Letort", "registeredAt": "2026-09-08", "amendments": [], "ruleHash": "9439f485df80c190cc0d031edf9eba040acaca5d3ae68cd7a31251873b0a27ab", "distance": 1.0986122886681098, "distanceLabel": "clearly above consensus", "engine": { "id": "RT-03", "prior": { "p": 0.719089730901256, "method": "drift-gbm-terminal:li-1", "note": "P(indicator li-1 satisfies onTrack >= 2e+21 at 2029-03-31); drift +0.97/yr, vol 1.42/sqrt(yr) from 10 points; proxy for the thesis, not its rule" }, "council": { "p": 0.3601379184845809, "votes": [ { "persona": "grid-and-power-planner", "model": "openai/gpt-5.6-sol", "p": 0.35, "rationale": "Using Epoch’s historical frontier-cluster step-ups as the reference class, I estimate qualifying-leap hazards near 0.12 in 2027 and 0.30 in 2028; compounding, plus a small late-2026 contribution, gives roughly 0.40 before resolution penalties. The latest listed maximum, 2.73e20, requires a 36.6-fold increase, while Epoch’s trend reaches only ~3e21 in mid-2028 and ~8e21 in mid-2029. Rapid accelerator revenue growth supports hardware availability but not its single-site concentration, power delivery, networking, commissioning, or Epoch’s treatment. Operational-date verification, dataset lag or discontinuation, and ambiguity resolving against the forecaster reduce my estimate below the stated 0.50 and well below the noisy GBM prior." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.39, "rationale": "The relevant reference class is frontier-cluster compute scaling, discounted for schedule slips in unprecedented accelerator deployments. Epoch’s observed maximum was 2.73e20 in July 2025, so the threshold requires a 36.6-fold increase in roughly 3.4 years. That resembles the noisy historical growth rate and explains the GBM prior, but Epoch’s smoother path reaches only 3e21 around mid-2028 and 8e21 around mid-2029. NVIDIA’s supplied revenue trajectory supports aggregate accelerator availability, yet success conjunctively requires sufficient packaging, HBM, networking, and energized power at one site, operation by year-end, and qualifying Epoch classification. Historical deployment slips justify discounting both the mechanical prior and the stated 0.5." }, { "persona": "enterprise-cio", "model": "openai/gpt-5.6-sol", "p": 0.29, "rationale": "Reference class: historical frontier-cluster scaling combined with multi-gigawatt technology-megaproject delivery, not announced capacity. The supplied Epoch maximum is 2.73e20 in July 2025, requiring 36.6x growth; Epoch’s extrapolation reaches only about 5e21 by end-2028, so success requires roughly a 2x positive deviation. I assign conditional crossing hazards of 0.02 for late 2026, 0.08 in 2027, and 0.23 in 2028. Compounding gives 0.30, followed by a modest discount for commissioning slippage, single-site classification, dataset lag, and discontinuation. The structural prior extrapolates volatile growth without sufficiently reflecting power, networking, and integration constraints." }, { "persona": "credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.42, "rationale": "Using the Epoch frontier-cluster series as the reference class, the target requires a 36.6-fold increase from 2.7288e20 in July 2025, or roughly 2.85x annual compounding through December 2028. That is near the 2020–2025 realized pace, while Epoch’s supplied central extrapolation is materially slower: about 3e21 by mid-2028 and below 1e22 through mid-2029. Rapid accelerator revenue growth and multi-gigawatt build plans create upside, but the claim is conjunctive: hardware, power, networking, commissioning, single-site classification, and Epoch’s timely operational labeling must all clear. This is below the forecaster’s 0.50 and well below the GBM prior, which likely overweights volatile historical jumps." }, { "persona": "geopolitics-policy-analyst", "model": "openai/gpt-5.6-sol", "p": 0.41, "rationale": "Reference class: Epoch’s 2020–2025 frontier-cluster series, tempered by its forward trend extrapolation and execution delays for multi-gigawatt infrastructure. Reaching 1e22 from the latest 2.73e20 observation requires roughly 37-fold scaling, or about 2.85-fold annually through 2028. Historical growth supports that pace, but Epoch’s extrapolation places the frontier around half the threshold at year-end 2028. Power delivery, accelerator availability, commissioning, single-site classification, and dataset-update timing all reduce the chance that announcements become a qualifying row. The structural prior likely overweights unconstrained historical growth. I therefore place the claim below the forecaster’s 0.5." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.31, "rationale": "Base rate: 0.25 for frontier-compute thresholds arriving at least one year ahead of a published central trend. Epoch’s extrapolation has the leader near 3e21 in mid-2028 and 8e21 in mid-2029; 1e22 by year-end 2028 therefore requires roughly a twofold upside surprise. The latest cluster reading, 2.7288e20 in July 2025, requires 36.6-fold growth in about 3.4 years. Rapid supplier revenue growth supports capacity expansion but does not verify one qualifying cluster. I adjust upward modestly for historically lumpy frontier jumps, then downward for single-entry treatment, operational status/date, Epoch classification, and dataset survival. The 0.7191 GBM prior overweights volatile tail growth and ignores disclosure risk." } ], "aggregation": "log-odds-mean", "extremization": 1, "dissent": null, "spread": 0.13, "warnings": [ "all 6 votes came from one vendor (openai); extremization disabled, persona diversity only" ] }, "final": { "p": 0.5455217538156822, "ci80": [ 0.33350071749812826, 0.7899178867325811 ], "method": "stacked-fixed-weights (radar: prior 0.5, council 0.5; no resolved radar items to learn from)", "weights": { "council": 0.5, "prior": 0.5 } }, "stated": 0.5, "history": [ { "asOf": "2026-09-07", "p": 0.5231461209559903, "source": "radar-engine", "runId": "radar-2026-09-07T23-12-09Z-06a58f" }, { "asOf": "2026-09-07", "p": 0.5455217538156822, "source": "radar-engine", "runId": "radar-2026-09-07T23-49-04Z-de8abf" } ], "indicatorStatus": "off-track", "milestones": [ { "id": "m1", "enginePrior": 0.65, "councilP": null }, { "id": "m2", "enginePrior": 0.65, "councilP": null }, { "id": "m3", "enginePrior": null, "councilP": null } ], "arena": "compute-supply", "status": "draft", "tier": "T2" }, "indicators": [ { "thesisId": "RT-03", "indicatorId": "li-1", "connector": "epoch", "seriesId": "gpu_clusters:performance_16bit_flop_s:max", "unit": "FLOP/s", "asOf": "2025-07-22", "current": 272880000000000000000, "status": "off-track", "history": [ { "asOf": "2016-11-01", "value": 57912000000000000 }, { "asOf": "2017-11-15", "value": 2750000000000000000 }, { "asOf": "2018-12-15", "value": 2160000000000000000 }, { "asOf": "2019-12-15", "value": 700000000000000000 }, { "asOf": "2020-11-30", "value": 1397760000000000000 }, { "asOf": "2021-12-20", "value": 6000000000000000000 }, { "asOf": "2022-12-15", "value": 14413056000000000000 }, { "asOf": "2023-12-15", "value": 14247360000000000000 }, { "asOf": "2024-12-15", "value": 98940000000000000000 }, { "asOf": "2025-07-22", "value": 272880000000000000000 } ], "url": "https://epoch.ai/data/gpu_clusters.csv", "note": "max of 'performance_16bit_flop_s' per calendar year" }, { "thesisId": "RT-03", "indicatorId": "li-2", "connector": "sec-xbrl", "seriesId": "0001045810:Revenues:USD", "unit": "USD billions", "asOf": "2026-07-26", "current": 96.221, "status": "between", "history": [ { "asOf": "2009-10-25", "value": 0.9032060000000001 }, { "asOf": "2010-01-31", "value": 0.982488 }, { "asOf": "2010-05-02", "value": 1.001813 }, { "asOf": "2010-07-31", "value": 0.811208 }, { "asOf": "2010-10-31", "value": 0.8439120000000001 }, { "asOf": "2011-01-30", "value": 0.886376 }, { "asOf": "2011-05-01", "value": 0.9620390000000001 }, { "asOf": "2011-07-31", "value": 1.0165170000000001 }, { "asOf": "2011-10-30", "value": 1.0661800000000001 }, { "asOf": "2012-01-29", "value": 0.9531940000000001 }, { "asOf": "2012-04-29", "value": 0.9248770000000001 }, { "asOf": "2012-07-29", "value": 1.04427 }, { "asOf": "2012-10-28", "value": 1.20411 }, { "asOf": "2013-01-27", "value": 1.106902 }, { "asOf": "2013-04-28", "value": 0.954739 }, { "asOf": "2013-07-28", "value": 0.977238 }, { "asOf": "2013-10-27", "value": 1.053967 }, { "asOf": "2014-01-26", "value": 1.1442180000000002 }, { "asOf": "2014-04-27", "value": 1.103 }, { "asOf": "2014-07-27", "value": 1.103 }, { "asOf": "2014-10-26", "value": 1.225 }, { "asOf": "2015-01-25", "value": 1.2510000000000001 }, { "asOf": "2015-04-26", "value": 1.151 }, { "asOf": "2015-07-26", "value": 1.153 }, { "asOf": "2015-10-25", "value": 1.3050000000000002 }, { "asOf": "2016-01-31", "value": 1.401 }, { "asOf": "2016-05-01", "value": 1.3050000000000002 }, { "asOf": "2016-07-31", "value": 1.4280000000000002 }, { "asOf": "2016-10-30", "value": 2.004 }, { "asOf": "2017-01-29", "value": 2.173 }, { "asOf": "2017-04-30", "value": 1.937 }, { "asOf": "2017-07-30", "value": 2.23 }, { "asOf": "2017-10-29", "value": 2.636 }, { "asOf": "2018-01-28", "value": 2.911 }, { "asOf": "2018-04-29", "value": 3.2070000000000003 }, { "asOf": "2018-07-29", "value": 3.123 }, { "asOf": "2018-10-28", "value": 3.181 }, { "asOf": "2019-04-28", "value": 2.22 }, { "asOf": "2019-07-28", "value": 2.579 }, { "asOf": "2019-10-27", "value": 3.0140000000000002 }, { "asOf": "2020-04-26", "value": 3.08 }, { "asOf": "2020-07-26", "value": 3.866 }, { "asOf": "2020-10-25", "value": 4.726 }, { "asOf": "2021-05-02", "value": 5.6610000000000005 }, { "asOf": "2021-08-01", "value": 6.507000000000001 }, { "asOf": "2021-10-31", "value": 7.103000000000001 }, { "asOf": "2022-05-01", "value": 8.288 }, { "asOf": "2022-07-31", "value": 6.704000000000001 }, { "asOf": "2022-10-30", "value": 5.931 }, { "asOf": "2023-04-30", "value": 7.192 }, { "asOf": "2023-07-30", "value": 13.507000000000001 }, { "asOf": "2023-10-29", "value": 18.12 }, { "asOf": "2024-04-28", "value": 26.044 }, { "asOf": "2024-07-28", "value": 30.040000000000003 }, { "asOf": "2024-10-27", "value": 35.082 }, { "asOf": "2025-04-27", "value": 44.062000000000005 }, { "asOf": "2025-07-27", "value": 46.743 }, { "asOf": "2025-10-26", "value": 57.006 }, { "asOf": "2026-04-26", "value": 81.61500000000001 }, { "asOf": "2026-07-26", "value": 96.221 } ], "url": "https://data.sec.gov/api/xbrl/companyconcept/CIK0001045810/us-gaap/Revenues.json", "note": "quarterly frames; raw USD facts scaled by 1e-09 to match unit 'USD billions'" } ], "indicatorStatus": "off-track", "chainEvents": [ { "seq": 202, "ts": "2026-09-07", "kind": "repriced", "hash": "28e4a9354e193a7c50842ae871183ed11043d5fb648be0d0909d384d3d1b19da" }, { "seq": 203, "ts": "2026-09-07", "kind": "repriced", "hash": "86336ad2dd60e88fa0ff7137fed300f0d5b57c359d41bbd418f287e7783679a8" }, { "seq": 260, "ts": "2026-09-08", "kind": "registered", "hash": "15996ecc4a4c59ffdb605eec962cc7487e0f2d4c3496c4994c3c6ae88a10ebc7" } ], "url": "/predictions/compute-supply/RT-03" }, { "id": "RT-04", "title": "NVIDIA falls below half of merchant accelerator revenue", "arena": "compute-supply", "tier": "T3", "tag": "THESIS", "direction": "shift", "status": "registered", "statement": "In at least one calendar quarter ending on or before 2031-12-31, NVIDIA's Data Center revenue is less than 50% of the merchant AI accelerator revenue pool, defined as NVIDIA Data Center plus AMD Data Center segment plus Broadcom AI semiconductor plus Marvell Data Center segment revenue.", "resolutionRule": "For each calendar quarter, take the four as-reported segment figures from each company's earnings press release for the fiscal quarter whose period end falls within 45 days of the calendar quarter end: NVIDIA \"Data Center\", AMD \"Data Center\" segment, Broadcom \"AI semiconductor revenue\", Marvell \"Data Center\" end market. Compute NVIDIA / (sum of four). Resolves HIT if the ratio is below 0.50 in any quarter ending on or before 2031-12-31. If any company stops disclosing the named line, substitute its total revenue (which raises the denominator and therefore cannot produce a false HIT); if NVIDIA stops disclosing Data Center, resolves MISS. Baseline Q2/Q3 2026 pair: approximately 89 / (89 + 16.7 + AMD + Marvell), roughly 75-78%.", "resolutionSource": { "name": "NVIDIA, AMD, Broadcom and Marvell quarterly earnings press releases", "url": "https://investor.nvidia.com/financial-info/quarterly-results/default.aspx" }, "ambiguity": "resolves-against-forecaster", "horizon": "2031-12-31", "p": 0.35, "ci80": [ 0.2, 0.5 ], "consensus": { "value": "Both leaders' guidance keeps NVIDIA near three quarters of the pool: Broadcom targets ~$230B AI revenue in FY2028 while NVIDIA targets ~70% growth in fiscal 2028 on a ~$400B base; no public sell-side model shows NVIDIA below 60% of merchant accelerator revenue before 2030.", "impliedP": 0.15, "impliedBy": "Broadcom's $230B FY2028 target plus AMD and Marvell at plausible run rates puts the non-NVIDIA pool near $90-100B/quarter in 2028 against NVIDIA Data Center of ~$150-200B/quarter, a 60-67% NVIDIA share. Falling below 50% by 2031 needs the rest of the pool to grow roughly twice as fast as NVIDIA for three more years; the guidance path gives that about one chance in seven.", "source": "Broadcom Q3 FY2026 and NVIDIA Q2 FY2027 earnings calls", "url": "https://www.fool.com/earnings/call-transcripts/2026/08/26/nvidia-nvda-q2-2027-earnings-call-transcript/", "asOf": "2026-09-04", "note": "No prediction market prices vendor share. Consensus is inferred from the two companies' multi-year guidance; treat as a soft baseline." }, "leadingIndicators": [ { "id": "li-1", "label": "NVIDIA share of four-vendor pool, quarterly", "connector": "manual", "seriesId": "nvda-share-merchant-accelerator-pool", "unit": "ratio", "onTrack": { "op": "<=", "value": 0.68 }, "offTrack": { "op": ">", "value": 0.76 }, "url": "https://investor.nvidia.com/financial-info/quarterly-results/default.aspx", "note": "Computed per the resolution rule from the four earnings releases each quarter. Thresholds are for the quarter ending 2027-12-31." }, { "id": "li-2", "label": "AMD Data Center segment revenue, quarterly", "connector": "sec-xbrl", "seriesId": "0000002488:RevenueFromContractWithCustomerExcludingAssessedTax:USD", "unit": "USD billions", "onTrack": { "op": ">=", "value": 20 }, "offTrack": { "op": "<", "value": 12 }, "url": "https://data.sec.gov/api/xbrl/companyconcept/CIK0000002488/us-gaap/RevenueFromContractWithCustomerExcludingAssessedTax.json", "note": "AMD total revenue as XBRL proxy (segment detail is in the press release). Thresholds are for the quarter ending 2027-12. AMD is the swing vendor for the pool." } ], "milestones": [ { "id": "m1", "date": "2027-12-31", "statement": "NVIDIA share of the four-vendor pool is at or below 70% in at least one quarter.", "p": 0.5 }, { "id": "m2", "date": "2029-12-31", "statement": "NVIDIA share of the four-vendor pool is at or below 60% in at least one quarter.", "p": 0.4 }, { "id": "m3", "date": "2031-12-31", "statement": "NVIDIA share of the four-vendor pool is below 50% in at least one quarter.", "p": 0.35 } ], "falsifier": "NVIDIA's share of the four-vendor pool remains at or above 70% in every quarter through 2029-12-31; the thesis is then withdrawn as falsified regardless of the 2031 horizon.", "whyItMatters": "The buildings, networks, and power contracts that outlast any chip generation are being designed around one vendor's rack architecture. If merchant supply splits below 50/50 by 2031, liquid-cooling density, fabric choice, and financing structures diversify with it. If it does not, single-vendor concentration remains the largest unpriced operational risk in the sector.", "whatWouldRaise": [ "Google, Amazon, or Meta disclose custom-ASIC accelerator capex exceeding their NVIDIA purchases.", "AMD Data Center revenue exceeds $25B in a quarter before 2029.", "A second frontier lab trains a flagship model on non-NVIDIA silicon end to end." ], "whatWouldCut": [ "NVIDIA Data Center revenue passes $200B in a quarter before 2029.", "A hyperscaler cancels a custom ASIC program and returns to merchant GPUs.", "Broadcom or Marvell guide AI revenue down for two consecutive quarters." ], "namedEntities": [ "NVIDIA", "AMD", "Broadcom", "Marvell" ], "marketCall": false, "recusalCheck": { "passed": true, "keysChecked": [ "employer-leasing", "named-competitor-bookings", "customer-pipeline", "ticker" ] }, "signedBy": "Brian Letort", "registeredAt": "2026-09-08", "amendments": [], "ruleHash": "3e4e68b25f799f8cd6e797fc6bd9475de93ae6c9854a6dc1abf354bb5df75aa9", "distance": 1.1155618469818829, "distanceLabel": "clearly above consensus", "engine": { "id": "RT-04", "prior": { "p": 0.98, "method": "sec-xbrl-loglinear-bootstrap:li-2", "note": "log-linear growth +0.165/yr on the last 16 points, extrapolated from 2026-06-27; P(li-2 satisfies onTrack >= 20.0 at 2031-12-31); proxy for the thesis, not its rule" }, "council": { "p": 0.24888639966912407, "votes": [ { "persona": "grid-and-power-planner", "model": "openai/gpt-5.6-sol", "p": 0.23, "rationale": "I use the reference class of dominant compute platforms facing custom-silicon substitution: roughly a 0.08 annual hazard of a greater-than-25-point share break over 2027–31, compounding to about 0.34. I reduce that because the rule requires reported revenue, NVIDIA’s line includes networking and systems, and the 2029 falsifier excludes late-only erosion. Moving from 75–78% to below 50% requires peers’ combined revenue relative to NVIDIA to more than triple. Broadcom’s supplied trajectory makes 70% plausible, but 50% still needs sustained differential growth or an NVIDIA downturn. The 0.98 structural prior extrapolates one XBRL series, not the four-company ratio. My estimate is below 0.35." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.23, "rationale": "Starting near 0.76, rivals are only about 0.32 times NVIDIA; a HIT requires them to exceed NVIDIA, more than tripling that relative ratio despite NVIDIA’s roadmap, software lock-in, and priority access to HBM and advanced packaging. I use the reference class of dominant compute platforms losing over 25 share points within five years, which is uncommon without major product disruption. Broadcom’s supplied FY2028 target, custom-ASIC growth, disclosure substitutions, and one-quarter testing create meaningful tail risk. I assign 0.55 to the required pre-2030 print below 0.70 and 0.42 conditional probability of reaching below 0.50 by 2031, or 0.23 compounded. AMD’s total-revenue extrapolation does not justify the structural prior." }, { "persona": "enterprise-cio", "model": "openai/gpt-5.6-sol", "p": 0.29, "rationale": "Using the reference class of dominant semiconductor-platform incumbents facing custom silicon, I apply roughly a 0.07 annual hazard of a >25-point share break over 5.3 years, then discount for the path-dependent falsifier. From a 75–78% baseline, rivals’ revenue relative to NVIDIA must roughly triple to exceed it. Broadcom’s stated ramp makes a sub-70 quarter by 2029 plausible, but NVIDIA’s software moat, annual product cadence, supply scale, and customers’ integration timelines make sub-50 by 2031 a tail. I estimate 0.62 for avoiding the 2029 falsifier and 0.46 conditional probability of a sub-50 print by the horizon, about 0.29. The 0.98 structural prior is uninformative because it extrapolates only Marvell total revenue. This is below the forecaster’s 0.35." }, { "persona": "credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.28, "rationale": "Using dominant compute-platform incumbents facing custom silicon and second-source entry as the reference class, I estimate an effective annual crossover hazard near 0.06, including the 2029 falsifier. From a 75–78% baseline, rivals’ combined revenue must rise from roughly 0.28–0.33 times NVIDIA’s to above 1.0 times it—a greater than threefold relative swing. Broadcom’s supplied guidance and rising AMD total revenue support erosion, but NVIDIA’s growth outlook and platform lock-in make sub-50 materially harder than reaching 60–70%. The 0.98 structural prior is inapplicable because AMD total revenue alone cannot forecast the four-vendor ratio. My estimate is below the forecaster’s 0.35." }, { "persona": "geopolitics-policy-analyst", "model": "openai/gpt-5.6-sol", "p": 0.26, "rationale": "I use the reference class of dominant computing platforms facing second-source and custom-silicon entry over five years. From the 75–78% baseline, rivals must roughly triple revenue relative to NVIDIA to force a sub-50% quarter. Broadcom’s stated trajectory and AMD’s growth make a sub-70% quarter by 2029 plausible, but NVIDIA’s scale, ecosystem, rapid product cadence, and Data Center networking revenue make the final step uncommon. An isolated product-transition quarter or competitor disclosure substitution raises the any-quarter probability. The 0.98 structural prior is not informative: it extrapolates one AMD total-revenue series rather than the four specified lines. The 2029 falsifier further limits paths to a HIT." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.21, "rationale": "Base rate: I use 0.20 for dominant semiconductor platforms losing at least 25 share points and falling below half within about five years. Starting near 75–78%, NVIDIA requires the other three reported pools to grow from roughly one-third of its revenue to more than parity. Supplied guidance still implies about 70% in FY2028, leaving a steep second leg. The 2029 falsifier makes this conjunctive: a sub-70 quarter must occur by then before a sub-50 quarter by 2031. “Any quarter” and competitor total-revenue substitutions raise odds somewhat; an NVIDIA disclosure change creates a MISS. The 0.98 prior extrapolates AMD total revenue, not the defined ratio, so receives negligible weight." } ], "aggregation": "log-odds-mean", "extremization": 1, "dissent": null, "spread": 0.08, "warnings": [ "all 6 votes came from one vendor (openai); extremization disabled, persona diversity only" ] }, "final": { "p": 0.8011711163903302, "ci80": [ 0.629685867710156, 0.8275882132609906 ], "method": "stacked-fixed-weights (radar: prior 0.5, council 0.5; no resolved radar items to learn from)", "weights": { "council": 0.5, "prior": 0.5 } }, "stated": 0.35, "history": [ { "asOf": "2026-09-07", "p": 0.8135886005396491, "source": "radar-engine", "runId": "radar-2026-09-07T23-12-09Z-06a58f" }, { "asOf": "2026-09-07", "p": 0.8011711163903302, "source": "radar-engine", "runId": "radar-2026-09-07T23-49-04Z-de8abf" } ], "indicatorStatus": "off-track", "milestones": [ { "id": "m1", "enginePrior": 0.5, "councilP": null }, { "id": "m2", "enginePrior": null, "councilP": null }, { "id": "m3", "enginePrior": null, "councilP": null } ], "arena": "compute-supply", "status": "draft", "tier": "T3" }, "indicators": [ { "thesisId": "RT-04", "indicatorId": "li-1", "connector": "manual", "seriesId": "nvda-share-merchant-accelerator-pool", "unit": "ratio", "asOf": "2026-09-07", "current": null, "status": "no-data", "history": [], "url": "https://investor.nvidia.com/financial-info/quarterly-results/default.aspx", "note": "manual: 'nvda-share-merchant-accelerator-pool' not in manual-series.json" }, { "thesisId": "RT-04", "indicatorId": "li-2", "connector": "sec-xbrl", "seriesId": "0000002488:RevenueFromContractWithCustomerExcludingAssessedTax:USD", "unit": "USD billions", "asOf": "2026-06-27", "current": 11.536000000000001, "status": "off-track", "history": [ { "asOf": "2017-04-01", "value": 1.1780000000000002 }, { "asOf": "2017-07-01", "value": 1.151 }, { "asOf": "2017-09-30", "value": 1.584 }, { "asOf": "2017-12-30", "value": 1.34 }, { "asOf": "2018-03-31", "value": 1.647 }, { "asOf": "2018-06-30", "value": 1.756 }, { "asOf": "2018-09-29", "value": 1.653 }, { "asOf": "2018-12-29", "value": 1.419 }, { "asOf": "2019-03-30", "value": 1.272 }, { "asOf": "2019-06-29", "value": 1.5310000000000001 }, { "asOf": "2019-09-28", "value": 1.8010000000000002 }, { "asOf": "2019-12-28", "value": 2.1270000000000002 }, { "asOf": "2020-03-28", "value": 1.786 }, { "asOf": "2020-06-27", "value": 1.9320000000000002 }, { "asOf": "2020-09-26", "value": 2.801 }, { "asOf": "2020-12-26", "value": 3.244 }, { "asOf": "2021-03-27", "value": 3.4450000000000003 }, { "asOf": "2021-06-26", "value": 3.85 }, { "asOf": "2021-09-25", "value": 4.313000000000001 }, { "asOf": "2022-03-26", "value": 5.8870000000000005 }, { "asOf": "2022-06-25", "value": 6.550000000000001 }, { "asOf": "2022-09-24", "value": 5.565 }, { "asOf": "2023-04-01", "value": 5.353000000000001 }, { "asOf": "2023-07-01", "value": 5.359 }, { "asOf": "2023-09-30", "value": 5.800000000000001 }, { "asOf": "2024-03-30", "value": 5.473000000000001 }, { "asOf": "2024-06-29", "value": 5.835 }, { "asOf": "2024-09-28", "value": 6.819000000000001 }, { "asOf": "2025-03-29", "value": 7.438000000000001 }, { "asOf": "2025-06-28", "value": 7.6850000000000005 }, { "asOf": "2025-09-27", "value": 9.246 }, { "asOf": "2026-03-28", "value": 10.253 }, { "asOf": "2026-06-27", "value": 11.536000000000001 } ], "url": "https://data.sec.gov/api/xbrl/companyconcept/CIK0000002488/us-gaap/RevenueFromContractWithCustomerExcludingAssessedTax.json", "note": "quarterly frames; raw USD facts scaled by 1e-09 to match unit 'USD billions'" } ], "indicatorStatus": "off-track", "chainEvents": [ { "seq": 204, "ts": "2026-09-07", "kind": "repriced", "hash": "3e27d46c3dd42292293ea196d8edf4c0ac15a5b7917cd03cf005ba6ff7938a50" }, { "seq": 205, "ts": "2026-09-07", "kind": "repriced", "hash": "791ac90442385d8c1f5f8b2bc4e7e50f2bdc943db440923cf4cde24eded24344" }, { "seq": 261, "ts": "2026-09-08", "kind": "registered", "hash": "d09bf1021b3ba9d6455b00dab863dd049dddee558d93093c3c71f02afa056c8c" } ], "url": "/predictions/compute-supply/RT-04" }, { "id": "RT-08", "title": "No native UALink switch ships in volume by 2027", "arena": "networking-fabric", "tier": "T1", "tag": "WARNING", "direction": "down", "status": "registered", "statement": "No UALink Consortium member announces general availability or volume production of a switch ASIC implementing the native UALink 200G specification (not an Ethernet-tunnelled or bridged variant) on or before 2027-12-31.", "resolutionRule": "Resolves MISS (the switch ships) if, on or before 2027-12-31, a UALink Consortium member company publishes a press release on its own newsroom or investor-relations site, or the Consortium posts on ualinkconsortium.org/news, stating that a switch ASIC natively implementing UALink 200G (the April 2025 1.0 specification or its successor) is \"generally available\", \"in volume production\", or \"shipping in production\" to customers. Announcements of sampling, early access, demonstrations, design wins, or of UALink-over-Ethernet bridge or tunnelling products do not count. Resolves HIT if no qualifying announcement exists by 2027-12-31. If the Consortium dissolves or its site goes offline, member newsrooms alone control.", "resolutionSource": { "name": "UALink Consortium news page and member press releases", "url": "https://ualinkconsortium.org/news/" }, "ambiguity": "resolves-against-forecaster", "horizon": "2027-12-31", "p": 0.5, "ci80": [ 0.35, 0.65 ], "consensus": { "value": "The Consortium's public roadmap at the 1.0 release targeted member silicon in 2026 and products in 2027; as of September 2026 announced deployments use UALink-over-Ethernet bridges, with native switch silicon still at sampling.", "impliedP": 0.35, "impliedBy": "The Consortium roadmap (silicon 2026, products 2027) taken at face value makes a 2027 GA announcement the base case; interconnect standards from PCIe 5 to CXL 3 have slipped GA by 6-18 months about a third of the time, so the roadmap implies roughly 0.35 that no native switch is GA by end-2027.", "source": "UALink Consortium 1.0 specification announcement and 2026 news posts", "url": "https://ualinkconsortium.org/news/", "asOf": "2026-09-07", "note": "No prediction market covers UALink. Honest coin flip: native switch silicon is at sampling with 15 months to run, and a vendor \"in production\" press release is cheap to issue, which pulls my number toward 0.50 rather than the 0.65 a pure slippage view would give. Left at 0.50 deliberately." }, "leadingIndicators": [ { "id": "li-1", "label": "Native UALink switch ASICs announced as sampling", "connector": "manual", "seriesId": "ualink-native-switch-sampling-count", "unit": "count", "onTrack": { "op": "<=", "value": 1 }, "offTrack": { "op": ">=", "value": 2 }, "url": "https://ualinkconsortium.org/news/", "note": "Count of distinct vendors (Astera Labs, Marvell, Broadcom, Cisco, Xconn, etc.) that have announced sampling of a native UALink switch. Thresholds are for 2026-12-31; two sampling vendors a year out makes 2027 GA likely." }, { "id": "li-2", "label": "Second non-Broadcom vendor cites 1.6T or CPO hyperscaler design win", "connector": "manual", "seriesId": "lq-p45-second-1-6t-design-win", "unit": "boolean", "onTrack": { "op": "==", "value": 0 }, "offTrack": { "op": "==", "value": 1 }, "url": "https://ualinkconsortium.org/news/", "note": "Mirrors Ledger live question lq-p45 (deadline 2026-09-30). Fast optics design wins outside Broadcom signal the scale-up ecosystem is maturing faster than this thesis assumes." } ], "milestones": [], "falsifier": "A UALink Consortium member announces general availability or volume production of a native UALink 200G switch ASIC on or before 2027-12-31.", "whyItMatters": "Scale-up fabric is where NVLink's moat lives. Until a native UALink switch ships in volume, every non-NVIDIA accelerator rack is bridged over Ethernet with a latency and power penalty, which caps how fast custom silicon can take share. Rack power density and cabling design for 2028 buildings depend on which fabric actually exists.", "whatWouldRaise": [ "Astera Labs or Marvell push native UALink switch sampling into 2027.", "AMD publicly positions Ethernet scale-up (UEC or Broadcom SUE) as its 2027 fabric instead of UALink.", "The Consortium publishes a 1.1 specification that changes the physical layer." ], "whatWouldCut": [ "Two vendors announce native UALink switch sampling before 2026-12-31.", "A hyperscaler publicly commits to a UALink-native rack for 2027 deployment.", "Broadcom announces a UALink-native Tomahawk variant." ], "namedEntities": [ "UALink Consortium" ], "marketCall": false, "recusalCheck": { "passed": true, "keysChecked": [ "employer-leasing", "named-competitor-bookings", "customer-pipeline", "ticker" ] }, "signedBy": "Brian Letort", "registeredAt": "2026-09-08", "amendments": [], "ruleHash": "326e66711c9355cc76ed2617dce9f634edd25211aca587ebcfa29a930872ab05", "distance": 0.6190392084062235, "distanceLabel": "leans above consensus", "engine": { "id": "RT-08", "prior": { "p": 0.5, "method": "reference-class:qtype:disclosure", "note": "ledger base rate, n=29, horizon 480d; the ledger has no multi-year history" }, "council": { "p": 0.4449647088988064, "votes": [ { "persona": "grid-and-power-planner", "model": "openai/gpt-5.6-sol", "p": 0.43, "rationale": "Reference class: new data-center switch ASICs moving from sampling to production, where 12–24 months is common but schedule slippage is frequent. Nearly 16 months remain. I move below the 0.50 structural prior for HIT because the consortium roadmap targets products in 2027, multiple members create several chances for a qualifying announcement, and issuing that announcement is inexpensive once production starts. I retain substantial HIT probability because native UALink is a new fabric, announced deployments remain bridged, sampling evidence is unverified here, and qualification, software, packaging, or system-readiness delays could push production beyond 2027. The no-data indicators add no readiness evidence. Against-forecaster ambiguity modestly reduces HIT probability." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.43, "rationale": "Reference class: first-generation interconnect switch ASICs typically require roughly 4–8 quarters from sampling through qualification and production, with frequent one-to-two-quarter slips. About five quarters remain, so any single programme could miss. However, the falsifier needs only one Consortium member to issue a qualifying announcement, and the supplied roadmap already targets products in 2027. Multiple potential vendors plus the low threshold of a production-status press release make a MISS slightly more likely than a HIT. Continued bridge-based deployments and absent indicator data preserve substantial slippage risk; strict, ambiguity-against-forecaster resolution also trims the HIT probability." }, { "persona": "enterprise-cio", "model": "openai/gpt-5.6-sol", "p": 0.46, "rationale": "My reference class is sampled networking switch ASICs with a public next-year product roadmap. I assign roughly a 0.44 annual hazard of a qualifying GA or volume announcement; over 15.8 months, compounding gives about 0.47 survival. The native-UALink requirement and current reliance on Ethernet bridges support slippage: silicon qualification, packaging, firmware, and customer validation often exceed a year. Conversely, the consortium targeted 2027 products, multiple members can trigger resolution, and a vendor need only publish qualifying production language—not prove broad deployment. Ambiguity against the forecaster trims my estimate to 0.46, below the stated 0.50." }, { "persona": "credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.45, "rationale": "I use sampled interconnect ASICs progressing to production announcements within 12–24 months as the reference class. A roughly 0.55 annual qualifying-announcement hazard over the remaining 15.8 months implies survival near 0.48. The consortium’s stated 2027 product target and multiple potential member vendors increase the chance that at least one issuer prints production language. Offsetting that, native fabrics commonly slip, and the rule excludes sampling, demonstrations, design wins, and Ethernet bridges while requiring an explicit member or consortium announcement. I therefore place the HIT probability slightly below the forecaster’s 0.50." }, { "persona": "geopolitics-policy-analyst", "model": "openai/gpt-5.6-sol", "p": 0.44, "rationale": "I use the reference class of sampled data-center switch ASICs approaching a publicly targeted product year: many reach production within 12–18 months, but schedule slips, qualification cycles, and ecosystem dependencies are common. Roughly 16 months remain, and multiple consortium members create several opportunities for a qualifying announcement. Conversely, the rule requires native UALink 200G silicon and explicit GA or production-shipping language; bridge products, demonstrations, and sampling do not count. The absence of usable leading-indicator data prevents raising confidence. The cheapness of issuing a production press release and ambiguity resolving against the forecaster put HIT modestly below even odds." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.46, "rationale": "Base rate: I use the supplied 0.50 reference-class prior for hard-date disclosure questions, implying roughly a 0.41 annual hazard of a qualifying announcement over the remaining 15.8 months. Sampling status and a public roadmap targeting products in 2027 raise that hazard, while multiple consortium members provide several opportunities to falsify this universal-negative claim. Conversely, the resolution requires native—not bridged—silicon and specific production language on an official site; actual shipment without that disclosure would not count. Missing indicators add no evidence. Ambiguity resolving against the forecaster trims the HIT probability further. No calibration table was supplied for an additional correction." } ], "aggregation": "log-odds-mean", "extremization": 1, "dissent": null, "spread": 0.03, "warnings": [ "all 6 votes came from one vendor (openai); extremization disabled, persona diversity only" ] }, "final": { "p": 0.47239849795619426, "ci80": [ 0.3722583083612603, 0.5748101521426079 ], "method": "stacked-fixed-weights (radar: prior 0.5, council 0.5; no resolved radar items to learn from)", "weights": { "council": 0.5, "prior": 0.5 } }, "stated": 0.5, "history": [ { "asOf": "2026-09-07", "p": 0.5075519470612959, "source": "radar-engine", "runId": "radar-2026-09-07T23-12-09Z-06a58f" }, { "asOf": "2026-09-07", "p": 0.47239849795619426, "source": "radar-engine", "runId": "radar-2026-09-07T23-49-04Z-de8abf" } ], "indicatorStatus": "no-data", "milestones": [], "arena": "networking-fabric", "status": "draft", "tier": "T1" }, "indicators": [ { "thesisId": "RT-08", "indicatorId": "li-1", "connector": "manual", "seriesId": "ualink-native-switch-sampling-count", "unit": "count", "asOf": "2026-09-07", "current": null, "status": "no-data", "history": [], "url": "https://ualinkconsortium.org/news/", "note": "manual: 'ualink-native-switch-sampling-count' not in manual-series.json" }, { "thesisId": "RT-08", "indicatorId": "li-2", "connector": "manual", "seriesId": "lq-p45-second-1-6t-design-win", "unit": "boolean", "asOf": "2026-09-07", "current": null, "status": "no-data", "history": [], "url": "https://ualinkconsortium.org/news/", "note": "manual: 'lq-p45-second-1-6t-design-win' not in manual-series.json" } ], "indicatorStatus": "no-data", "chainEvents": [ { "seq": 206, "ts": "2026-09-07", "kind": "repriced", "hash": "364391d8eba3a2a1fa125a8b8098a7c72943961228ab600b1772f9f1396c09bc" }, { "seq": 207, "ts": "2026-09-07", "kind": "repriced", "hash": "abf586857b7678ebc709918d5dd1c2dbeb6bf29f68baf77010089c5043f811d0" }, { "seq": 262, "ts": "2026-09-08", "kind": "registered", "hash": "e84837567049181a00fb67f49a98a941c2073007e6d21ab6ec722c6a4e463086" } ], "url": "/predictions/networking-fabric/RT-08" }, { "id": "RT-09", "title": "Co-packaged optics ships from three vendors by 2028", "arena": "networking-fabric", "tier": "T2", "tag": "CALL", "direction": "up", "status": "registered", "statement": "By 2028-12-31 at least three of NVIDIA, Broadcom, Cisco, Marvell, and Arista have announced general availability or volume production of a switch with co-packaged optics (CPO) on their own newsrooms, and at least one names a hyperscaler production deployment.", "resolutionRule": "Count vendors among {NVIDIA, Broadcom, Cisco, Marvell, Arista} that, on or before 2028-12-31, publish a press release on their own newsroom or investor-relations site stating a co-packaged-optics switch (optical engines integrated in the switch package, not pluggable transceivers) is \"generally available\", \"in full production\", \"in volume production\", or \"shipping in production\". Sampling, early access, demonstrations, and design wins do not count. Separately, at least one such release (or a hyperscaler's own newsroom) must name a specific hyperscaler (Microsoft, Google, Amazon, Meta, Oracle, or xAI) as running the CPO switch in production. Resolves HIT only if both the count >= 3 and the named deployment exist. Baseline September 2026: NVIDIA states Spectrum-X Photonics is in full production (one vendor); Broadcom Tomahawk 6 Davisson is at early access.", "resolutionSource": { "name": "NVIDIA newsroom (with Broadcom, Cisco, Marvell, Arista newsrooms)", "url": "https://nvidianews.nvidia.com/" }, "ambiguity": "resolves-against-forecaster", "horizon": "2028-12-31", "p": 0.55, "ci80": [ 0.4, 0.7 ], "consensus": { "value": "Vendor guidance treats 2026 as the first production year for CPO (NVIDIA Spectrum-X Photonics in full production August 2026; Broadcom Davisson early access), with pluggables expected to remain the volume default through 2028.", "impliedP": 0.45, "impliedBy": "Vendor schedules put NVIDIA in production now and Broadcom at volume in 2027, so two vendors by 2028 is the base case; the third vendor (Cisco, Marvell, or Arista) and a named hyperscaler production deployment are each roughly two-in-three on public roadmaps, giving about 0.45 for the conjunction.", "source": "NVIDIA Q2 FY2027 earnings call remarks on Spectrum-X Photonics", "url": "https://www.fool.com/earnings/call-transcripts/2026/08/26/nvidia-nvda-q2-2027-earnings-call-transcript/", "asOf": "2026-08-26", "note": "Optics market-share forecasts (LightCounting, Dell'Oro) are paywalled and cannot be cited for resolution. Honest coin flip: the conjunction (three vendors AND a named hyperscaler) is what makes this 0.55 rather than 0.75; I do not have a view on which of Cisco, Marvell, or Arista reaches GA first, so the number stays near consensus by design." }, "leadingIndicators": [ { "id": "li-1", "label": "Vendors with CPO switch in GA or volume production", "connector": "manual", "seriesId": "cpo-switch-ga-vendor-count", "unit": "count", "onTrack": { "op": ">=", "value": 2 }, "offTrack": { "op": "<=", "value": 1 }, "url": "https://nvidianews.nvidia.com/", "note": "Count per the resolution rule, checked against the five named newsrooms. Thresholds are for 2027-06-30." }, { "id": "li-2", "label": "Coherent quarterly revenue (optics demand proxy)", "connector": "sec-xbrl", "seriesId": "0000820318:RevenueFromContractWithCustomerExcludingAssessedTax:USD", "unit": "USD billions", "onTrack": { "op": ">=", "value": 2 }, "offTrack": { "op": "<", "value": 1.6 }, "url": "https://data.sec.gov/api/xbrl/companyconcept/CIK0000820318/us-gaap/RevenueFromContractWithCustomerExcludingAssessedTax.json", "note": "Coherent Corp quarterly revenue from XBRL as a proxy for datacenter optics volume (pluggables and CPO engines). Thresholds are for the quarter ending 2027-06. Falling optics revenue would signal a slower CPO ramp too." } ], "milestones": [ { "id": "m1", "date": "2026-09-30", "statement": "A second non-Broadcom vendor cites a 1.6T or co-packaged-optics hyperscaler design win.", "p": 0.55, "liveQuestionId": "lq-p45" }, { "id": "m2", "date": "2027-06-30", "statement": "Broadcom announces Tomahawk 6 Davisson (or successor CPO switch) in volume production.", "p": 0.65 }, { "id": "m3", "date": "2027-12-31", "statement": "A hyperscaler is publicly named as running a CPO switch in production.", "p": 0.6 }, { "id": "m4", "date": "2028-12-31", "statement": "Three of the five named vendors have CPO switches in GA or volume production.", "p": 0.55 } ], "falsifier": "On 2027-12-31 only one of the five named vendors has a CPO switch in GA or volume production and no hyperscaler production deployment has been named.", "whyItMatters": "Pluggable optics are roughly a fifth of a modern AI cluster's networking power budget and the leading field-failure item. If three vendors ship CPO by 2028, per-rack power and cooling assumptions for 2029 buildings change, and the transceiver supply chain that grew with the buildout loses its volume growth.", "whatWouldRaise": [ "Cisco or Marvell announce CPO switch sampling with a named hyperscaler before mid-2027.", "Microsoft or Meta publish an engineering blog on CPO in production.", "TSMC COUPE packaging capacity announcements exceed 2027 demand." ], "whatWouldCut": [ "Field reliability issues (laser failures, serviceability) reported publicly on first-generation CPO.", "1.6T pluggable pricing falls faster than CPO cost, removing the economic case.", "Broadcom delays Davisson volume production past 2027." ], "namedEntities": [ "NVIDIA", "Broadcom", "Cisco", "Marvell", "Arista", "Microsoft", "Google", "Amazon", "Meta", "Oracle", "xAI" ], "marketCall": false, "recusalCheck": { "passed": true, "keysChecked": [ "employer-leasing", "named-competitor-bookings", "customer-pipeline", "ticker" ] }, "signedBy": "Brian Letort", "registeredAt": "2026-09-08", "amendments": [], "ruleHash": "00aa87338a447ebdbca086d4b2eb237675e6233ce53bc5585b765e3e8dc80ac8", "distance": 0.40134139092430265, "distanceLabel": "leans above consensus", "engine": { "id": "RT-09", "prior": { "p": 0.98, "method": "sec-xbrl-loglinear-bootstrap:li-2", "note": "log-linear growth +0.157/yr on the last 16 points, extrapolated from 2026-03-31; P(li-2 satisfies onTrack >= 2.0 at 2028-12-31); proxy for the thesis, not its rule" }, "council": { "p": 0.3308335274283271, "votes": [ { "persona": "grid-and-power-planner", "model": "openai/gpt-5.6-sol", "p": 0.3, "rationale": "Start from one qualifying vendor. Using the reference class of early-access networking products reaching vendor-declared volume production, I assign Broadcom about 0.80 by the deadline, but only about 0.50 that at least one of Cisco, Marvell, or Arista issues the exact qualifying CPO-switch announcement within 27 months. Including paths where Broadcom misses but two others ship puts the three-vendor leg near 0.52. Publicly naming a hyperscaler as running the switch in production is an additional disclosure hurdle; conditional on three vendors, I put it near 0.58. That gives roughly 0.30 overall. The revenue-based structural prior does not measure CPO qualification, release wording, or customer disclosure." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.38, "rationale": "Reference class: leading-edge networking platforms commonly require 4–8 quarters to move from early access to production, with system-level CPO adding packaging, thermal, fiber-attach, and qualification risk. NVIDIA already satisfies one slot, and Broadcom is more likely than not to become the second by 2028. The third slot is less secure: Cisco, Marvell, or Arista must both commercialize a qualifying switch and use the resolution’s explicit production language on an official site. I estimate roughly 0.58 for the three-vendor count. Publicly naming a hyperscaler production deployment is a separate disclosure hurdle; conditional on three vendors, I assign about 0.65. The supplied revenue trend is not a meaningful CPO commercialization prior. Their conjunction yields about 0.38." }, { "persona": "enterprise-cio", "model": "openai/gpt-5.6-sol", "p": 0.31, "rationale": "Reference class: merchant-switch and optical-hardware transitions from early access to production. Over the remaining 2.3 years, I use roughly a 0.45 annual qualifying hazard for Broadcom and 0.12–0.18 for each of Cisco, Marvell, and Arista, discounted for correlated schedules and exact-wording risk. With NVIDIA already qualifying, that puts the three-vendor leg near 0.50. A specifically named hyperscaler production deployment is a separate, positively correlated but disclosure-constrained leg; customer confidentiality and strict “production” wording reduce the joint outcome to about 0.31. The revenue series and 0.98 structural prior are not causally informative about CPO qualification. The stated 0.55 underweights conjunction and documentation risk." }, { "persona": "credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.38, "rationale": "The relevant reference class is datacenter-networking products advancing from early access or demonstrations to explicitly documented production within two to three years. NVIDIA already qualifies, and Broadcom has a credible path from early access, but the thesis still needs a qualifying release from Cisco, Marvell, or Arista. Annual-hazard compounding provides time for that transition, yet exact newsroom wording and vendor-level system announcements create resolution risk. The named-hyperscaler production disclosure is an additional, correlated bottleneck because customers may deploy without permitting identification. The supplied revenue trend does not measure CPO commercialization and should not support the 0.98 structural prior. These conjunctive requirements put the estimate below the stated 0.55." }, { "persona": "geopolitics-policy-analyst", "model": "openai/gpt-5.6-sol", "p": 0.39, "rationale": "I use the reference class of networking hardware moving from early access to qualified volume production within roughly 2–3 years. NVIDIA already satisfies one slot, and Broadcom is a plausible second, but the thesis still requires Cisco, Marvell, or Arista to issue an expressly qualifying production announcement by end-2028. CPO’s operational complexity and continued preference for pluggables lower that transition rate. The separate customer-disclosure condition is material: hyperscalers often avoid naming production architectures, and design wins or trials do not qualify. Treating vendor count and named deployment as correlated but conjunctive yields a probability below the forecaster’s 0.55. The revenue-series structural prior is not technically specific to CPO and receives little weight." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.24, "rationale": "My reference-class base rate is 0.40 for a newly shipping switch technology to gain two additional qualifying vendors within roughly 28 months. NVIDIA already satisfies one slot and Broadcom’s early access improves the count leg, but Cisco, Marvell, or Arista must still cross the rule’s strict production threshold. I assign about 0.50 conditional probability that qualifying maturity also produces a timely newsroom disclosure naming a hyperscaler production user. Customer confidentiality and exact-source wording make disclosure harder than deployment itself. The generic revenue series and its 0.98 structural prior are not product-specific evidence. No calibration table was supplied for further correction." } ], "aggregation": "log-odds-mean", "extremization": 1, "dissent": null, "spread": 0.15, "warnings": [ "all 6 votes came from one vendor (openai); extremization disabled, persona diversity only" ] }, "final": { "p": 0.83113622291762, "ci80": [ 0.6796260652561734, 0.8547314962868717 ], "method": "stacked-fixed-weights (radar: prior 0.5, council 0.5; no resolved radar items to learn from)", "weights": { "council": 0.5, "prior": 0.5 } }, "stated": 0.55, "history": [ { "asOf": "2026-09-07", "p": 0.8276524277315835, "source": "radar-engine", "runId": "radar-2026-09-07T23-12-09Z-06a58f" }, { "asOf": "2026-09-07", "p": 0.83113622291762, "source": "radar-engine", "runId": "radar-2026-09-07T23-49-04Z-de8abf" } ], "indicatorStatus": "mixed", "milestones": [ { "id": "m1", "enginePrior": 0.3446075428814045, "councilP": 0.15642706295357858 }, { "id": "m2", "enginePrior": 0.5, "councilP": null }, { "id": "m3", "enginePrior": 0.5, "councilP": null }, { "id": "m4", "enginePrior": null, "councilP": null } ], "arena": "networking-fabric", "status": "draft", "tier": "T2" }, "indicators": [ { "thesisId": "RT-09", "indicatorId": "li-1", "connector": "manual", "seriesId": "cpo-switch-ga-vendor-count", "unit": "count", "asOf": "2026-09-07", "current": null, "status": "no-data", "history": [], "url": "https://nvidianews.nvidia.com/", "note": "manual: 'cpo-switch-ga-vendor-count' not in manual-series.json" }, { "thesisId": "RT-09", "indicatorId": "li-2", "connector": "sec-xbrl", "seriesId": "0000820318:RevenueFromContractWithCustomerExcludingAssessedTax:USD", "unit": "USD billions", "asOf": "2026-03-31", "current": 1.805641, "status": "between", "history": [ { "asOf": "2016-09-30", "value": 0.22152000000000002 }, { "asOf": "2016-12-31", "value": 0.23182200000000003 }, { "asOf": "2017-03-31", "value": 0.244987 }, { "asOf": "2017-06-30", "value": 0.27371700000000004 }, { "asOf": "2017-09-30", "value": 0.26150300000000004 }, { "asOf": "2017-12-31", "value": 0.28147 }, { "asOf": "2018-03-31", "value": 0.294746 }, { "asOf": "2018-06-30", "value": 0.321075 }, { "asOf": "2018-09-30", "value": 0.314433 }, { "asOf": "2018-12-31", "value": 0.342839 }, { "asOf": "2019-03-31", "value": 0.342496 }, { "asOf": "2019-06-30", "value": 0.362728 }, { "asOf": "2019-09-30", "value": 0.340409 }, { "asOf": "2019-12-31", "value": 0.666331 }, { "asOf": "2020-03-31", "value": 0.6270410000000001 }, { "asOf": "2020-06-30", "value": 0.74629 }, { "asOf": "2020-09-30", "value": 0.7280840000000001 }, { "asOf": "2020-12-31", "value": 0.7865690000000001 }, { "asOf": "2021-03-31", "value": 0.783232 }, { "asOf": "2021-06-30", "value": 0.808006 }, { "asOf": "2021-09-30", "value": 0.795111 }, { "asOf": "2021-12-31", "value": 0.8068190000000001 }, { "asOf": "2022-03-31", "value": 0.827724 }, { "asOf": "2022-09-30", "value": 1.34457 }, { "asOf": "2022-12-31", "value": 1.370285 }, { "asOf": "2023-03-31", "value": 1.240194 }, { "asOf": "2023-09-30", "value": 1.053083 }, { "asOf": "2023-12-31", "value": 1.131434 }, { "asOf": "2024-03-31", "value": 1.208809 }, { "asOf": "2024-09-30", "value": 1.348135 }, { "asOf": "2024-12-31", "value": 1.434665 }, { "asOf": "2025-03-31", "value": 1.4978790000000002 }, { "asOf": "2025-09-30", "value": 1.5813780000000002 }, { "asOf": "2025-12-31", "value": 1.685629 }, { "asOf": "2026-03-31", "value": 1.805641 } ], "url": "https://data.sec.gov/api/xbrl/companyconcept/CIK0000820318/us-gaap/RevenueFromContractWithCustomerExcludingAssessedTax.json", "note": "quarterly frames; raw USD facts scaled by 1e-09 to match unit 'USD billions'" } ], "indicatorStatus": "mixed", "chainEvents": [ { "seq": 208, "ts": "2026-09-07", "kind": "repriced", "hash": "f543399112ca3aedeb07032e23d3965a935025bc7960099bd178d24bae38044d" }, { "seq": 209, "ts": "2026-09-07", "kind": "repriced", "hash": "d24ae265f169b49e2fdfaccd1a6ac76c130a7d1ae3a21498b3f4a9e1d14025c0" }, { "seq": 263, "ts": "2026-09-08", "kind": "registered", "hash": "a786545a7cee128c7607832650d51dc99bd6a777fda8db62fd8a4c64371fb5ff" } ], "url": "/predictions/networking-fabric/RT-09" }, { "id": "RT-10", "title": "Open scale-up fabrics run in two hyperscalers by 2031", "arena": "networking-fabric", "tier": "T3", "tag": "THESIS", "direction": "shift", "status": "registered", "statement": "By 2031-12-31 at least two of Amazon, Microsoft, Meta, and Oracle publicly disclose a production AI training or inference deployment whose scale-up domain of 64 or more accelerators is connected by an open-standard fabric (UALink or an Ethernet scale-up specification published by UEC or the Open Compute Project) rather than NVLink or a proprietary in-house interconnect.", "resolutionRule": "Resolves HIT if, on or before 2031-12-31, at least two of the four named companies publish on their own newsroom, engineering blog, investor site, or a conference paper with a company author a statement that a production (not pilot or lab) deployment uses UALink (native or the UALink protocol carried over Ethernet), UEC Scale-Up Ethernet, or an OCP-published scale-up specification to connect a scale-up domain of at least 64 accelerators. Google is excluded because its ICI fabric is proprietary and would not test the thesis. NVLink Fusion deployments and undisclosed in-house protocols do not count as open. Press reports without a company statement do not count. If the UALink Consortium and UEC both cease publishing specifications before 2029-12-31, resolves MISS.", "resolutionSource": { "name": "Hyperscaler newsrooms and engineering blogs (Amazon, Microsoft, Meta, Oracle)", "url": "https://engineering.fb.com/" }, "ambiguity": "resolves-against-forecaster", "horizon": "2031-12-31", "p": 0.75, "ci80": [ 0.6, 0.88 ], "consensus": { "value": "NVIDIA management describes NVLink and NVLink Fusion as the scale-up standard and reports Spectrum-X Ethernet revenue up 2.6x year over year; the UALink Consortium roadmap targets products in 2027 with no hyperscaler production commitment disclosed as of September 2026.", "impliedP": 0.4, "impliedBy": "With no hyperscaler having disclosed an open scale-up deployment and the incumbent positioning NVLink Fusion as the path for custom silicon, the consensus path has open fabrics reaching one hyperscaler in production by 2029 with roughly even odds and a second by 2031 conditional on the first at ~0.8, giving about 0.40.", "source": "NVIDIA Q2 FY2027 earnings call and UALink Consortium news", "url": "https://ualinkconsortium.org/news/", "asOf": "2026-09-07", "note": "No prediction market covers scale-up fabric adoption. My 0.75 rests on AMD's Helios rack (72 GPUs, UALink protocol over Ethernet) already being ordered by Oracle and OpenAI for 2026-2027, so one qualifying disclosure is close to certain and the bet is really on a second hyperscaler by 2031." }, "leadingIndicators": [ { "id": "li-1", "label": "Hyperscalers disclosing open scale-up fabric in production", "connector": "manual", "seriesId": "open-scaleup-fabric-hyperscaler-count", "unit": "count", "onTrack": { "op": ">=", "value": 1 }, "offTrack": { "op": "==", "value": 0 }, "url": "https://ualinkconsortium.org/news/", "note": "Count per the resolution rule across the four named companies. Thresholds are for 2029-06-30." }, { "id": "li-2", "label": "Native UALink switch GA (from RT-08)", "connector": "manual", "seriesId": "ualink-native-switch-ga", "unit": "boolean", "onTrack": { "op": "==", "value": 1 }, "offTrack": { "op": "==", "value": 0 }, "url": "https://ualinkconsortium.org/news/", "note": "1 once any member announces GA or volume production of a native UALink switch (the RT-08 resolution event). Threshold is for 2028-06-30." } ], "milestones": [ { "id": "m1", "date": "2027-12-31", "statement": "A UALink Consortium member announces GA or volume production of a native UALink switch ASIC.", "p": 0.5 }, { "id": "m2", "date": "2029-06-30", "statement": "One of Amazon, Microsoft, Meta, or Oracle discloses a production deployment on an open scale-up fabric with 64 or more accelerators.", "p": 0.8 }, { "id": "m3", "date": "2031-12-31", "statement": "Two of the four named hyperscalers have disclosed such deployments.", "p": 0.75 } ], "falsifier": "On 2029-12-31 none of Amazon, Microsoft, Meta, or Oracle has disclosed a production open-standard scale-up deployment of 64 or more accelerators; the thesis is then withdrawn as falsified.", "whyItMatters": "Scale-up interconnect is the last proprietary layer in the AI rack. If open fabrics reach production at two hyperscalers, custom accelerators can be mixed into the same domain and the rack stops being a single-vendor appliance. That changes procurement, cooling design, and the residual value of GPU-backed collateral. If it does not happen by 2031, NVLink lock-in should be treated as structural for the decade.", "whatWouldRaise": [ "AMD Helios or a Trainium rack ships with a native UALink switch in 2028.", "Meta or Microsoft publish an OCP contribution for a scale-up Ethernet rack.", "Broadcom reports scale-up Ethernet switch revenue as a distinct line." ], "whatWouldCut": [ "Two or more custom-ASIC programs adopt NVLink Fusion for scale-up.", "UALink native silicon slips past 2028.", "Hyperscalers standardize on proprietary in-house fabrics and publish nothing." ], "namedEntities": [ "Amazon", "Microsoft", "Meta", "Oracle", "UALink Consortium", "Ultra Ethernet Consortium", "Open Compute Project", "NVIDIA" ], "marketCall": false, "recusalCheck": { "passed": true, "keysChecked": [ "employer-leasing", "named-competitor-bookings", "customer-pipeline", "ticker" ] }, "signedBy": "Brian Letort", "registeredAt": "2026-09-08", "amendments": [], "ruleHash": "22110c4d163dc751d9916f494890d717383a7bb4c62a12f4a1812f633178a26a", "distance": 1.5040773967762737, "distanceLabel": "far above consensus", "engine": { "id": "RT-10", "prior": null, "council": { "p": 0.5218210588691014, "votes": [ { "persona": "grid-and-power-planner", "model": "openai/gpt-5.6-sol", "p": 0.58, "rationale": "Reference class: consortium interconnects attempting to displace an incumbent proprietary fabric in hyperscale production. From 2027 product availability to 2031, I assign roughly a 0.18 annual hazard for each of Meta and Microsoft to make a qualifying disclosure, lower for Amazon, with strongly correlated adoption; Oracle’s announced Helios path gives it about a 0.70 chance of becoming the first qualifier. Qualification and disclosure penalties matter: orders are not production, “Ethernet” alone does not prove use of a published scale-up specification, and first-party posts may omit the 64-accelerator domain. Combining ecosystem-success and failure branches yields 0.58. Consortium cessation before 2029 adds a small miss risk." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.56, "rationale": "Reference class: new datacenter interconnect standards typically need one to two accelerator generations after initial silicon before documented production adoption. Oracle is a plausible first qualifier because the cited Helios configuration exceeds 64 accelerators, but resolution still requires Oracle’s own production disclosure explicitly identifying the qualifying fabric. A second adopter among Meta, Microsoft, and Amazon is materially less certain: proprietary fabrics, NVLink Fusion, schedule slips, and disclosures omitting domain size or protocol all cause misses. The 2029 falsifier further excludes scenarios where adoption arrives only late. Shared ecosystem progress creates positive correlation, but the conjunctive two-company and documentation requirements make 0.75 too high." }, { "persona": "enterprise-cio", "model": "openai/gpt-5.6-sol", "p": 0.44, "rationale": "My reference class is open server/interconnect standards progressing from first silicon to specifically documented hyperscale production within roughly five years. That transition often takes two to four procurement cycles; named production disclosures lag pilots. The supplied Helios/Oracle claim materially raises the chance of one qualifier, but an order is not proof of production, fabric semantics, or a 64-accelerator scale-up domain. A second named company is a conjunctive hurdle: AWS has a proprietary path, while Microsoft and Meta may deploy open technology without publishing every required detail. No indicator data and only 0.5 stated odds for 2027 native-switch GA add schedule risk. Moderate annual adoption hazards still leave substantial risk of only one disclosure or none by the 2029 falsifier." }, { "persona": "credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.54, "rationale": "Reference class: five-year adoption of new datacenter interconnect standards from first silicon to customer-attributed production disclosure. The supplied Oracle–Helios path makes one qualifying deployment plausible before the 2029 falsifier, but ordering is not the required company-authored production disclosure. A second named company is a meaningful conjunction: products arrive no earlier than 2027, integration takes time, and NVLink plus proprietary fabrics remain viable. Exact requirements—64 accelerators, scale-up use, open specification, production status and approved publication channel—create substantial documentation risk. Correlated ecosystem delay could prevent both disclosures. This is materially below 0.75." }, { "persona": "geopolitics-policy-analyst", "model": "openai/gpt-5.6-sol", "p": 0.58, "rationale": "My reference class is new interconnect standards progressing from specification to publicly disclosed hyperscaler production within four to five years. I assign each non-Oracle company roughly a 0.08–0.12 annual qualifying-disclosure hazard after viable products emerge, with Oracle higher given the supplied Helios evidence. Compounding creates a credible two-company path, but adoption outcomes are correlated: switch delays, weak software support, or continued NVLink/proprietary use could affect several companies simultaneously. The requirement for production, 64 accelerators, an explicitly named open specification, and a company-authored disclosure materially lowers resolution odds. Neither leading indicator confirms switch availability or a qualifying deployment, making 0.75 too high." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.43, "rationale": "Base rate: 0.32, derived from an 0.08 annual per-company hazard for a nascent hardware standard to reach qualifying production adoption and primary disclosure over roughly four post-product years, then requiring at least two of four companies. I raise to 0.43 because the supplied Oracle/Helios order gives UALink a credible first path and standardization could create correlated uptake. However, the indicators show no qualifying deployment, switch GA, or production commitment; products start no earlier than 2027, while proprietary and NVLink alternatives remain strong. The claim requires two adopters, production, 64+ accelerators, an eligible specification, and timely company-authored disclosure. The 2029 falsifier further makes 0.75 overconfident." } ], "aggregation": "log-odds-mean", "extremization": 1, "dissent": null, "spread": 0.15, "warnings": [ "all 6 votes came from one vendor (openai); extremization disabled, persona diversity only" ] }, "final": { "p": 0.5218210588691014, "ci80": [ 0.4335420911352798, 0.6087577131656302 ], "method": "council-only:log-odds-mean", "weights": { "council": 1 } }, "stated": 0.75, "history": [ { "asOf": "2026-09-07", "p": 0.5048537242113821, "source": "radar-engine", "runId": "radar-2026-09-07T23-12-09Z-06a58f" }, { "asOf": "2026-09-07", "p": 0.5218210588691014, "source": "radar-engine", "runId": "radar-2026-09-07T23-49-04Z-de8abf" } ], "indicatorStatus": "no-data", "milestones": [ { "id": "m1", "enginePrior": 0.3888888888888889, "councilP": null }, { "id": "m2", "enginePrior": null, "councilP": null }, { "id": "m3", "enginePrior": null, "councilP": null } ], "arena": "networking-fabric", "status": "draft", "tier": "T3" }, "indicators": [ { "thesisId": "RT-10", "indicatorId": "li-1", "connector": "manual", "seriesId": "open-scaleup-fabric-hyperscaler-count", "unit": "count", "asOf": "2026-09-07", "current": null, "status": "no-data", "history": [], "url": "https://ualinkconsortium.org/news/", "note": "manual: 'open-scaleup-fabric-hyperscaler-count' not in manual-series.json" }, { "thesisId": "RT-10", "indicatorId": "li-2", "connector": "manual", "seriesId": "ualink-native-switch-ga", "unit": "boolean", "asOf": "2026-09-07", "current": null, "status": "no-data", "history": [], "url": "https://ualinkconsortium.org/news/", "note": "manual: 'ualink-native-switch-ga' not in manual-series.json" } ], "indicatorStatus": "no-data", "chainEvents": [ { "seq": 210, "ts": "2026-09-07", "kind": "repriced", "hash": "0c16e491dc2fba3a8e01dbb8c19bea3ac412a9ab47d0ffba890a5d2248474b48" }, { "seq": 211, "ts": "2026-09-07", "kind": "repriced", "hash": "11e33634cacfb93b21077b38e11b6f9671c8b66b182ef4734fe2736f2c750bc4" }, { "seq": 264, "ts": "2026-09-08", "kind": "registered", "hash": "52861e09e5c8bbf130dff2d2d9515b88f9315e5bce8719b6b0427536eeae0507" } ], "url": "/predictions/networking-fabric/RT-10" }, { "id": "RT-11", "title": "Open weights stay four points behind closed frontier", "arena": "model-frontier", "tier": "T1", "tag": "WARNING", "direction": "shift", "status": "registered", "statement": "On 2027-12-31 the gap between the top-scoring model and the top-scoring open-weights model on the Artificial Analysis Intelligence Index is at least 4 points on the index version then live.", "resolutionRule": "On 2027-12-31 (or the first day after on which the page loads), open the Artificial Analysis Intelligence Index leaderboard (artificialanalysis.ai, models leaderboard, default view, all models). Record the highest index score among all models (S_all) and the highest score among models Artificial Analysis labels open weights (S_open, the open-weights filter or the /models/open-source view). Resolves HIT if S_all - S_open >= 4.0 on the index version then displayed. Version changes are expected and do not void the question; the check is made on whatever version is live that day. Preview or unreleased models count only if Artificial Analysis lists them with a score. If the index is discontinued or the open-weights label is removed, resolves MISS. Baseline 2026-09-07 (leaderboard default view): S_all 66 (Claude Fable 5.1, max effort), S_open 60 (Kimi K3 max, GLM-5.3 max), gap 6; a year earlier the gap was about 13.", "resolutionSource": { "name": "Artificial Analysis Intelligence Index leaderboard", "url": "https://artificialanalysis.ai/leaderboards/models" }, "ambiguity": "resolves-against-forecaster", "horizon": "2027-12-31", "p": 0.6, "ci80": [ 0.45, 0.75 ], "consensus": { "value": "Artificial Analysis's index shows the open/closed gap falling from ~13 points to 6 in twelve months; FutureSearch (mid-2026) put a 35-45% chance on an open-weight model temporarily reaching #1 during 2027 and 13% on holding #1 for a quarter of the year. A continuation of the halving trend implies a gap near 3 points at end-2027.", "impliedP": 0.35, "impliedBy": "If the gap halves again over the next fifteen months, the central estimate on 2027-12-31 is about 3 points; with FutureSearch's 35-45% chance that an open model briefly reaches #1 (gap zero or negative) in 2027, the probability that the snapshot still shows 4 or more points is roughly one in three.", "source": "FutureSearch, \"Will an open-weight LLM be the single top-ranked model ... for at least 25% of days in 2027?\" and Artificial Analysis open-weights page", "url": "https://futuresearch.ai/app/p/a/open-weight-llm-overtaking-closed-weight", "asOf": "2026-09-07", "note": "The FutureSearch question uses a days-at-#1 framing, so its 13% is not the probability for this statement; impliedP converts its gap reasoning to the point-gap snapshot. My 0.60 is above it because the two labs at the top ship max-effort reasoning variants that they do not open-weight, and the index keeps re-weighting toward agentic evals where those variants lead." }, "leadingIndicators": [ { "id": "li-1", "label": "Top open-weights Intelligence Index score", "connector": "artificial-analysis", "seriesId": "index:open-weights-top-intelligence", "unit": "index points", "onTrack": { "op": "<=", "value": 64 }, "offTrack": { "op": ">", "value": 67 }, "url": "https://artificialanalysis.ai/models/open-source", "note": "Highest index score among open-weights models on the live index version. Thresholds are for 2027-03-31 and assume the top overall score is near 70 by then; the engine should evaluate the gap, not the raw level, when both are available." }, { "id": "li-2", "label": "Top overall Intelligence Index score", "connector": "artificial-analysis", "seriesId": "index:top-intelligence", "unit": "index points", "onTrack": { "op": ">=", "value": 69 }, "offTrack": { "op": "<", "value": 67 }, "url": "https://artificialanalysis.ai/leaderboards/models", "note": "Highest index score among all models on the live index version, 2027-03-31. A stalled frontier lets open models close the gap regardless of their own pace." }, { "id": "li-3", "label": "Hugging Face downloads, leading Chinese open-weights orgs", "connector": "hf-hub", "seriesId": "downloads:Qwen", "unit": "downloads per month", "onTrack": { "op": "<", "value": 60000000 }, "offTrack": { "op": ">=", "value": 100000000 }, "url": "https://huggingface.co/Qwen", "note": "Sum of last-30-day downloads across models under the Qwen organization, as shown on the Hub. Thresholds are for 2027-03-31; a surge signals open models closing the gap in practice." } ], "milestones": [], "falsifier": "On 2027-12-31 the top open-weights model scores within 3.9 points of the top overall model on the live Artificial Analysis index.", "whyItMatters": "Enterprises deciding between hosted frontier APIs and self-hosted open weights are pricing that choice on a gap that has halved in a year. If the gap holds at four or more points, the closed frontier keeps its pricing power and inference stays concentrated in a few clouds. If it closes, on-premises and sovereign inference becomes the default for most workloads and the demand map for capacity shifts.", "whatWouldRaise": [ "Two frontier labs ship new flagship reasoning models in 2027 without open counterparts.", "Chinese labs face tighter export limits on training compute, slowing release cadence.", "Artificial Analysis raises index difficulty (new version) in a way that widens gaps at the top." ], "whatWouldCut": [ "DeepSeek, Qwen, Moonshot, or Z.ai release a model within 2 points of the top overall score before mid-2027.", "A US lab open-weights a model within one generation of its flagship.", "The top overall score stalls below 68 through 2027." ], "namedEntities": [ "Artificial Analysis" ], "marketCall": false, "recusalCheck": { "passed": true, "keysChecked": [ "employer-leasing", "named-competitor-bookings", "customer-pipeline", "ticker" ] }, "signedBy": "Brian Letort", "registeredAt": "2026-09-08", "amendments": [], "ruleHash": "fca6ce012b0fcde9833c5cd20c052f3eed70eb73e1bd6e22d7fe2a171f41f644", "distance": 1.0245043165143877, "distanceLabel": "clearly above consensus", "engine": { "id": "RT-11", "prior": { "p": 0.65, "method": "reference-class:qtype:threshold", "note": "ledger base rate, n=19, horizon 480d; the ledger has no multi-year history" }, "council": { "p": 0.5600406417448298, "votes": [ { "persona": "grid-and-power-planner", "model": "openai/gpt-5.6-sol", "p": 0.55, "rationale": "My reference class is fixed-date, one-year frontier-leaderboard snapshots: open models usually trail, but gaps move sharply with lumpy release cycles. Starting at 6 points favors HIT, and closed labs’ max-effort variants provide a structural advantage. Against that, the supplied gap compressed from about 13 to 6 in one year; even partial continuation crosses the 4-point threshold. Qwen downloads do not measure frontier capability and receive little weight. Over the 16-month horizon, I estimate a 0.57 conditional HIT probability, reduced to 0.55 for the small compounded hazard of leaderboard discontinuation or label removal, both automatic misses. This is below the structural prior and forecaster’s 0.60." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.57, "rationale": "I start from the supplied threshold-question reference-class prior of 0.65. The current 6-point gap provides only a 2-point cushion, while its decline from roughly 13 in one year suggests meaningful catch-up risk across several release cycles before the snapshot. I adjust upward for closed labs’ persistent advantage in proprietary max-effort reasoning and agentic systems, which need not transfer into released weights. I adjust downward for rapid open-model diffusion, volatile leaderboard timing, index reweighting, and the resolves-against-forecaster discontinuation/label risk. Qwen downloads indicate ecosystem strength but are weak evidence about frontier index scores. Overall, a gap of at least 4 remains slightly more likely than not." }, { "persona": "enterprise-cio", "model": "openai/gpt-5.6-sol", "p": 0.56, "rationale": "I use the reference class of 12–18 month persistence of a narrow benchmark lead during a fast platform race, starting from the supplied structural prior of 0.65. The current six-point gap and closed labs’ advantage in costly max-effort variants support a HIT. Against that, the margin over the four-point threshold is only two points, while the reported gap compressed from about 13 to six in one year. Over the remaining 16 months, compounding continued catch-up and leaderboard/version volatility makes crossing below four materially likely. Qwen downloads measure distribution, not frontier index performance, so carry little weight. I land slightly below the stated 0.60." }, { "persona": "credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.58, "rationale": "Using frontier-leaderboard gaps across release cycles as the reference class, the supplied six-point baseline gives the claim a modest cushion, while the fall from roughly 13 points shows rapid convergence. Over the 1.31-year horizon, I estimate about a 0.30 annual hazard of open weights compressing the gap below four; compounding and allowing for subsequent closed-model releases leaves the endpoint close to balanced. Closed labs’ proprietary max-effort variants support persistence, but release timing, index reweighting, and the resolves-MISS discontinuation clause reduce confidence. The Qwen download reading is only a weak capability proxy. My estimate is slightly below both the 0.60 forecast and 0.65 structural prior." }, { "persona": "geopolitics-policy-analyst", "model": "openai/gpt-5.6-sol", "p": 0.56, "rationale": "I anchor on the supplied 0.65 threshold-forecast reference-class prior, then lower it because the observed gap compressed from roughly 13 to 6 in one year. Another 16 months gives open-weight labs multiple release opportunities, and a continuation—though likely slower—would cross below the 4-point threshold. Offsetting this, frontier gaps are sawtoothed: closed labs can restore a lead with proprietary max-effort releases, and the fixed year-end snapshot may catch such a cycle. Qwen downloads indicate broad open-model adoption but weakly predict frontier score parity. Index redesign, discontinuation, or label removal adds a small miss risk under the strict rule." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.54, "rationale": "Base rate: the supplied reference class for threshold questions is 0.65. I adjust downward because the observed gap reportedly fell from about 13 to 6 in one year, leaving only a 2-point cushion above the threshold with 16 months remaining. Benchmark reweighting and the exact-date snapshot add substantial variance, while discontinuation or removal of the label automatically causes a miss. I adjust upward slightly because open weights need to close more than 2 additional points, and the current closed leaders’ max-effort variants may preserve differentiation. Hugging Face downloads are not a reliable index-performance indicator. The forecaster’s 0.60 looks modestly high; no calibration table was supplied to justify a further correction." } ], "aggregation": "log-odds-mean", "extremization": 1, "dissent": null, "spread": 0.04, "warnings": [ "all 6 votes came from one vendor (openai); extremization disabled, persona diversity only" ] }, "final": { "p": 0.6059173912481528, "ci80": [ 0.49213532516982356, 0.7177634245244716 ], "method": "stacked-fixed-weights (radar: prior 0.5, council 0.5; no resolved radar items to learn from)", "weights": { "council": 0.5, "prior": 0.5 } }, "stated": 0.6, "history": [ { "asOf": "2026-09-07", "p": 0.6026806787958939, "source": "radar-engine", "runId": "radar-2026-09-07T23-12-09Z-06a58f" }, { "asOf": "2026-09-07", "p": 0.6059173912481528, "source": "radar-engine", "runId": "radar-2026-09-07T23-49-04Z-de8abf" } ], "indicatorStatus": "off-track", "milestones": [], "arena": "model-frontier", "status": "draft", "tier": "T1" }, "indicators": [ { "thesisId": "RT-11", "indicatorId": "li-1", "connector": "artificial-analysis", "seriesId": "index:open-weights-top-intelligence", "unit": "index points", "asOf": "2026-09-07", "current": null, "status": "no-data", "history": [], "url": "https://artificialanalysis.ai/leaderboards/models", "note": "ARTIFICIAL_ANALYSIS_API_KEY not set; the data API answers 401 and the leaderboard is JS-rendered. Enter readings under 'artificial-analysis:' in data/manual-series.json with a public source" }, { "thesisId": "RT-11", "indicatorId": "li-2", "connector": "artificial-analysis", "seriesId": "index:top-intelligence", "unit": "index points", "asOf": "2026-09-07", "current": null, "status": "no-data", "history": [], "url": "https://artificialanalysis.ai/leaderboards/models", "note": "ARTIFICIAL_ANALYSIS_API_KEY not set; the data API answers 401 and the leaderboard is JS-rendered. Enter readings under 'artificial-analysis:' in data/manual-series.json with a public source" }, { "thesisId": "RT-11", "indicatorId": "li-3", "connector": "hf-hub", "seriesId": "downloads:Qwen", "unit": "downloads per month", "asOf": "2026-09-07", "current": 317585926, "status": "off-track", "history": [ { "asOf": "2026-09-07", "value": 317585926 } ], "url": "https://huggingface.co/Qwen", "note": "rolling 30-day downloads" } ], "indicatorStatus": "off-track", "chainEvents": [ { "seq": 212, "ts": "2026-09-07", "kind": "repriced", "hash": "6e284d37c1eceb78c32b92de22d9b0aa44771842ee6227100a955147c9b534e9" }, { "seq": 213, "ts": "2026-09-07", "kind": "repriced", "hash": "db897fc156c0ee9c2fac2ac72fc6f2e8bf324455d5475d25305a442006c850cc" }, { "seq": 265, "ts": "2026-09-08", "kind": "registered", "hash": "48463726f35cdbc8318c50b84b7de23a4fe26306340076c6f60d47aab25c5190" } ], "url": "/predictions/model-frontier/RT-11" }, { "id": "RT-12", "title": "The frontier tier keeps its price through 2028", "arena": "model-frontier", "tier": "T2", "tag": "THESIS", "direction": "up", "status": "registered", "statement": "On 2028-12-31 the top-ranked model on the Artificial Analysis Intelligence Index carries a list output price of at least $20 per million tokens on its first-party API, even as trailing-tier prices keep falling.", "resolutionRule": "On 2028-12-31 (or the first day after on which the page loads), identify the model with the highest score on the Artificial Analysis Intelligence Index (ties: any tied model qualifies if it meets the price test). Take the USD output price per million tokens shown by Artificial Analysis for that model's first-party endpoint (the model developer's own API, standard tier, not batch or cached). Resolves HIT if the price is >= $20.00. If the top model has no public API price (unreleased or private), the next-highest model with a public first-party price is used. If the index is discontinued, resolves MISS. Baseline 2026-09-07: Claude Fable 5.1 (max effort) leads the default leaderboard at 66 and lists at $10 input / $50 output per million tokens; GPT-6 Astra lists at the same $10 / $50.", "resolutionSource": { "name": "Artificial Analysis Intelligence Index leaderboard (score and pricing columns)", "url": "https://artificialanalysis.ai/leaderboards/models" }, "ambiguity": "resolves-against-forecaster", "horizon": "2028-12-31", "p": 0.7, "ci80": [ 0.55, 0.85 ], "consensus": { "value": "a16z \"LLMflation\" (Nov 2024): inference cost at constant capability falls about 10x per year; Epoch AI's inference price trend page shows 9-900x annual declines at fixed benchmark scores. Applied to a $50 frontier price, the consensus implies the top tier at well under $20 by 2028.", "impliedP": 0.2, "impliedBy": "Read literally, a 10x-per-year decline takes $50 to $0.50 in two years, so the consensus narrative gives the #1 model at $20 or more almost no chance. Allowing that the narrative is about constant capability and that the frontier model has always carried a premium, the \"intelligence too cheap to meter\" view still puts the top list price above $20 at end-2028 at roughly one in five.", "source": "a16z, LLMflation, and Epoch AI LLM inference price trends", "url": "https://a16z.com/llmflation-llm-inference-cost/", "asOf": "2024-11-11", "note": "Consensus is a trend about constant-capability prices; this thesis is about the moving frontier, where the #1 list price has been flat-to-up for three years (Opus 4 $75, Fable 5 and 5.1 $50, GPT-6 Astra $50, 2.5x its predecessor). No prediction market prices frontier list prices." }, "leadingIndicators": [ { "id": "li-1", "label": "Output price of the top-ranked AA model", "connector": "artificial-analysis", "seriesId": "index:top-model-output-price-usd-per-1m", "unit": "USD per 1M output tokens", "onTrack": { "op": ">=", "value": 25 }, "offTrack": { "op": "<", "value": 15 }, "url": "https://artificialanalysis.ai/leaderboards/models", "note": "First-party output price of the current #1 model on the live index. Thresholds are for 2027-12-31." }, { "id": "li-2", "label": "Cheapest model within 5 index points of the top", "connector": "artificial-analysis", "seriesId": "index:cheapest-within-5-points-output-price-usd-per-1m", "unit": "USD per 1M output tokens", "onTrack": { "op": ">=", "value": 5 }, "offTrack": { "op": "<", "value": 2 }, "url": "https://artificialanalysis.ai/leaderboards/models", "note": "Lowest first-party output price among models scoring within 5 points of the top overall score. If near-frontier intelligence gets very cheap, the leader's price umbrella collapses. Thresholds are for 2027-12-31." } ], "milestones": [ { "id": "m1", "date": "2027-06-30", "statement": "The top-ranked Artificial Analysis model lists at $30 or more per million output tokens.", "p": 0.8 }, { "id": "m2", "date": "2027-12-31", "statement": "The top-ranked Artificial Analysis model lists at $20 or more per million output tokens.", "p": 0.75 }, { "id": "m3", "date": "2028-12-31", "statement": "The top-ranked Artificial Analysis model lists at $20 or more per million output tokens.", "p": 0.7 } ], "falsifier": "On 2027-12-31 the top-ranked model on the Artificial Analysis index lists below $15 per million output tokens.", "whyItMatters": "Capacity plans assume inference revenue per token collapses toward zero. That is true for last year's capability and false, so far, for this year's. If the frontier tier holds $20 or more, the labs keep gross margin to fund training and the demand for the highest-density racks stays anchored to a few closed models. If it breaks, the buildout is being financed on a price that no longer exists.", "whatWouldRaise": [ "OpenAI, Anthropic, or Google raise list prices on a new flagship in 2027.", "Frontier labs restrict top models to enterprise tiers with minimum commitments.", "Compute supply stays constrained (NVIDIA fiscal 2028 remains supply-limited)." ], "whatWouldCut": [ "An open-weights model reaches the top of the index and is served below $5 per million tokens.", "A frontier lab cuts its flagship price by more than half within six months of launch.", "The top index score stalls, letting cheaper distilled models tie the leader." ], "namedEntities": [ "Artificial Analysis" ], "marketCall": false, "recusalCheck": { "passed": true, "keysChecked": [ "employer-leasing", "named-competitor-bookings", "customer-pipeline", "ticker" ] }, "signedBy": "Brian Letort", "registeredAt": "2026-09-08", "amendments": [], "ruleHash": "3619779f20f7470e4b582bdc15475f96bcd6f92d78d2bb29a71bdd5c1ee2323d", "distance": 2.2335922215070942, "distanceLabel": "far above consensus", "engine": { "id": "RT-12", "prior": null, "council": { "p": 0.6219515905460664, "votes": [ { "persona": "grid-and-power-planner", "model": "openai/gpt-5.6-sol", "p": 0.57, "rationale": "Using premium flagship first-party API launches—not fixed-capability inference costs—as the reference class, I assign about a 0.20 annual hazard that competition or efficiency pushes the index leader below $20. Compounded over 2.31 years, that gives roughly 0.60 survival; I trim for leaderboard discontinuation and resolution-against ambiguity. Starting at $50 provides substantial cushion, and moving-frontier models can retain scarcity pricing while older capabilities cheapen. Conversely, a 60% list-price reduction is modest relative to historical inference-cost declines, and a cheaper model could take first place. Neither leading indicator supplied a current reading. The named index and pricing field must also remain available." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.64, "rationale": "Reference class: moving-frontier flagship API list prices, rather than constant-capability inference prices. The supplied $50 baseline leaves a 60% decline before failure, while frontier leaders can preserve premiums through scarce compute, high-effort inference, and product segmentation. Against that, rapid hardware/software cost declines and competition make a sub-$20 leader plausible by late 2028. I assign roughly a 15% annual hazard that frontier pricing crosses below $20; compounded over 2.3 years this gives about 0.69 survival, reduced to 0.64 for leaderboard discontinuation, pricing-definition changes, and snapshot risk. This is below the forecaster’s 0.70." }, { "persona": "enterprise-cio", "model": "openai/gpt-5.6-sol", "p": 0.64, "rationale": "Reference class: successive first-party frontier API generations since commercial LLMs began, a sparse and volatile sample. Starting at $50, failure requires more than a 60% list-price cut while the measured frontier advances; constant-capability cost declines therefore overstate downward pressure. Premium prices can persist because frontier models consume more inference compute and vendors can price-discriminate, but competition, distillation, and falling serving costs create regime-change risk. I assign a 0.16 annual hazard of falling below $20; compounded over 2.31 years gives about 0.67 survival. Allowing 0.03 for index discontinuation, endpoint ambiguity, and snapshot noise yields 0.64." }, { "persona": "credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.62, "rationale": "My reference class is premium flagship API pricing across successive frontier-model generations over two-to-three years. Starting from $50 provides a substantial cushion: resolution requires the qualifying leader’s standard output price to fall more than 60%. Moving-frontier capability, scarcity and high inference intensity support continued premium pricing, unlike constant-capability price trends. I assign roughly a 17% annual hazard of a pricing-regime shift below $20; compounded through 2028, then allowing modest index-discontinuation and resolution-ambiguity risk, yields 0.62. Competition, cheaper near-frontier substitutes and falling compute costs make the forecaster’s 0.7 somewhat optimistic, especially with no current indicator readings." }, { "persona": "geopolitics-policy-analyst", "model": "openai/gpt-5.6-sol", "p": 0.64, "rationale": "Reference class: successive frontier API launches in the three-year history supplied, where top nominal output prices stayed at $50–$75 even while constant-capability inference cheapened. The current $50 price has a 60% cushion over the threshold, and the moving leaderboard tends to select costly new capability rather than fixed capability. I nevertheless use about a 17% annual hazard that competition, architectural efficiency, or a low-priced leader pushes the top qualifying model below $20; compounded over 2.3 years, plus a small discontinuation risk, this leaves 0.64. Two or more release cycles and absent indicator data justify discounting the forecaster’s 0.70." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.62, "rationale": "I use a 0.60 base rate for a two-to-three-year window in which the year-end capability leader among major first-party APIs retains an output price above $20. The supplied $50 starting point helps: a miss requires a fall below $20, and moving-frontier pricing is stickier than constant-capability pricing. Against that, two model generations create meaningful chances that a cheaper rival tops the index; benchmark-normalized inference declines remain directionally relevant. I model about a 0.17 annual hazard of crossing below $20, plus a small penalty for index discontinuation or resolution ambiguity. With no live indicator readings or calibration table supplied, the forecaster’s 0.70 appears too confident." } ], "aggregation": "log-odds-mean", "extremization": 1, "dissent": null, "spread": 0.07, "warnings": [ "all 6 votes came from one vendor (openai); extremization disabled, persona diversity only" ] }, "final": { "p": 0.6219515905460664, "ci80": [ 0.5368928298346535, 0.700113838284626 ], "method": "council-only:log-odds-mean", "weights": { "council": 1 } }, "stated": 0.7, "history": [ { "asOf": "2026-09-07", "p": 0.6168688442041397, "source": "radar-engine", "runId": "radar-2026-09-07T23-12-09Z-06a58f" }, { "asOf": "2026-09-07", "p": 0.6219515905460664, "source": "radar-engine", "runId": "radar-2026-09-07T23-49-04Z-de8abf" } ], "indicatorStatus": "no-data", "milestones": [ { "id": "m1", "enginePrior": 0.65, "councilP": null }, { "id": "m2", "enginePrior": 0.65, "councilP": null }, { "id": "m3", "enginePrior": null, "councilP": null } ], "arena": "model-frontier", "status": "draft", "tier": "T2" }, "indicators": [ { "thesisId": "RT-12", "indicatorId": "li-1", "connector": "artificial-analysis", "seriesId": "index:top-model-output-price-usd-per-1m", "unit": "USD per 1M output tokens", "asOf": "2026-09-07", "current": null, "status": "no-data", "history": [], "url": "https://artificialanalysis.ai/leaderboards/models", "note": "ARTIFICIAL_ANALYSIS_API_KEY not set; the data API answers 401 and the leaderboard is JS-rendered. Enter readings under 'artificial-analysis:' in data/manual-series.json with a public source" }, { "thesisId": "RT-12", "indicatorId": "li-2", "connector": "artificial-analysis", "seriesId": "index:cheapest-within-5-points-output-price-usd-per-1m", "unit": "USD per 1M output tokens", "asOf": "2026-09-07", "current": null, "status": "no-data", "history": [], "url": "https://artificialanalysis.ai/leaderboards/models", "note": "ARTIFICIAL_ANALYSIS_API_KEY not set; the data API answers 401 and the leaderboard is JS-rendered. Enter readings under 'artificial-analysis:' in data/manual-series.json with a public source" } ], "indicatorStatus": "no-data", "chainEvents": [ { "seq": 214, "ts": "2026-09-07", "kind": "repriced", "hash": "3822f769738fbd54aaa31c84a4e4b478fd7cbe2479f3594ff6835434ed294017" }, { "seq": 215, "ts": "2026-09-07", "kind": "repriced", "hash": "63c99ca8923cc3ea2a97bc3f6292d75ba378384c613d7ccbc30921a80b007d8e" }, { "seq": 266, "ts": "2026-09-08", "kind": "registered", "hash": "fcbf4d65b805ae27922b1e01940157b4b5d0c4d9b17db1f336b2d0aef22710a6" } ], "url": "/predictions/model-frontier/RT-12" }, { "id": "RT-13", "title": "Chinese labs hold the open-weights lead through 2029", "arena": "model-frontier", "tier": "T2", "tag": "THESIS", "direction": "shift", "status": "registered", "statement": "On each of 2027-12-31, 2028-12-31, and 2029-12-31 the highest-scoring open-weights model on the Artificial Analysis Intelligence Index is developed by a lab headquartered in mainland China.", "resolutionRule": "On each of the three snapshot dates (or the first day after on which the page loads), open the Artificial Analysis open-weights models view and record the developer of the highest-scoring model (ties: HIT for that snapshot if any tied model qualifies). A lab qualifies if its parent company is headquartered in mainland China per its own corporate disclosures (DeepSeek, Alibaba/Qwen, Moonshot, Z.ai/Zhipu, MiniMax, ByteDance, Tencent, Baidu, StepFun qualify; Hong Kong and Taiwan do not). \"Open weights\" follows Artificial Analysis's label regardless of licence restrictiveness. Resolves HIT only if all three snapshots qualify. If the index or the open-weights label is discontinued before any snapshot, resolves MISS. Baseline 2026-09-07: Kimi K3 (Moonshot) and GLM-5.3 (Z.ai) tie at 60 on the default leaderboard; the best non-Chinese open model (NVIDIA Nemotron 3 Ultra) scores 38.", "resolutionSource": { "name": "Artificial Analysis open-weights models leaderboard", "url": "https://artificialanalysis.ai/models/open-source" }, "ambiguity": "resolves-against-forecaster", "horizon": "2029-12-31", "p": 0.5, "ci80": [ 0.35, 0.65 ], "consensus": { "value": "Artificial Analysis's open-weights view has been led by Chinese labs since early 2025 (DeepSeek R1, Qwen, Kimi, GLM); US open-weight entrants (gpt-oss, Llama, Gemma, NVIDIA Nemotron) have not held the top spot. No public forecast puts a probability on the lead persisting three more year-ends.", "impliedP": 0.5, "impliedBy": "The observed streak (Chinese labs at the top of the open-weights view at every quarter-end since early 2025, currently 22 points clear of the best US open model) implies roughly 0.8 per year-end that the lead survives; three consecutive year-ends compound to about 0.5.", "source": "Artificial Analysis open-weights models page", "url": "https://artificialanalysis.ai/models/open-source", "asOf": "2026-09-07", "note": "No Metaculus, Polymarket, or Kalshi question matches. Honest coin flip: I have no edge on whether Meta, OpenAI, or a European lab decides to open-weight a near-frontier model in 2028-2029, and that single decision is the whole question. Left at 0.50 deliberately; the year-by-year milestones carry the conviction." }, "leadingIndicators": [ { "id": "li-1", "label": "Developer HQ of the top open-weights AA model", "connector": "artificial-analysis", "seriesId": "index:open-weights-top-model-developer", "unit": "boolean (1 = mainland China HQ)", "onTrack": { "op": "==", "value": 1 }, "offTrack": { "op": "==", "value": 0 }, "url": "https://artificialanalysis.ai/models/open-source", "note": "1 when the highest-scoring open-weights model's developer qualifies under the resolution rule, checked quarterly." }, { "id": "li-2", "label": "Hugging Face downloads, Meta Llama organization", "connector": "hf-hub", "seriesId": "downloads:meta-llama", "unit": "downloads per month", "onTrack": { "op": "<", "value": 30000000 }, "offTrack": { "op": ">=", "value": 60000000 }, "url": "https://huggingface.co/meta-llama", "note": "Sum of last-30-day downloads for models under meta-llama on the Hub, as a proxy for a US open-weights comeback. Thresholds are for 2027-06-30." }, { "id": "li-3", "label": "Count of Chinese-lab open-weights models in the AA top 20 overall", "connector": "artificial-analysis", "seriesId": "index:top20-open-weights-china-count", "unit": "count", "onTrack": { "op": ">=", "value": 3 }, "offTrack": { "op": "<=", "value": 1 }, "url": "https://artificialanalysis.ai/leaderboards/models", "note": "Number of open-weights models from qualifying Chinese labs among the top 20 overall scores on the live index. Thresholds are for 2027-12-31." } ], "milestones": [ { "id": "m1", "date": "2026-09-30", "statement": "DeepSeek publishes the V4-Pro-0813 build weights to Hugging Face.", "p": 0.65, "liveQuestionId": "lq-p83" }, { "id": "m2", "date": "2027-12-31", "statement": "Top open-weights model on the AA index is from a mainland-China lab.", "p": 0.75 }, { "id": "m3", "date": "2028-12-31", "statement": "Top open-weights model on the AA index is from a mainland-China lab.", "p": 0.65 }, { "id": "m4", "date": "2029-12-31", "statement": "Top open-weights model on the AA index is from a mainland-China lab.", "p": 0.55 } ], "falsifier": "On any of the three snapshot dates the top open-weights model is from a lab headquartered outside mainland China.", "whyItMatters": "If the best weights an enterprise can run on its own hardware keep coming from China, sovereign-AI and procurement policy collide with engineering reality for every regulated buyer in the US and Europe. That tension shapes where inference capacity gets built and which models it is allowed to run. A US open-weights comeback would remove the tension; the thesis says it does not arrive.", "whatWouldRaise": [ "DeepSeek or Qwen release a model within 4 points of the overall AA leader in 2027.", "Meta or OpenAI publicly deprioritize open-weights releases.", "Chinese labs keep quarterly flagship cadence despite export controls." ], "whatWouldCut": [ "A US or European lab open-weights a model that takes the top open spot for a full quarter.", "Chinese labs move flagship releases to closed APIs only.", "Artificial Analysis reclassifies restrictive-licence Chinese models as not open." ], "namedEntities": [ "Artificial Analysis", "DeepSeek", "Alibaba", "Moonshot", "Z.ai", "MiniMax", "ByteDance", "Tencent", "Baidu", "StepFun" ], "marketCall": false, "recusalCheck": { "passed": true, "keysChecked": [ "employer-leasing", "named-competitor-bookings", "customer-pipeline", "ticker" ] }, "signedBy": "Brian Letort", "registeredAt": "2026-09-08", "amendments": [], "ruleHash": "812d1f306d08262ee2e5043a3db5575d60a48e2acdb9e9a3bbc1a2e3e0ab90ce", "distance": 0, "distanceLabel": "in line with consensus", "engine": { "id": "RT-13", "prior": null, "council": { "p": 0.4264880081903964, "votes": [ { "persona": "grid-and-power-planner", "model": "openai/gpt-5.6-sol", "p": 0.45, "rationale": "The relevant reference class is annual retention of a technology-leader ecosystem, not any single model. Starting from the stated 22-point lead and China’s multi-lab open-weight cadence, I assign roughly 0.78 for 2027, then conditional retention probabilities of 0.78 in 2028 and 0.75 in 2029. Their product is about 0.46; I shave slightly for leaderboard/label discontinuation and adverse ambiguity. The snapshots are highly path-dependent, but the conjunctive rule still creates three opportunities for one near-frontier release from Meta, OpenAI, Google, NVIDIA, or a European lab to cause a miss. Meta downloads measure adoption, not AA leadership. My estimate is modestly below the stated 0.50." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.43, "rationale": "The relevant reference class is country-level benchmark leadership in fast-moving model ecosystems, not persistence by one lab. China benefits from several qualifying developers, a sustained lead since 2025, and a large current 60-to-38 gap. I estimate roughly 0.78 for 2027, then conditional retention probabilities near 0.76 and 0.73 as rival-release and compute-access hazards accumulate. Those risks are correlated: a durable shift toward near-frontier US or European open-weight releases could spoil multiple snapshots, while continued Chinese openness supports all three. Compounding the conditional estimates and allowing a small probability that the required leaderboard or label disappears yields about 0.43." }, { "persona": "enterprise-cio", "model": "openai/gpt-5.6-sol", "p": 0.42, "rationale": "I use frontier-model leaderboard persistence under rapid release cycles as the reference class. China’s reported 22-point baseline advantage, multiple qualifying labs, and leadership since early 2025 make 2027 favorable. However, three conjunctive year-end snapshots expose the thesis to repeated release-cycle risk: I estimate conditional qualification probabilities near 0.80, 0.76, and 0.72. Their product is 0.44, reduced slightly for index/label discontinuation and resolution ambiguity. Meta download volume indicates ecosystem adoption, not a model capable of topping this index. A single near-frontier open-weight release outside mainland China causes a miss." }, { "persona": "credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.44, "rationale": "Reference class: persistence of open-weight leaderboard regimes since 2025, adjusted for three conjunctive year-end snapshots. The supplied baseline shows a large 22-point Chinese lead and several qualifying labs, making 2027 leadership likely. However, the horizon allows Meta, OpenAI, Google, NVIDIA, or a European lab to release near-frontier weights; only one non-Chinese leader on one snapshot causes a miss. I estimate conditional success near 0.80, 0.76, and 0.73 across the three dates. Compounding gives roughly 0.44 after a small allowance for index or label discontinuation. Positive correlation across years prevents treating each snapshot as an independent coin flip." }, { "persona": "geopolitics-policy-analyst", "model": "openai/gpt-5.6-sol", "p": 0.44, "rationale": "The reference class is annual leadership persistence on fast-moving AI model benchmarks: incumbency matters, but leads are vulnerable to infrequent major releases. The supplied 60-versus-38 gap and multiple qualifying Chinese labs support about 0.78 for 2027. Conditional on retaining leadership, I use 0.77 for 2028 and 0.76 for 2029; correlation makes this more appropriate than multiplying unconditional milestone estimates. Their product is 0.46, reduced slightly for leaderboard or label discontinuation. Meta’s download volume shows ecosystem strength, not near-frontier benchmark leadership. A single US or European near-frontier open-weight release causes a miss, so I am below the forecaster’s 0.50." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.38, "rationale": "Base rate: 0.30 for a currently leading national bloc remaining first at all three annual snapshots of a fast-moving technology benchmark. I adjust upward because the supplied baseline shows a large 60–38 gap, several independent qualifying Chinese labs, and leadership since early 2025. I adjust downward for the strict three-leg conjunction, plausible near-frontier open-weight releases by non-Chinese labs, exact-date exposure, and automatic failure if the index or label disappears. The milestone probabilities multiply to 0.27 under independence; positive persistence across years raises the joint estimate, but not to 0.50. The key leaderboard indicators were unavailable, and no calibration history was supplied." } ], "aggregation": "log-odds-mean", "extremization": 1, "dissent": null, "spread": 0.07, "warnings": [ "all 6 votes came from one vendor (openai); extremization disabled, persona diversity only" ] }, "final": { "p": 0.4264880081903964, "ci80": [ 0.34384758795484904, 0.5134481066593181 ], "method": "council-only:log-odds-mean", "weights": { "council": 1 } }, "stated": 0.5, "history": [ { "asOf": "2026-09-07", "p": 0.4261432996416763, "source": "radar-engine", "runId": "radar-2026-09-07T23-12-09Z-06a58f" }, { "asOf": "2026-09-07", "p": 0.4264880081903964, "source": "radar-engine", "runId": "radar-2026-09-07T23-49-04Z-de8abf" } ], "indicatorStatus": "on-track", "milestones": [ { "id": "m1", "enginePrior": 0.46063666269036246, "councilP": 0.3315466546345604 }, { "id": "m2", "enginePrior": 0.5681818181818182, "councilP": null }, { "id": "m3", "enginePrior": null, "councilP": null }, { "id": "m4", "enginePrior": null, "councilP": null } ], "arena": "model-frontier", "status": "draft", "tier": "T2" }, "indicators": [ { "thesisId": "RT-13", "indicatorId": "li-1", "connector": "artificial-analysis", "seriesId": "index:open-weights-top-model-developer", "unit": "boolean (1 = mainland China HQ)", "asOf": "2026-09-07", "current": null, "status": "no-data", "history": [], "url": "https://artificialanalysis.ai/leaderboards/models", "note": "ARTIFICIAL_ANALYSIS_API_KEY not set; the data API answers 401 and the leaderboard is JS-rendered. Enter readings under 'artificial-analysis:' in data/manual-series.json with a public source" }, { "thesisId": "RT-13", "indicatorId": "li-2", "connector": "hf-hub", "seriesId": "downloads:meta-llama", "unit": "downloads per month", "asOf": "2026-09-07", "current": 25459569, "status": "on-track", "history": [ { "asOf": "2026-09-07", "value": 25459569 } ], "url": "https://huggingface.co/meta-llama", "note": "rolling 30-day downloads" }, { "thesisId": "RT-13", "indicatorId": "li-3", "connector": "artificial-analysis", "seriesId": "index:top20-open-weights-china-count", "unit": "count", "asOf": "2026-09-07", "current": null, "status": "no-data", "history": [], "url": "https://artificialanalysis.ai/leaderboards/models", "note": "ARTIFICIAL_ANALYSIS_API_KEY not set; the data API answers 401 and the leaderboard is JS-rendered. Enter readings under 'artificial-analysis:' in data/manual-series.json with a public source" } ], "indicatorStatus": "on-track", "chainEvents": [ { "seq": 216, "ts": "2026-09-07", "kind": "repriced", "hash": "173f2af8fce371e58c59e8c8d83023aecb2bfd7af0bca2ca8e77e63a4bfa774a" }, { "seq": 217, "ts": "2026-09-07", "kind": "repriced", "hash": "cabe0c91f2bbda26819fcad7f19dc8da72848abd64ea6dc512110ee9812bc1d4" }, { "seq": 267, "ts": "2026-09-08", "kind": "registered", "hash": "21f7c7c7d87cb714949e2396284d7c5f78816306f2f5f98fd02c6ec08bc90e59" } ], "url": "/predictions/model-frontier/RT-13" }, { "id": "RT-14", "title": "SWE-bench Pro public set passes 90% by 2027", "arena": "agi-capabilities", "tier": "T1", "tag": "CALL", "direction": "up", "status": "registered", "statement": "A model posts a resolved rate of at least 90.0% on the SWE-bench Pro public leaderboard (Scale AI, public problem set) on or before 2027-12-31.", "resolutionRule": "Open the SWE-bench Pro public leaderboard at scale.com/leaderboard/swe_bench_pro_public on 2028-01-02 (or the first day after on which it loads). Resolves HIT if any entry dated on or before 2027-12-31 shows a resolved rate (the leaderboard's headline score) >= 90.0% on the public set, under the leaderboard's standard evaluation (any scaffold the leaderboard accepts; self-reported entries count only if the leaderboard lists them). The private or commercial SWE-bench Pro subsets do not count. If Scale AI retires the public leaderboard before 2027-12-31 and no successor page lists the same metric, resolves MISS. Baseline 2026-09-07: top entry 80.3% (Claude Fable 5 Deep Thinking Mode, with tools).", "resolutionSource": { "name": "Scale AI SWE-bench Pro public leaderboard", "url": "https://scale.com/leaderboard/swe_bench_pro_public" }, "ambiguity": "resolves-against-forecaster", "horizon": "2027-12-31", "p": 0.8, "ci80": [ 0.65, 0.9 ], "consensus": { "value": "The public-set top score moved from about 23% at launch (Sept 2025) to 80.3% (Sept 2026); OpenAI now recommends SWE-bench Pro over Verified because Verified is contaminated. Straight-line extrapolation reaches ~90% by end-2027 but the Metaculus community historically expected 90% on Verified only by 2029.", "impliedP": 0.55, "impliedBy": "The public-set top score has gained ~57 points in twelve months; even at a third of that pace the straight line crosses 90% in late 2027, so trend puts the threshold right at the horizon with roughly even odds. The Metaculus community's prior expectation of 2029 for 90% on Verified (a benchmark that then saturated three years early) argues for discounting the trend, not the threshold.", "source": "Metaculus, \"When will AIs hit 90% on SWE-bench Verified\" (resolved 2026-04-07) and OpenAI, \"Why SWE-bench Verified no longer measures frontier coding capabilities\"", "url": "https://www.metaculus.com/questions/28610/", "asOf": "2026-04-07", "note": "No live market exists for a SWE-bench Pro threshold; the resolved Verified question and OpenAI's benchmark note are the nearest dated public references. Rung 1 of the capability ladder. My 0.80 rests on labs now reporting SWE-bench Pro as the headline coding metric, which historically saturates a benchmark within 18 months of adoption." }, "leadingIndicators": [ { "id": "li-1", "label": "Top SWE-bench Pro public resolved rate", "connector": "manual", "seriesId": "swebench-pro-public-top-score", "unit": "percent", "onTrack": { "op": ">=", "value": 86 }, "offTrack": { "op": "<", "value": 83 }, "url": "https://scale.com/leaderboard/swe_bench_pro_public", "note": "Headline resolved rate of the top row on the public leaderboard. Thresholds are for 2027-03-31." }, { "id": "li-2", "label": "Artificial Analysis Terminal-Bench 2.1 top score", "connector": "artificial-analysis", "seriesId": "eval:terminalbench-v2-1:top", "unit": "percent", "onTrack": { "op": ">=", "value": 85 }, "offTrack": { "op": "<", "value": 78 }, "url": "https://artificialanalysis.ai/evaluations/terminalbench-v2-1", "note": "Highest Terminal-Bench 2.1 score on Artificial Analysis, an independent agentic-coding measure that moves with SWE-bench Pro. Thresholds are for 2027-03-31." } ], "milestones": [], "falsifier": "On 2027-12-31 no entry on the SWE-bench Pro public leaderboard shows a resolved rate at or above 90.0%.", "whyItMatters": "This is the first rung of the capability ladder: can a model close real, uncontaminated repository issues at the rate a strong senior engineer would. Passing it moves the enterprise question from whether coding agents work to how much software headcount they displace, which is what rung four measures.", "whatWouldRaise": [ "Top public score exceeds 86% before April 2027.", "Scale AI reports that remaining failures are concentrated in environment setup rather than code reasoning.", "Two labs publish SWE-bench Pro scores as headline model-card metrics." ], "whatWouldCut": [ "Scale AI publishes an audit showing a hard ceiling from ambiguous tasks below 90%.", "Top score stalls within 2 points for two consecutive quarters.", "Labs shift reporting to a successor benchmark and stop submitting to SWE-bench Pro." ], "namedEntities": [ "Scale AI" ], "marketCall": false, "recusalCheck": { "passed": true, "keysChecked": [ "employer-leasing", "named-competitor-bookings", "customer-pipeline", "ticker" ] }, "signedBy": "Brian Letort", "registeredAt": "2026-09-08", "amendments": [], "ruleHash": "00a4931c9dbb311c7035fd99648cd99811484676b87345e61e13635d497d593f", "distance": 1.1856236656577395, "distanceLabel": "clearly above consensus", "engine": { "id": "RT-14", "prior": { "p": 0.65, "method": "reference-class:qtype:threshold", "note": "ledger base rate, n=19, horizon 480d; the ledger has no multi-year history" }, "council": { "p": 0.7071582692810922, "votes": [ { "persona": "grid-and-power-planner", "model": "openai/gpt-5.6-sol", "p": 0.74, "rationale": "Using the reference class of public coding benchmarks already above 80%, I estimate a 0.79 chance of gaining the remaining 9.7 points within 16 months. The supplied rise from roughly 23% to 80.3% supports rapid improvement, and accepted scaffolds plus a public problem set increase benchmark-specific optimization. However, progress usually slows near saturation because remaining tasks are unusually difficult or evaluation-sensitive. I apply about a 0.94 probability that a qualifying result is actually dated, listed under Scale’s standard evaluation, and the leaderboard or same-metric successor remains available. Multiplying gives approximately 0.74, below the forecaster’s 0.80." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.72, "rationale": "My reference class is threshold crossings on rapidly improving but saturating software-agent benchmarks, anchored to the 0.65 structural prior. Moving from 80.3% to 90.0% requires roughly halving the remaining error rate over 16 months. New model generations, inference scaling, and accepted scaffolds make that plausible, while the public set permits optimization and possible contamination. However, the hardest residual tasks can create a long tail, and the observed one-year gain should not be extrapolated linearly. I estimate about 0.77 for a qualifying technical result, discounted for leaderboard retirement, methodology changes, submission delays, or failure to list the result by the deadline." }, { "persona": "enterprise-cio", "model": "openai/gpt-5.6-sol", "p": 0.68, "rationale": "My reference class is frontier software-engineering benchmarks after broad lab adoption: headline scores often saturate within 12–24 months, especially when public tasks and accepted scaffolds permit targeted optimization. Moving from 80.3 to 90.0 requires eliminating roughly half the remaining failures, harder than a straight-line extrapolation suggests because of irreducible, flaky, or environment-sensitive tasks. The 16-month window still allows multiple model and agent generations. I discount for diminishing returns, absent confirming indicators, leaderboard submission discretion, and the conjunctive requirement that Scale continue listing the qualifying result. This is above the 0.65 structural prior but below the stated 0.80." }, { "persona": "credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.68, "rationale": "My reference class is public coding benchmarks after frontier-lab adoption: rapid early gains followed by sharply diminishing returns near saturation. From 80.3%, reaching 90% requires halving the unresolved-task rate, materially harder than adding 9.7 linear points. Still, 16 months, public-task optimization, and accepted tool scaffolds make crossing more likely than not. I estimate a 0.62 chance of crossing in the next 12 months and a 0.24 conditional chance in the final four, then discount roughly 0.03 for leaderboard retirement, methodology, or qualifying-entry risks. The 0.65 structural prior is consistent; without fresh indicators, 0.80 is too confident." }, { "persona": "geopolitics-policy-analyst", "model": "openai/gpt-5.6-sol", "p": 0.72, "rationale": "Reference class: frontier software-engineering benchmarks after broad lab adoption, which typically show rapid early gains followed by diminishing returns near saturation. Advancing from 80.3 to 90.0 within roughly 16 months is plausible because accepted tool scaffolds, public tasks, and competition among several labs create repeated attempts. The prior year’s unusually large gain warrants an uplift from the 0.65 structural prior. I discount for increasingly difficult residual tasks, evaluation variance, possible rule tightening, and literal-resolution risk from leaderboard retirement or metric replacement. A roughly 0.62 annual success hazard compounded across 1.32 years yields about 0.72, below the forecaster’s 0.80." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.7, "rationale": "Base rate: 0.65 from the supplied reference class for threshold questions. I adjust upward because the stated leaderboard gain from roughly 23% to 80.3% in one year leaves only 9.7 points over nearly 16 months, and accepted scaffolds broaden the paths to success. I limit that adjustment because the remaining improvement requires about halving the current error rate, benchmark progress commonly slows near saturation, and both qualifying performance and timely listing on the named public leaderboard are required. Retirement without a same-metric successor also causes failure. With no fresh indicator data, the forecaster’s 0.80 relies too heavily on straight-line extrapolation." } ], "aggregation": "log-odds-mean", "extremization": 1, "dissent": null, "spread": 0.06, "warnings": [ "all 6 votes came from one vendor (openai); extremization disabled, persona diversity only" ] }, "final": { "p": 0.6792507543854072, "ci80": [ 0.5716744891851333, 0.7779123936330596 ], "method": "stacked-fixed-weights (radar: prior 0.5, council 0.5; no resolved radar items to learn from)", "weights": { "council": 0.5, "prior": 0.5 } }, "stated": 0.8, "history": [ { "asOf": "2026-09-07", "p": 0.6792553358295149, "source": "radar-engine", "runId": "radar-2026-09-07T23-12-09Z-06a58f" }, { "asOf": "2026-09-07", "p": 0.6792507543854072, "source": "radar-engine", "runId": "radar-2026-09-07T23-49-04Z-de8abf" } ], "indicatorStatus": "no-data", "milestones": [], "arena": "agi-capabilities", "status": "draft", "tier": "T1" }, "indicators": [ { "thesisId": "RT-14", "indicatorId": "li-1", "connector": "manual", "seriesId": "swebench-pro-public-top-score", "unit": "percent", "asOf": "2026-09-07", "current": null, "status": "no-data", "history": [], "url": "https://scale.com/leaderboard/swe_bench_pro_public", "note": "manual: 'swebench-pro-public-top-score' not in manual-series.json" }, { "thesisId": "RT-14", "indicatorId": "li-2", "connector": "artificial-analysis", "seriesId": "eval:terminalbench-v2-1:top", "unit": "percent", "asOf": "2026-09-07", "current": null, "status": "no-data", "history": [], "url": "https://artificialanalysis.ai/leaderboards/models", "note": "ARTIFICIAL_ANALYSIS_API_KEY not set; the data API answers 401 and the leaderboard is JS-rendered. Enter readings under 'artificial-analysis:' in data/manual-series.json with a public source" } ], "indicatorStatus": "no-data", "chainEvents": [ { "seq": 218, "ts": "2026-09-07", "kind": "repriced", "hash": "ed996775e7d642eb2cf9467c03c5bffe588e600400a9ec6a35527a7afe1d0d46" }, { "seq": 219, "ts": "2026-09-07", "kind": "repriced", "hash": "f8ea9814adb7ac78139c68facd095b31ae17b62d36cdd0e77525f1437c124b8d" }, { "seq": 268, "ts": "2026-09-08", "kind": "registered", "hash": "071e685655b2d0d08e956f138e14f7ca8db0d6742675ad2f99a33fcec8a4fc26" } ], "url": "/predictions/agi-capabilities/RT-14" }, { "id": "RT-15", "title": "AI solves ten FrontierMath open problems by 2028", "arena": "agi-capabilities", "tier": "T2", "tag": "CALL", "direction": "up", "status": "registered", "statement": "Epoch AI's FrontierMath Open Problems page credits AI systems with verified solutions to at least 10 of the listed unsolved research-level problems on or before 2028-12-31.", "resolutionRule": "Open epoch.ai/frontiermath/open-problems on 2029-01-02 (or the first day after on which it loads). Count problems whose status Epoch marks as solved by an AI system and verified by the problem's author or Epoch's reviewers (Epoch's own \"solved\" designation; partial progress, human-assisted solutions, or solutions Epoch lists as under review do not count). Resolves HIT if the count is >= 10 with solution dates on or before 2028-12-31. If Epoch retires the page or stops tracking solutions before that date, resolves MISS. Baseline 2026-09-07: 2 of about 50 problems marked solved. This rung replaces the original \"GPQA Diamond >= 90% third-party verified\" rung, which had already resolved (Artificial Analysis reports GPT-6 Astra at 96.0% and several models above 90%).", "resolutionSource": { "name": "Epoch AI FrontierMath Open Problems", "url": "https://epoch.ai/frontiermath/open-problems" }, "ambiguity": "resolves-against-forecaster", "horizon": "2028-12-31", "p": 0.65, "ci80": [ 0.5, 0.8 ], "consensus": { "value": "Epoch's page shows 2 of ~50 open problems solved by AI as of September 2026, with the first solutions arriving in 2026; Epoch's own commentary describes the set as designed to resist current models for years. GPQA Diamond, the original rung, saturated at 96% (Artificial Analysis, GPT-6 Astra).", "impliedP": 0.35, "impliedBy": "Two solutions in the first nine months of tracking extrapolate linearly to about 6-8 solved by end-2028; reaching 10 needs the solve rate to roughly double, which the linear reading gives about one chance in three.", "source": "Epoch AI FrontierMath Open Problems page and Artificial Analysis GPQA Diamond results", "url": "https://epoch.ai/frontiermath/open-problems", "asOf": "2026-09-07", "note": "No Metaculus or market question targets this count. Rung 2 of the capability ladder (replacing GPQA Diamond, which saturated). My 0.65 is above the linear read because FrontierMath Tier 4 went from 25% to 97.6% (GPT-6 Astra) in a year, and open-problem solves follow Tier 4 saturation with a lag." }, "leadingIndicators": [ { "id": "li-1", "label": "FrontierMath open problems marked solved by AI", "connector": "epoch", "seriesId": "frontiermath_open_problems:solved_by_ai:count", "unit": "count", "onTrack": { "op": ">=", "value": 4 }, "offTrack": { "op": "<=", "value": 2 }, "url": "https://epoch.ai/frontiermath/open-problems", "note": "Count of problems with Epoch status solved-by-AI. Thresholds are for 2027-06-30." } ], "milestones": [ { "id": "m1", "date": "2027-06-30", "statement": "Epoch credits AI with at least 4 verified open-problem solutions.", "p": 0.75 }, { "id": "m2", "date": "2027-12-31", "statement": "Epoch credits AI with at least 6 verified open-problem solutions.", "p": 0.7 }, { "id": "m3", "date": "2028-12-31", "statement": "Epoch credits AI with at least 10 verified open-problem solutions.", "p": 0.65 } ], "falsifier": "On 2027-12-31 Epoch credits AI with fewer than 4 verified open-problem solutions.", "whyItMatters": "Multiple-choice science benchmarks are exhausted; GPQA Diamond fell before this book was written. Solving research problems nobody has solved is the first rung that cannot be reached by memorization. Ten verified solutions would mean models are producing new mathematics at a rate that changes how research labs staff.", "whatWouldRaise": [ "A frontier lab publishes a verified open-problem solution as a headline result in 2027.", "Epoch verifies a third and fourth solution before mid-2027.", "Mathematicians submit new open problems faster than Epoch can verify solutions, signalling demand." ], "whatWouldCut": [ "Epoch tightens verification and reverses a credited solution.", "Author verification of submitted solutions takes longer than six months.", "Labs stop evaluating on FrontierMath after saturating Tiers 1-3." ], "namedEntities": [ "Epoch AI" ], "marketCall": false, "recusalCheck": { "passed": true, "keysChecked": [ "employer-leasing", "named-competitor-bookings", "customer-pipeline", "ticker" ] }, "signedBy": "Brian Letort", "registeredAt": "2026-09-08", "amendments": [], "ruleHash": "c3322c4421b33d1038fc5cebc42cff35972376b4a20ed2d6ef894eeb013d3053", "distance": 1.238078416812447, "distanceLabel": "clearly above consensus", "engine": { "id": "RT-15", "prior": null, "council": { "p": 0.41798843399860575, "votes": [ { "persona": "grid-and-power-planner", "model": "openai/gpt-5.6-sol", "p": 0.43, "rationale": "Starting from 2, this requires eight additional verified solves in 28 months, about 3.4 per year. My reference class is early research-benchmark progress: capability gains are lumpy, while verification and publication lag model performance. A Poisson-style extrapolation from roughly two solves during 2026 lands near, but below, the threshold. Tier-4 saturation is supportive, yet transfer from bounded benchmark questions to genuinely open problems is weak, and the remaining set may be selected for resistance. The literal rule also requires Epoch—not merely papers or announcements—to mark ten qualifying, non-human-assisted solutions; page retirement adds another failure mode. Thus 0.65 appears optimistic." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.41, "rationale": "The reference class is sparse early successes on newly launched, adversarial research benchmarks, where closed-benchmark saturation transfers poorly to genuinely open problems. From 2 solved, the thesis needs 8 additional verified solves in roughly 28 months—about 3.4 annually—with harder-tail depletion and review lag. Capability acceleration and a pool of roughly 48 remaining problems make that plausible, but the strict rule excludes partial or human-assisted work. Resolution also conjunctively requires Epoch’s designation, qualifying solution dates, and continued page tracking. I estimate about 0.44 conditional on continued tracking, reduced to 0.41 overall, below the stated 0.65." }, { "persona": "enterprise-cio", "model": "openai/gpt-5.6-sol", "p": 0.38, "rationale": "Starting from 2, the thesis requires 8 additional verified solves in roughly 28 months, about 3.4 per year. My reference class is transitions from benchmark saturation to independently verified research results: benchmark scores can improve quickly, but novel proofs face search, evaluation, attribution, and publication lags. A count model centered on 2–3 accepted solves in 2027 and 3–4 in 2028 puts the threshold in the upper tail even with acceleration. Strict exclusion of human-assisted and under-review work, plus page-continuity risk, lowers the odds further. Tier 4 saturation is supportive but not a reliable linear predictor of autonomous open-problem closure." }, { "persona": "credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.44, "rationale": "Using sparse, independently verified research-mathematics breakthroughs as the reference class, the target requires eight additional credited solutions in about 2.3 years, or roughly 3.5 annually. A Poisson-style baseline near three annual additions puts the threshold below even odds. Rapid capability gains create substantial upside, but benchmark saturation does not reliably translate into autonomous theorem solving. Correlated problem difficulty, verification delays, exclusion of human-assisted work, and the page-continuity condition reduce resolution probability. With no indicator update beyond the baseline count of two, I place the claim below the forecaster’s 0.65." }, { "persona": "geopolitics-policy-analyst", "model": "openai/gpt-5.6-sol", "p": 0.47, "rationale": "Starting from 2, the thesis needs 8 additional verified solutions in about 28 months. A Poisson-style reference class for rare, independently verified research breakthroughs, calibrated loosely to two arrivals during 2026, puts the required pace near the upper-middle of the plausible range. Rapid benchmark gains support an accelerating arrival rate, but Tier 4 saturation is weaker evidence for solving genuinely open problems. Epoch’s verification delay, exclusion of human-assisted or under-review work, and the requirement that the page remain active all reduce resolution probability. The missing indicator update adds uncertainty rather than positive evidence. This makes 0.65 too optimistic under strict resolution." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.38, "rationale": "Base rate: I use a 0.30 chance that a curated set of research-level open problems reaches 20% verified autonomous completion within roughly 28 months after only two initial machine solves. The target needs eight additional successes among about 48, while review latency, exclusion of human-assisted work, and page-retirement risk all reduce resolution odds. Rapid FrontierMath Tier 4 saturation is positive, but benchmark performance is a weaker reference class than novel, author-verified research, and the connector supplies no confirming trajectory. Capability acceleration raises my estimate above the base rate, but 0.65 underweights threshold and verification risk." } ], "aggregation": "log-odds-mean", "extremization": 1, "dissent": null, "spread": 0.09, "warnings": [ "all 6 votes came from one vendor (openai); extremization disabled, persona diversity only" ] }, "final": { "p": 0.41798843399860575, "ci80": [ 0.33602999284577545, 0.5047407184059167 ], "method": "council-only:log-odds-mean", "weights": { "council": 1 } }, "stated": 0.65, "history": [ { "asOf": "2026-09-07", "p": 0.41476561784507004, "source": "radar-engine", "runId": "radar-2026-09-07T23-12-09Z-06a58f" }, { "asOf": "2026-09-07", "p": 0.41798843399860575, "source": "radar-engine", "runId": "radar-2026-09-07T23-49-04Z-de8abf" } ], "indicatorStatus": "no-data", "milestones": [ { "id": "m1", "enginePrior": 0.65, "councilP": null }, { "id": "m2", "enginePrior": 0.65, "councilP": null }, { "id": "m3", "enginePrior": null, "councilP": null } ], "arena": "agi-capabilities", "status": "draft", "tier": "T2" }, "indicators": [ { "thesisId": "RT-15", "indicatorId": "li-1", "connector": "epoch", "seriesId": "frontiermath_open_problems:solved_by_ai:count", "unit": "count", "asOf": "2026-09-07", "current": null, "status": "no-data", "history": [], "url": "https://epoch.ai/frontiermath/open-problems", "note": "epoch: unknown dataset 'frontiermath_open_problems'; known: notable_ai_models, large_scale_ai_models, ml_hardware, gpu_clusters, benchmarks" } ], "indicatorStatus": "no-data", "chainEvents": [ { "seq": 220, "ts": "2026-09-07", "kind": "repriced", "hash": "1800f700495403b1dbfeee0603c145e15751a6c05aac958e7d9fb2e06951d539" }, { "seq": 221, "ts": "2026-09-07", "kind": "repriced", "hash": "e83d38673f5fe04fbda782200c431cf91e67cf45daf7009de6572351f19e6287" }, { "seq": 269, "ts": "2026-09-08", "kind": "registered", "hash": "a57cda39d9e87f142fb6255953650289b90824844d81d58fde7b99f7096192d3" } ], "url": "/predictions/agi-capabilities/RT-15" }, { "id": "RT-16", "title": "METR does not certify a one-month agent horizon", "arena": "agi-capabilities", "tier": "T2", "tag": "WARNING", "direction": "down", "status": "registered", "statement": "METR does not publish a 50%-success time-horizon point estimate of 160 work-hours (one working month) or more for any publicly released model on its Time Horizons page on or before 2028-12-31.", "resolutionRule": "Open metr.org/time-horizons (or METR's successor page for the time-horizon methodology) on 2029-01-02 or the first day after on which it loads. Resolves MISS (the horizon is certified) if METR's own published point estimate of the 50%-success time horizon for any model that is publicly available via API or download, evaluated under METR's then-current Time Horizon methodology (1.1 or later), is >= 160 hours (9,600 minutes) with a publication date on or before 2028-12-31. Confidence-interval upper bounds, 80%-success horizons, lab self-evaluations, and estimates METR labels as unreliable or out of range do not count as certification. Resolves HIT if no such estimate exists. If METR stops publishing time horizons before that date, resolves MISS (against the forecaster). Baseline 2026-09-07: Claude Mythos Preview ~17.4 hours, with METR stating estimates above ~16 hours are unreliable under the current task suite.", "resolutionSource": { "name": "METR Time Horizons page", "url": "https://metr.org/time-horizons/" }, "ambiguity": "resolves-against-forecaster", "horizon": "2028-12-31", "p": 0.65, "ci80": [ 0.5, 0.8 ], "consensus": { "value": "METR's published doubling time of roughly 4-7 months, applied to a 17-hour horizon, reaches 160 hours between mid-2027 and mid-2028; METR's own March 2026 note warns that its task suite cannot reliably measure horizons much above 16 hours. An informal METR staff poll quoted on Manifold put 29% on exceeding 32 hours by end-2026.", "impliedP": 0.35, "impliedBy": "From ~17 hours, 160 hours is 3.2 doublings; at METR's 4-7 month doubling time that lands between mid-2027 and mid-2028, so the trend alone gives ~0.8 that the horizon is certified and ~0.2 that it is not. Discounting for METR's statement that horizons above 16 hours are unmeasurable on the current suite raises the consensus-path probability of no certification to about 0.35.", "source": "METR, Time Horizons page and methodology note (2026-03-20); Manifold, \"Best METR 50% Time Horizon in 2026\"", "url": "https://manifold.markets/Bayesian/best-metr-time-horizons-in-2026", "asOf": "2026-09-07", "note": "Rung 3 of the capability ladder, stated as the skeptical side. No market prices the 160-hour threshold directly. My 0.65 is above the trend read because certifying a month-long horizon needs a task suite with multi-week human baselines that METR has not announced, and a published point estimate is the resolving event, not the capability." }, "leadingIndicators": [ { "id": "li-1", "label": "Longest METR 50% time horizon for a public model", "connector": "manual", "seriesId": "metr-th-50pct-max-hours", "unit": "hours", "onTrack": { "op": "<", "value": 40 }, "offTrack": { "op": ">=", "value": 60 }, "url": "https://metr.org/time-horizons/", "note": "Largest published 50% point estimate across public models on METR's page. Thresholds are for 2027-06-30; a certified 60-hour horizon by then puts 160 hours within reach." }, { "id": "li-2", "label": "Longest human-baseline task in METR's suite", "connector": "manual", "seriesId": "metr-suite-max-task-hours", "unit": "hours", "onTrack": { "op": "<", "value": 80 }, "offTrack": { "op": ">=", "value": 200 }, "url": "https://metr.org/time-horizons/", "note": "Longest human-time task METR reports in its evaluation suite; the suite must contain tasks well beyond 160 hours before a 160-hour horizon is measurable. Thresholds are for 2027-12-31." } ], "milestones": [ { "id": "m1", "date": "2027-06-30", "statement": "METR publishes a 50% time horizon of at least 40 hours for a public model.", "p": 0.6 }, { "id": "m2", "date": "2027-12-31", "statement": "METR publishes a task suite with human-baseline tasks longer than 80 hours.", "p": 0.5 }, { "id": "m3", "date": "2028-12-31", "statement": "METR publishes a 50% time horizon of at least 160 hours for a public model (rung 3 achieved).", "p": 0.35 } ], "falsifier": "METR publishes a 50% time horizon of 160 hours or more for a public model on or before 2028-12-31, or publishes a task suite with multi-week human baselines before 2027-12-31 (which would make certification in 2028 likely).", "whyItMatters": "A month-long autonomous horizon is the rung where agents stop being tools and become staff. Everything downstream, including the labor-market rung and the capacity to run always-on agents per employee, depends on whether this is reached in 2028 or measured only in retrospect. The thesis bets that measurement lags capability, which matters for how anyone reads AGI timelines.", "whatWouldRaise": [ "METR revises its doubling-time estimate longer than 8 months.", "METR publishes that 50% horizons plateau near 30-50 hours across two model generations.", "Frontier labs stop granting METR pre-release access, delaying public estimates." ], "whatWouldCut": [ "METR releases a task suite with multi-week human baselines in 2027.", "A public model reaches 40 hours before mid-2027.", "Independent replications (Epoch, Apollo) publish horizon estimates consistent with METR's." ], "namedEntities": [ "METR" ], "marketCall": false, "recusalCheck": { "passed": true, "keysChecked": [ "employer-leasing", "named-competitor-bookings", "customer-pipeline", "ticker" ] }, "signedBy": "Brian Letort", "registeredAt": "2026-09-08", "amendments": [], "ruleHash": "c438106ad332f00fbf93d56724d18ae8607cf009e86aaa876092d7b5b2247056", "distance": 1.238078416812447, "distanceLabel": "clearly above consensus", "engine": { "id": "RT-16", "prior": null, "council": { "p": 0.5570641685163671, "votes": [ { "persona": "grid-and-power-planner", "model": "openai/gpt-5.6-sol", "p": 0.47, "rationale": "I use rapidly iterated AI benchmarks facing measurement saturation as the reference class. Moving from 17.4 to 160 hours requires about 3.2 doublings; the quoted 4–7-month trend makes underlying capability by end-2028 plausible. Certification is harder: METR must add sufficiently long tasks, obtain costly human baselines, evaluate a public model, deem the result reliable and in-range, and publish the point estimate. A longer task suite alone does not resolve MISS. With no new indicator data, I assign certification hazards of roughly 0.03 in late 2026, 0.20 in 2027, and 0.38 in 2028. Compounding, plus about 0.02 risk that publication ceases, leaves 0.47 for HIT." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.54, "rationale": "Starting from the supplied 17.4-hour estimate, 160 hours requires about 3.2 doublings. At METR’s cited 4–7-month pace, raw extrapolation crosses the threshold around late 2027 to mid-2028. Capability alone is insufficient, however: METR must extend its roughly 16-hour measurement ceiling, validate long-baseline tasks, test a public model, and publish an in-range 50% point estimate by the cutoff. Using METR benchmark-refresh cycles and capability extrapolations as the reference class, I assign conditional certification hazards near 0.18 through 2027 and 0.31 in 2028, compounded, plus a small adverse publication-discontinuation risk. Non-certification remains slightly more likely." }, { "persona": "enterprise-cio", "model": "openai/gpt-5.6-sol", "p": 0.62, "rationale": "I use the reference class of frontier-benchmark transitions: capability often outruns a benchmark, while redesign, validation, and publication take roughly 1–2 years. A 17.4-to-160-hour increase is about 3.2 doublings, so the quoted 4–7-month trend makes underlying capability plausible before 2029. But certification is conjunctive: METR must deploy longer-baseline tasks, obtain a reliable point estimate, test a publicly released model, and publish by the deadline. I assign about 0.35 cumulative probability to that sequence, plus roughly 0.03 to METR ceasing publication, yielding 0.62 for HIT. This is slightly below the forecaster’s 0.65 because over two budget years a suite expansion is quite feasible." }, { "persona": "credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.58, "rationale": "I use METR’s own frontier-model horizon progression and benchmark-refresh lag as the reference class. The supplied 4–7-month doubling trend can reach 160 hours before the cutoff, but certification additionally requires a substantially extended task suite, reliable in-range estimation, a public model, and timely publication. Treating qualifying certification as a rising hazard averaging roughly 0.20 annually over the remaining 2.3 years gives about 0.40 cumulative risk; a small probability that METR stops publishing also resolves against the statement. Missing indicator data provides no basis to raise confidence. The conjunction therefore makes HIT modestly more likely than MISS, but below the forecaster’s 0.65." }, { "persona": "geopolitics-policy-analyst", "model": "openai/gpt-5.6-sol", "p": 0.56, "rationale": "Using METR’s reported 4–7-month doubling trend as the capability reference class, moving from 17.4 to 160 hours requires about 3.2 doublings, implying a raw crossing around late 2027 to mid-2028. That weighs against the thesis. But resolution requires a conjunction: a public model, an upgraded suite capable of reliable month-scale measurement, a qualifying 50% point estimate, and publication by the cutoff. Benchmark redesign and long human baselines can lag capability by years; the current >16-hour unreliability warning is therefore material. With 27 months remaining, I put certification slightly below even odds. The indicators provide no update." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.57, "rationale": "Base rate: I use 0.45 for a fast-moving capability benchmark to publish a ninefold threshold crossing within 27 months when its fitted trend projects crossing inside that window (about 0.26 annual hazard). The 17.4-hour baseline and 4–7-month doubling trend increase certification risk. However, a MISS requires several timed legs: a substantially expanded task suite, a qualifying public model, a reliable 50%-success estimate, and METR publication by the deadline. The current ~16-hour reliability ceiling makes that measurement chain a major constraint. METR discontinuation adds a small adverse tail. With no new indicator data, 0.57 is below the forecaster’s 0.65." } ], "aggregation": "log-odds-mean", "extremization": 1, "dissent": null, "spread": 0.15, "warnings": [ "all 6 votes came from one vendor (openai); extremization disabled, persona diversity only" ] }, "final": { "p": 0.5570641685163671, "ci80": [ 0.4698505026668333, 0.6408960852292583 ], "method": "council-only:log-odds-mean", "weights": { "council": 1 } }, "stated": 0.65, "history": [ { "asOf": "2026-09-07", "p": 0.5733906163405034, "source": "radar-engine", "runId": "radar-2026-09-07T23-12-09Z-06a58f" }, { "asOf": "2026-09-07", "p": 0.5570641685163671, "source": "radar-engine", "runId": "radar-2026-09-07T23-49-04Z-de8abf" } ], "indicatorStatus": "no-data", "milestones": [ { "id": "m1", "enginePrior": 0.65, "councilP": null }, { "id": "m2", "enginePrior": 0.3888888888888889, "councilP": null }, { "id": "m3", "enginePrior": null, "councilP": null } ], "arena": "agi-capabilities", "status": "draft", "tier": "T2" }, "indicators": [ { "thesisId": "RT-16", "indicatorId": "li-1", "connector": "manual", "seriesId": "metr-th-50pct-max-hours", "unit": "hours", "asOf": "2026-09-07", "current": null, "status": "no-data", "history": [], "url": "https://metr.org/time-horizons/", "note": "manual: 'metr-th-50pct-max-hours' not in manual-series.json" }, { "thesisId": "RT-16", "indicatorId": "li-2", "connector": "manual", "seriesId": "metr-suite-max-task-hours", "unit": "hours", "asOf": "2026-09-07", "current": null, "status": "no-data", "history": [], "url": "https://metr.org/time-horizons/", "note": "manual: 'metr-suite-max-task-hours' not in manual-series.json" } ], "indicatorStatus": "no-data", "chainEvents": [ { "seq": 222, "ts": "2026-09-07", "kind": "repriced", "hash": "352abf957a15e96593171d1ca367403f44bae0a3b3732f9bbe0161ce60a32363" }, { "seq": 223, "ts": "2026-09-07", "kind": "repriced", "hash": "f25aab9d4dc6e1167ab79a4c7de0050bb5fe5fb90dddb50923ca2f5a155b30c3" }, { "seq": 270, "ts": "2026-09-08", "kind": "registered", "hash": "81aaee36f804a50faa9402ba2267e47bb8ba2a54a196448ac2e6f99ddf60a88a" } ], "url": "/predictions/agi-capabilities/RT-16" }, { "id": "RT-17", "title": "No S&P 500 filing blames AI for 10% cut", "arena": "agi-capabilities", "tier": "T2", "tag": "WARNING", "direction": "down", "status": "registered", "statement": "Through 2028-12-31 no S&P 500 company files a 10-K, 10-Q, or 8-K with the SEC that attributes a workforce reduction of 10% or more of its total headcount principally to AI or AI-enabled automation.", "resolutionRule": "Resolves MISS (rung 4 achieved) if a Form 10-K, 10-Q, or 8-K filed on EDGAR on or before 2028-12-31 by a company that is an S&P 500 constituent on the filing date (a) states a workforce reduction of 10% or more of total employees, or gives a headcount figure equal to 10% or more of the employee count in the company's most recent prior 10-K, and (b) in the same filing names AI, generative AI, or AI-enabled automation as the primary or principal driver of that reduction, or quantifies the AI-attributed portion at 10% or more of total headcount. Language naming AI as \"a factor\" or \"one of several factors\" without quantification does not count (Oracle's FY2026 10-K, which names AI as a cause of unquantified reductions, is the baseline and does not qualify). Resolves HIT if no qualifying filing exists. Score by SEC EDGAR full-text search on filing text; press releases not filed with the SEC do not count. If EDGAR full-text search is discontinued, the filings themselves control.", "resolutionSource": { "name": "SEC EDGAR full-text search", "url": "https://efts.sec.gov/LATEST/search-index?q=%22artificial%20intelligence%22%20%22workforce%22%20%22reduction%22&forms=10-K%2C10-Q%2C8-K" }, "ambiguity": "resolves-against-forecaster", "horizon": "2028-12-31", "p": 0.75, "ci80": [ 0.6, 0.88 ], "consensus": { "value": "As of August 2026 one S&P 500 annual report (Oracle, FY2026 10-K) names AI as a cause of workforce reductions without quantifying the share of its ~13% headcount decline; Challenger, Gray & Christmas counts AI among stated reasons in a rising share of announced US job cuts, and press trackers attribute roughly 45,000 2026 layoffs to AI with company attribution.", "impliedP": 0.55, "impliedBy": "The press narrative treats AI-attributed layoffs as established and rising; if one partial attribution (Oracle) became a quantified or principal-driver 10% attribution at even one of 500 companies over the next 28 months, the event fires. Extrapolating the tracker's Tier-1 count from one filer toward several makes the consensus-path chance of no qualifying filing roughly 0.55, i.e. the positive event near 0.45.", "source": "Axis Intelligence AI Layoff Tracker, documentary-attribution tier", "url": "https://axis-intelligence.com/ai-layoff-tracker/", "asOf": "2026-08-13", "note": "Rung 4 of the capability ladder, stated as the skeptical side. No prediction market covers SEC attribution language. My 0.75 rests on the legal asymmetry: naming AI as the principal driver of a 10% cut invites WARN Act, securities, and political scrutiny that \"efficiency\" language does not, so companies will keep the causation unquantified even when it is true. Sister thesis RT-20 (Fortune 100, 2029 horizon) covers the same event on a narrower universe and a later date." }, "leadingIndicators": [ { "id": "li-1", "label": "S&P 500 SEC filings naming AI as a cause of workforce reduction", "connector": "manual", "seriesId": "edgar-ai-workforce-reduction-filers-count", "unit": "count of companies", "onTrack": { "op": "<=", "value": 3 }, "offTrack": { "op": ">=", "value": 6 }, "url": "https://efts.sec.gov/LATEST/search-index?q=%22artificial%20intelligence%22%20%22workforce%22%20%22reduction%22&forms=10-K%2C10-Q%2C8-K", "note": "Count of distinct S&P 500 constituents whose 10-K, 10-Q, or 8-K filed since 2026-01-01 states that AI adoption caused or contributed to a workforce reduction, at any level of quantification. Thresholds are for 2027-12-31; many unquantified attributions make a quantified one more likely." }, { "id": "li-2", "label": "BLS employment, computer systems design and related services", "connector": "bls", "seriesId": "CES6054150001", "unit": "thousands of employees", "onTrack": { "op": ">", "value": 2450 }, "offTrack": { "op": "<=", "value": 2350 }, "url": "https://data.bls.gov/timeseries/CES6054150001", "note": "Seasonally adjusted all-employees series for NAICS 5415. A steep decline is the macro signature a principal-driver filing would describe. Thresholds are for the December 2027 reading." } ], "milestones": [ { "id": "m1", "date": "2027-12-31", "statement": "At least three S&P 500 companies name AI as a cause of workforce reductions in SEC filings, at any level of quantification.", "p": 0.6 }, { "id": "m2", "date": "2028-06-30", "statement": "An S&P 500 filing quantifies an AI-attributed workforce reduction at 5% or more of total headcount.", "p": 0.35 }, { "id": "m3", "date": "2028-12-31", "statement": "An S&P 500 filing attributes a 10%-or-greater workforce reduction principally to AI (rung 4 achieved).", "p": 0.25 } ], "falsifier": "An S&P 500 company files a 10-K, 10-Q, or 8-K on or before 2028-12-31 that meets both conditions of the resolution rule.", "whyItMatters": "Filed attribution is the only labor-market signal that has passed legal review, and it is the rung where capability becomes economics rather than benchmarks. If no company will write the causation down through 2028, either AI is not yet displacing headcount at the 10% scale or the disclosure regime hides it; both readings matter for anyone sizing enterprise adoption and the capacity behind it.", "whatWouldRaise": [ "Two or more companies retract or soften AI attribution after litigation or political pressure.", "Companies with large announced cuts (Oracle, Meta, Block) omit AI from the filed restructuring language.", "SEC comment letters press filers on unsubstantiated AI attribution." ], "whatWouldCut": [ "A tech-sector S&P 500 company quantifies AI-attributed reductions above 5% in a 2027 filing.", "Three or more S&P 500 filings name AI as a cause of reductions during 2027.", "A WARN Act notice or 8-K restructuring exhibit names AI as the principal reason." ], "namedEntities": [ "Oracle", "Meta", "Block", "SEC" ], "marketCall": false, "recusalCheck": { "passed": true, "keysChecked": [ "employer-leasing", "named-competitor-bookings", "customer-pipeline", "ticker" ] }, "signedBy": "Brian Letort", "registeredAt": "2026-09-08", "amendments": [], "ruleHash": "8c8b8757631d88279e570eb07c641223693baf26cec1d3f2c7733099a56368a4", "distance": 0.8979415932059581, "distanceLabel": "clearly above consensus", "engine": { "id": "RT-17", "prior": { "p": 0.02, "method": "drift-gbm-terminal:li-2", "note": "P(indicator li-2 satisfies onTrack > 2450.0 at 2028-12-31); drift -0.0136/yr, vol 0.00589/sqrt(yr) from 32 points; proxy for the thesis, not its rule" }, "council": { "p": 0.7340249473683669, "votes": [ { "persona": "grid-and-power-planner", "model": "openai/gpt-5.6-sol", "p": 0.72, "rationale": "I use large public-company restructuring disclosures as the reference class and estimate a pooled S&P 500 qualifying-event hazard near 0.13 annually. Compounded over roughly 2.3 years, that implies about 0.72 probability of no qualifying filing. The falsifier is unusually conjunctive: a 10% companywide cut, explicit AI causation, principal-driver or quantified language, qualifying form, and index membership. Oracle’s near miss shows the pathway is plausible but also demonstrates issuers’ preference for unquantified, multi-cause language. The modest decline in the supplied BLS employment series does not measure SEC attribution and adds little. Resolves-against-forecaster ambiguity modestly increases miss risk." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.76, "rationale": "I use the reference class of S&P 500 restructurings exceeding 10% of company headcount that also receive single-cause attribution in mandatory filings. Such cuts are uncommon, and explicit principal-cause language is rarer: issuers usually cite several strategic and macro factors. Oracle is a meaningful near-miss, raising the hazard, but it also illustrates the likely disclosure pattern. The supplied BLS series has fallen only modestly since March, not indicating a broad displacement shock. I estimate an aggregate qualifying-filing hazard near 0.11 annually over the remaining 2.3 years. The mechanical 0.02 prior is poorly matched because employment drift cannot model legal wording." }, { "persona": "enterprise-cio", "model": "openai/gpt-5.6-sol", "p": 0.77, "rationale": "Using large-company restructuring filings as the reference class, I estimate an average 0.11 annual falsifier hazard through the remaining 2.3 years; compounding implies about 0.23 probability of at least one qualifying filing. Cuts exceeding 10% occur, but explicit SEC attribution principally to AI is much rarer: issuers generally describe restructuring through cost, strategy, and efficiency language, while AI deployment, integration, controls, and change management delay measurable enterprise-wide substitution. Oracle is a meaningful near miss and raises the hazard, but its missing quantification illustrates the disclosure barrier. The modest decline in the supplied employment series does not establish company-level AI causation. I therefore discount the engine prior as mismatched to the filing-language resolution rule." }, { "persona": "credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.73, "rationale": "I use the reference class of S&P 500 issuers undertaking at least 10% workforce reductions, then require the additional conjunction of explicit principal AI causation and disclosure in a covered SEC filing. That language remains rare because restructuring filings typically cite cost reduction, strategy, or automation among multiple factors. I estimate roughly a 0.14 annual qualifying-event hazard; compounded over about 2.3 years, that implies a 0.27 miss probability and 0.73 hit probability. Oracle’s near-miss indicates rising risk, while the modest BLS employment decline is weak evidence. This is slightly below the forecaster’s 0.75 because filed 8-K exhibits and increasing AI-productivity claims broaden the paths to qualification." }, { "persona": "geopolitics-policy-analyst", "model": "openai/gpt-5.6-sol", "p": 0.71, "rationale": "I use the reference class of large S&P 500 restructurings and their typically cautious SEC causation language. Across roughly 2.3 years, I estimate a 0.15 annual hazard of at least one qualifying filing, implying about 0.71 survival. The event is strongly conjunctive: a 10% companywide cut, explicit principal AI causation or qualifying attribution, and disclosure in a specified filing while the issuer is an index constituent. Oracle’s near miss and increasing public AI-layoff attribution raise the hazard, but legal and political incentives favor unquantified “efficiency” language. The modest BLS decline is weak evidence. This is slightly below the stated 0.75 because one close precedent exists and only one filer is needed." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.71, "rationale": "Base rate: I use roughly one qualifying filing per 4,000 S&P 500 company-years; 10% companywide cuts are uncommon, and principal-AI attribution in an SEC filing is rarer. About 1,160 company-years remain, implying a 0.25 miss chance before adjustments and a 0.75 HIT chance. Oracle’s near-miss and increasing public AI-layoff attribution raise the hazard, while the same-filing, principal-driver, constituency, form, and hard-date conjunction sharply limits it; modest BLS employment erosion is weak evidence. Ambiguity against the forecaster trims the estimate. The 0.02 structural prior is poorly matched because aggregate employment drift does not model legal attribution language. No calibration table was supplied." } ], "aggregation": "log-odds-mean", "extremization": 1, "dissent": null, "spread": 0.06, "warnings": [ "all 6 votes came from one vendor (openai); extremization disabled, persona diversity only" ] }, "final": { "p": 0.19180255627202072, "ci80": [ 0.16612542866507377, 0.22039948294741282 ], "method": "stacked-fixed-weights (radar: prior 0.5, council 0.5; no resolved radar items to learn from)", "weights": { "council": 0.5, "prior": 0.5 } }, "stated": 0.75, "history": [ { "asOf": "2026-09-07", "p": 0.1876822330887209, "source": "radar-engine", "runId": "radar-2026-09-07T23-12-09Z-06a58f" }, { "asOf": "2026-09-07", "p": 0.19180255627202072, "source": "radar-engine", "runId": "radar-2026-09-07T23-49-04Z-de8abf" } ], "indicatorStatus": "mixed", "milestones": [ { "id": "m1", "enginePrior": 0.5, "councilP": null }, { "id": "m2", "enginePrior": 0.5, "councilP": null }, { "id": "m3", "enginePrior": null, "councilP": null } ], "arena": "agi-capabilities", "status": "draft", "tier": "T2" }, "indicators": [ { "thesisId": "RT-17", "indicatorId": "li-1", "connector": "manual", "seriesId": "edgar-ai-workforce-reduction-filers-count", "unit": "count of companies", "asOf": "2026-09-07", "current": null, "status": "no-data", "history": [], "url": "https://efts.sec.gov/LATEST/search-index?q=%22artificial%20intelligence%22%20%22workforce%22%20%22reduction%22&forms=10-K%2C10-Q%2C8-K", "note": "manual: 'edgar-ai-workforce-reduction-filers-count' not in manual-series.json" }, { "thesisId": "RT-17", "indicatorId": "li-2", "connector": "bls", "seriesId": "CES6054150001", "unit": "thousands of employees", "asOf": "2026-08-31", "current": 2362.7, "status": "between", "history": [ { "asOf": "2024-01-31", "value": 2446.9 }, { "asOf": "2024-02-29", "value": 2444.6 }, { "asOf": "2024-03-31", "value": 2439.8 }, { "asOf": "2024-04-30", "value": 2441 }, { "asOf": "2024-05-31", "value": 2440.6 }, { "asOf": "2024-06-30", "value": 2444.7 }, { "asOf": "2024-07-31", "value": 2442.6 }, { "asOf": "2024-08-31", "value": 2437.2 }, { "asOf": "2024-09-30", "value": 2437.6 }, { "asOf": "2024-10-31", "value": 2433.4 }, { "asOf": "2024-11-30", "value": 2429.2 }, { "asOf": "2024-12-31", "value": 2431.4 }, { "asOf": "2025-01-31", "value": 2426.3 }, { "asOf": "2025-02-28", "value": 2416.7 }, { "asOf": "2025-03-31", "value": 2408.9 }, { "asOf": "2025-04-30", "value": 2411.3 }, { "asOf": "2025-05-31", "value": 2410.3 }, { "asOf": "2025-06-30", "value": 2399.9 }, { "asOf": "2025-07-31", "value": 2395.7 }, { "asOf": "2025-08-31", "value": 2394 }, { "asOf": "2025-09-30", "value": 2387.6 }, { "asOf": "2025-10-31", "value": 2379.7 }, { "asOf": "2025-11-30", "value": 2381.5 }, { "asOf": "2025-12-31", "value": 2380.6 }, { "asOf": "2026-01-31", "value": 2374.2 }, { "asOf": "2026-02-28", "value": 2379.1 }, { "asOf": "2026-03-31", "value": 2369.6 }, { "asOf": "2026-04-30", "value": 2369.5 }, { "asOf": "2026-05-31", "value": 2370 }, { "asOf": "2026-06-30", "value": 2368.3 }, { "asOf": "2026-07-31", "value": 2363.6 }, { "asOf": "2026-08-31", "value": 2362.7 } ], "url": "https://data.bls.gov/timeseries/CES6054150001", "note": "BLS v1, last 10 years" } ], "indicatorStatus": "mixed", "chainEvents": [ { "seq": 224, "ts": "2026-09-07", "kind": "repriced", "hash": "8174d941dd81ab68be75319baca1d8e09d94c2dcc2e0d7ea9502faa1df04413c" }, { "seq": 225, "ts": "2026-09-07", "kind": "repriced", "hash": "f0bb73a451de87deeee9e96bea2633fac688e943ad821dc8b7983453c0b403c9" }, { "seq": 271, "ts": "2026-09-08", "kind": "registered", "hash": "4e592cd9cb7f2787baa8339443950bf93ad5f9740fa0280337f5c30830c467f0" } ], "url": "/predictions/agi-capabilities/RT-17" }, { "id": "RT-18", "title": "Enterprise AI use stalls under 30% of firms", "arena": "enterprise-absorption", "tier": "T1", "tag": "WARNING", "direction": "down", "status": "registered", "statement": "The share of US businesses reporting current AI use in the Census Bureau's Business Trends and Outlook Survey stays below 30% in every biweekly collection period through the end of 2027.", "resolutionRule": "Score against the national \"Current AI Use (last two weeks)\" estimate on the Census BTOS data page. The thesis hits if no national estimate with a reference period ending on or before 2027-12-31 is at or above 30.0%. A single published national estimate at or above 30.0% resolves the thesis as a miss. If Census changes the question wording again, use the estimate it labels as the headline AI-use rate. If the series is discontinued before 2027-12-31 with no successor estimate, resolve against the forecaster (miss).", "resolutionSource": { "name": "US Census Bureau, Business Trends and Outlook Survey (BTOS) data page", "url": "https://www.census.gov/hfp/btos/" }, "ambiguity": "resolves-against-forecaster", "horizon": "2027-12-31", "p": 0.55, "ci80": [ 0.38, 0.7 ], "consensus": { "value": "BTOS respondents expected about 24.5% AI use within six months as of the July 2026 collection (21.7% current plus 2.8 points expected). A straight line through the January to July 2026 gain (+4.0 points per six months) lands at roughly 33% by the end of 2027, above the 30% bar.", "impliedP": 0.35, "impliedBy": "A straight line through the 2026 gain (+4 points per six months) crosses 30% around July 2027, five months before the deadline, so the trend baseline gives the thesis (no reading at or above 30% through 2027) about one chance in three.", "source": "Census BTOS current and expected AI use, as summarized by CRE Daily from WeWork's analysis", "url": "https://www.credaily.com/briefs/ai-adoption-slows-concentrates-in-us-knowledge-sectors/", "asOf": "2026-08-25", "note": "No prediction market covers BTOS. The consensus figure is the survey population's own six-month expectation plus a linear extrapolation." }, "leadingIndicators": [ { "id": "li-1", "label": "BTOS national current AI use rate", "connector": "manual", "seriesId": "census-btos-ai-use-national", "unit": "percent", "onTrack": { "op": "<", "value": 26 }, "offTrack": { "op": ">=", "value": 30 }, "url": "https://www.census.gov/hfp/btos/", "note": "National share of businesses answering yes to current AI use in the last two weeks, from the BTOS dashboard \"AI Use\" panel (biweekly, wording revised Nov 2025). Read the latest national estimate; the downloadable data tables on the same page carry the history." }, { "id": "li-2", "label": "US nonfarm business labor productivity, quarterly growth", "connector": "bls", "seriesId": "PRS85006092", "unit": "percent-annualized", "onTrack": { "op": "<", "value": 2.5 }, "offTrack": { "op": ">=", "value": 3.5 }, "url": "https://data.bls.gov/timeseries/PRS85006092", "note": "BLS Productivity and Costs, nonfarm business sector output per hour, percent change from previous quarter at an annual rate, seasonally adjusted. A sustained jump above 3.5% would signal diffusion faster than this thesis assumes." } ], "milestones": [], "falsifier": "Any BTOS national current-AI-use estimate at or above 30.0% with a reference period on or before 2027-12-31.", "whyItMatters": "Nearly every capex model assumes enterprise adoption broadens fast. The Census series is the only nationally representative, biweekly read on whether firms outside the Information and Finance sectors are actually using AI in operations. If four in five businesses are still not using it at the end of 2027, the demand side of the buildout is narrower than the narrative.", "whatWouldRaise": [ "Two consecutive BTOS readings where the six-month expected-use gap shrinks below 2 points.", "Firms with 1 to 19 employees flat or down in AI use for two quarters while large firms keep rising.", "Census working papers showing adoption breadth (functions per firm) stalling among existing users." ], "whatWouldCut": [ "A national reading of 27% or higher before mid-2027.", "Retail Trade or Construction AI use crossing 20%, which would mean diffusion beyond knowledge sectors.", "Nonfarm productivity growth above 3.5% annualized for two consecutive quarters." ], "namedEntities": [ "US Census Bureau" ], "marketCall": false, "recusalCheck": { "passed": true, "keysChecked": [ "employer-leasing", "named-competitor-bookings", "customer-pipeline", "ticker" ] }, "signedBy": "Brian Letort", "registeredAt": "2026-09-08", "amendments": [], "ruleHash": "b8a9d06b0dc43f739e48b6633edbdd0ae341d92a7f00bb0495f71a96fd3da451", "distance": 0.819709903868375, "distanceLabel": "clearly above consensus", "engine": { "id": "RT-18", "prior": { "p": 0.5171498123650154, "method": "drift-bm-terminal:li-2", "note": "P(indicator li-2 satisfies onTrack < 2.5 at 2027-12-31); drift +0.534/yr, vol 5.64/sqrt(yr) from 10 points; proxy for the thesis, not its rule" }, "council": { "p": 0.39782784494188816, "votes": [ { "persona": "grid-and-power-planner", "model": "openai/gpt-5.6-sol", "p": 0.41, "rationale": "My reference class is S-curve adoption of broadly defined business software across a firm-count survey, where small businesses slow the later diffusion phase. From 21.7% in July 2026, avoiding the falsifier requires gains below 8.3 points over roughly 17 months. The recent six-month gain annualizes above that pace, while respondents’ six-month expectation implies about 5.6 points annually—almost exactly the pace needed to reach 30% by year-end 2027. Thus modest persistence or definitional broadening produces a crossing. Intention-realization shortfalls and diffusion deceleration support staying below, but the single-print rule and discontinuation penalty reduce that probability. The productivity indicator is weakly relevant." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.37, "rationale": "The relevant reference class is early-stage business-technology diffusion, where adoption usually keeps rising but survey-reported intentions overstate realized uptake. From 21.7%, the series needs only 8.3 points over roughly 16 months to falsify the thesis. The January–July 2026 pace would cross 30% during 2027; respondents’ slower six-month expectation still puts the threshold near year-end. Requiring every biweekly estimate to remain below 30% is conjunctive, and one noisy print or series discontinuation causes a miss. Weak productivity growth is only indirect evidence and does not materially offset the observed adoption trend. Thus I am below both the structural prior and the forecaster’s 0.55." }, { "persona": "enterprise-cio", "model": "openai/gpt-5.6-sol", "p": 0.43, "rationale": "Using enterprise-technology diffusion and Census survey persistence as the reference class, 21.7% must gain less than 8.3 points over roughly 17 months. The recent six-month gain of 4.0 points extrapolates past 30%, while respondents’ six-month expectation implies slower growth to 24.5%. Procurement, security, integration, and change-management constraints support further deceleration; weak 1.4% productivity growth provides little evidence of broad production deployment. However, BTOS measures reported use, not scaled productivity, so lightweight adoption can spread faster than enterprise transformation. The single-crossing rule and discontinuation clause also reduce the chance of a hit. My 0.43 is below the forecaster’s 0.55." }, { "persona": "credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.41, "rationale": "Using the reference class of self-reported enterprise technology diffusion, adoption usually slows after an initial surge, but the literal all-period condition is demanding. July 2026 current use was 21.7%; respondents’ six-month expectation implies roughly 24.5% by early 2027. Avoiding 30% then requires growth below about 5.5 points through the rest of 2027, versus 4.0 points in the preceding six months. Deceleration could achieve that, but continued diffusion, sampling volatility, or one broadened headline estimate can cause a miss. Compounding a low crossing hazard in late 2026 with a materially higher 2027 hazard, plus the discontinuation penalty, puts survival below even odds. The productivity indicator is weakly informative." }, { "persona": "geopolitics-policy-analyst", "model": "openai/gpt-5.6-sol", "p": 0.43, "rationale": "The July 2026 level of 21.7 requires an 8.3-point gain within roughly 17 months to trigger a miss. Recent growth of 4.0 points per six months would cross 30 during 2027, while respondents’ six-month expectations imply slower growth and a year-end value near the threshold. Using broad business-software diffusion as the reference class, adoption usually decelerates as expansion reaches smaller firms, but generative AI remains unusually fast-moving. Because one noisy print suffices, wording changes could raise the headline rate, and discontinuation also resolves as a miss, I put the cumulative miss probability modestly above one-half. Productivity data are weak corroboration either way." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.34, "rationale": "Base rate: 0.40 for a rising enterprise-technology survey series starting 8–10 points below a threshold and avoiding it for roughly 17 months. BTOS is already 21.7, up 4.0 points in six months; continued linear growth would cross 30 before the deadline. Respondents’ 24.5 six-month expectation implies slower growth, making crossing uncertain rather than inevitable. I reduce the estimate because one biweekly print suffices to falsify the thesis, sampling variation matters near 30, and discontinuation also resolves as a miss. The productivity indicator is only weakly relevant. The stated 0.55 underweights these repeated-opportunity and resolution risks." } ], "aggregation": "log-odds-mean", "extremization": 1, "dissent": null, "spread": 0.09, "warnings": [ "all 6 votes came from one vendor (openai); extremization disabled, persona diversity only" ] }, "final": { "p": 0.45687022157298535, "ci80": [ 0.24040651072541855, 0.7557798547363112 ], "method": "stacked-fixed-weights (radar: prior 0.5, council 0.5; no resolved radar items to learn from)", "weights": { "council": 0.5, "prior": 0.5 } }, "stated": 0.55, "history": [ { "asOf": "2026-09-07", "p": 0.45311897922639005, "source": "radar-engine", "runId": "radar-2026-09-07T23-12-09Z-06a58f" }, { "asOf": "2026-09-07", "p": 0.45687022157298535, "source": "radar-engine", "runId": "radar-2026-09-07T23-49-04Z-de8abf" } ], "indicatorStatus": "on-track", "milestones": [], "arena": "enterprise-absorption", "status": "draft", "tier": "T1" }, "indicators": [ { "thesisId": "RT-18", "indicatorId": "li-1", "connector": "manual", "seriesId": "census-btos-ai-use-national", "unit": "percent", "asOf": "2026-09-07", "current": null, "status": "no-data", "history": [], "url": "https://www.census.gov/hfp/btos/", "note": "manual: 'census-btos-ai-use-national' not in manual-series.json" }, { "thesisId": "RT-18", "indicatorId": "li-2", "connector": "bls", "seriesId": "PRS85006092", "unit": "percent-annualized", "asOf": "2026-06-30", "current": 1.4, "status": "on-track", "history": [ { "asOf": "2024-03-31", "value": 0.2 }, { "asOf": "2024-06-30", "value": 3.7 }, { "asOf": "2024-09-30", "value": 3.7 }, { "asOf": "2024-12-31", "value": 1.4 }, { "asOf": "2025-03-31", "value": -0.9 }, { "asOf": "2025-06-30", "value": 4.2 }, { "asOf": "2025-09-30", "value": 5.2 }, { "asOf": "2025-12-31", "value": 1.6 }, { "asOf": "2026-03-31", "value": 0.8 }, { "asOf": "2026-06-30", "value": 1.4 } ], "url": "https://data.bls.gov/timeseries/PRS85006092", "note": "BLS v1, last 10 years" } ], "indicatorStatus": "on-track", "chainEvents": [ { "seq": 226, "ts": "2026-09-07", "kind": "repriced", "hash": "bf42b1c2780511963be1766f5ac27f111403ba0f57bee4c7af7d93f66952cd35" }, { "seq": 227, "ts": "2026-09-07", "kind": "repriced", "hash": "aebad45aa9d09f287bdaa17fcae41e7ab6159584fe7310a197e24116efe48f8f" }, { "seq": 272, "ts": "2026-09-08", "kind": "registered", "hash": "2df01379b18ffd6f4e9ca5653abde83c2f390b5ec354179f64d5f8bc71ba38f0" } ], "url": "/predictions/enterprise-absorption/RT-18" }, { "id": "RT-19", "title": "Microsoft reports 100 million paid Copilot seats", "arena": "enterprise-absorption", "tier": "T2", "tag": "CALL", "direction": "up", "status": "registered", "statement": "Microsoft publicly states at least 100 million paid Microsoft 365 Copilot seats in an earnings release, earnings call, or SEC filing dated on or before 2028-12-31.", "resolutionRule": "Hit if a Microsoft earnings press release, prepared remarks or Q&A on an earnings call, or SEC filing dated on or before 2028-12-31 states a paid Microsoft 365 Copilot seat count of 100,000,000 or more (phrases such as \"over 100 million paid seats\" count). Consumer Copilot subscriptions, Copilot Chat users, GitHub Copilot seats, and agent counts do not count. If Microsoft stops disclosing a paid seat number before stating 100 million, resolve against the forecaster (miss).", "resolutionSource": { "name": "Microsoft Investor Relations, quarterly earnings releases and webcasts", "url": "https://www.microsoft.com/en-us/investor/earnings/" }, "ambiguity": "resolves-against-forecaster", "horizon": "2028-12-31", "p": 0.6, "ci80": [ 0.45, 0.74 ], "consensus": { "value": "Morgan Stanley (July 2026) models Microsoft 365 Copilot revenue of about $22.5 billion in FY2029, which at roughly $25 to $30 per seat per month implies 60 to 75 million seat-equivalents by mid-2029, including E7 and consumption revenue. Bernstein said the market expected \"over 25 million\" seats going into the FY26 Q4 print; Microsoft reported over 30 million.", "impliedP": 0.25, "impliedBy": "The sell-side revenue model implies 60 to 75 million seat-equivalents by mid-2029, six months after this thesis's deadline and short of 100 million; reaching 100 million by end-2028 would need adds well above the modelled path, so the model implies roughly one chance in four.", "source": "Morgan Stanley Copilot revenue forecast, as reported by Bitget News", "url": "https://www.bitget.com/amp/news/detail/12560605527166", "asOf": "2026-07-21", "note": "No sell-side house publishes a public seat-count forecast; the implied seat figure is derived from the revenue forecast and stated price points." }, "leadingIndicators": [ { "id": "li-1", "label": "Paid Microsoft 365 Copilot seats (latest disclosed)", "connector": "manual", "seriesId": "msft-m365-copilot-paid-seats", "unit": "millions", "onTrack": { "op": ">=", "value": 50 }, "offTrack": { "op": "<", "value": 40 }, "url": "https://www.microsoft.com/en-us/investor/earnings/", "note": "Paid seat count stated by Microsoft in the quarterly earnings press release or call (15M at FY26 Q2, 20M at FY26 Q3, 30M at FY26 Q4). Thresholds are set for the FY27 Q4 print in July 2027." }, { "id": "li-2", "label": "Microsoft 365 commercial paid seats", "connector": "manual", "seriesId": "msft-m365-commercial-paid-seats", "unit": "millions", "onTrack": { "op": ">=", "value": 480 }, "offTrack": { "op": "<", "value": 460 }, "url": "https://www.microsoft.com/en-us/investor/earnings/", "note": "Denominator for Copilot attach. Stated on the earnings call (about 464 million at FY26 Q4, growing about 6% a year). A shrinking base would cap the seat thesis." } ], "milestones": [ { "id": "m1", "date": "2027-07-31", "statement": "Microsoft states at least 50 million paid Microsoft 365 Copilot seats at or before its FY2027 Q4 earnings release.", "p": 0.7 }, { "id": "m2", "date": "2028-01-31", "statement": "Microsoft states at least 75 million paid Microsoft 365 Copilot seats at or before its FY2028 Q2 earnings release.", "p": 0.55 } ], "falsifier": "Microsoft's FY2028 Q2 disclosure (January 2028) shows fewer than 60 million paid seats, or Microsoft stops disclosing a seat count.", "whyItMatters": "Paid Copilot seats are the only enterprise AI adoption number a hyperscaler publishes on a fixed quarterly cadence. Crossing 100 million, about one in five commercial Microsoft 365 seats, would confirm that knowledge-work AI is being bought as a standard license, not a pilot. It also anchors the inference-demand side of Azure's capex.", "whatWouldRaise": [ "Net seat adds above 10 million in any single quarter.", "E7 bundle pricing that folds Copilot into the default enterprise SKU.", "Microsoft disclosing a customer count above 5,000 organizations with more than 50,000 seats each." ], "whatWouldCut": [ "Two consecutive quarters of sequential net adds below 5 million.", "Microsoft shifting disclosure from seats to agents or consumption without a seat figure.", "Published renewal or seat-reduction data from large tenants showing net churn." ], "namedEntities": [ "Microsoft" ], "marketCall": false, "recusalCheck": { "passed": true, "keysChecked": [ "employer-leasing", "named-competitor-bookings", "customer-pipeline", "ticker" ] }, "signedBy": "Brian Letort", "registeredAt": "2026-09-08", "amendments": [], "ruleHash": "5fa6267e74264669630621a885c523d4e8e2af70a5dd657456d9fbae88b68420", "distance": 1.504077396776274, "distanceLabel": "far above consensus", "engine": { "id": "RT-19", "prior": null, "council": { "p": 0.2915478048433887, "votes": [ { "persona": "grid-and-power-planner", "model": "openai/gpt-5.6-sol", "p": 0.29, "rationale": "The supplied baseline is just over 30 million seats in mid-2026. Reaching 100 million by year-end 2028 requires roughly 62% annualized growth for 2.5 years, unusually sustained for a high-priced enterprise add-on. My reference class is rapid enterprise-software seat expansion after initial large-customer adoption, where growth normally decelerates as penetration rises. The cited FY2029 revenue model implies only 60–75 million seat-equivalents after the resolution deadline and includes revenue that may not represent qualifying seats. The claim is also conjunctive: adoption must reach 100 million and Microsoft must explicitly disclose a qualifying paid-seat count. Those factors put me materially below the forecaster’s 0.6." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.33, "rationale": "Using the reference class of enterprise add-on adoption, growth usually decelerates as deployments move beyond early adopters. From the supplied FY2026 Q4 level of over 30 million, Microsoft must exceed 100 million by the likely October 2028 earnings print—a 3.3-fold increase in roughly nine quarters. The cited FY2029 revenue model implies only 60–75 million seat-equivalents even later, and includes non-seat revenue. Bundling or aggressive discounting could accelerate adoption, but resolution also requires Microsoft to keep publishing an unambiguous paid Microsoft 365 Copilot seat count. That disclosure conjunction materially lowers the probability below the forecaster’s 0.6." }, { "persona": "enterprise-cio", "model": "openai/gpt-5.6-sol", "p": 0.31, "rationale": "Using enterprise productivity add-ons as the reference class, the supplied FY2026 Q4 level of over 30 million must more than triple in roughly 29 months, requiring about 60% annualized seat growth despite procurement, security, integration, and change-management constraints. Growth normally decelerates after early adopters. The cited FY2029 revenue model implies only 60–75 million seat-equivalents by mid-2029, making 100 million by December 2028 an aggressive upside case. Resolution is additionally conjunctive: Microsoft must both achieve the count and explicitly disclose qualifying paid Microsoft 365 Copilot seats. Bundling could accelerate adoption, but could also make a clean seat metric less likely." }, { "persona": "credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.34, "rationale": "Using enterprise-software seat ramps and disclosure persistence as the reference class, the supplied FY2026 Q4 count of over 30 million must grow roughly 3.3x by December 2028, equivalent to about 1.6x annually despite an expanding absolute base. The quoted FY2029 revenue model implies only 60–75 million seat-equivalents by mid-2029 and includes non-seat revenue, materially below the threshold. Bundling, discounts, and Microsoft 365 distribution could produce faster seat growth than revenue implies, but resolution additionally requires an explicit qualifying disclosure. A shift toward E7, consumption, or agent metrics creates meaningful disclosure risk. No current indicator data corroborates the milestone path." }, { "persona": "geopolitics-policy-analyst", "model": "openai/gpt-5.6-sol", "p": 0.27, "rationale": "The supplied FY2026 Q4 baseline is over 30 million. Reaching 100 million by December 2028 requires roughly 3.3x growth, or about 60% annualized, despite enterprise rollout friction and possible saturation among suitable Microsoft 365 users. My reference class is large enterprise-software add-ons scaling from early adoption: growth can remain high, but successive doubling rates usually slow. The cited FY2029 revenue model implies only 60–75 million seat-equivalents later than the deadline and includes non-seat revenue, making 100 million an upside case. Resolution also requires Microsoft to keep publishing a specifically paid Microsoft 365 Copilot seat count; metric changes or bundling create additional miss risk even if adoption is strong." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.22, "rationale": "Base rate: about 0.30 for enterprise add-ons already at 30 million paid seats to triple within 2.5 years; I multiply by roughly 0.75 for timely, qualifying primary-source disclosure. Reaching 100 million requires approximately 3.3x growth, or 65% annualized through end-2028. Microsoft’s installed base supports an upward adjustment, but the quoted FY2029 model implies only 60–75 million seat-equivalents by mid-2029 and includes non-seat revenue. The literal contract adds substantial disclosure risk: Microsoft must print a qualifying paid-seat count in a named source. No indicator data corroborates acceleration. These factors leave the forecaster’s 0.60 materially overconfident." } ], "aggregation": "log-odds-mean", "extremization": 1, "dissent": null, "spread": 0.12, "warnings": [ "all 6 votes came from one vendor (openai); extremization disabled, persona diversity only" ] }, "final": { "p": 0.2915478048433887, "ci80": [ 0.22480546279951902, 0.36868121294385797 ], "method": "council-only:log-odds-mean", "weights": { "council": 1 } }, "stated": 0.6, "history": [ { "asOf": "2026-09-07", "p": 0.3203288084514032, "source": "radar-engine", "runId": "radar-2026-09-07T23-12-09Z-06a58f" }, { "asOf": "2026-09-07", "p": 0.2915478048433887, "source": "radar-engine", "runId": "radar-2026-09-07T23-49-04Z-de8abf" } ], "indicatorStatus": "no-data", "milestones": [ { "id": "m1", "enginePrior": 0.5, "councilP": null }, { "id": "m2", "enginePrior": 0.5, "councilP": null } ], "arena": "enterprise-absorption", "status": "draft", "tier": "T2" }, "indicators": [ { "thesisId": "RT-19", "indicatorId": "li-1", "connector": "manual", "seriesId": "msft-m365-copilot-paid-seats", "unit": "millions", "asOf": "2026-09-07", "current": null, "status": "no-data", "history": [], "url": "https://www.microsoft.com/en-us/investor/earnings/", "note": "manual: 'msft-m365-copilot-paid-seats' not in manual-series.json" }, { "thesisId": "RT-19", "indicatorId": "li-2", "connector": "manual", "seriesId": "msft-m365-commercial-paid-seats", "unit": "millions", "asOf": "2026-09-07", "current": null, "status": "no-data", "history": [], "url": "https://www.microsoft.com/en-us/investor/earnings/", "note": "manual: 'msft-m365-commercial-paid-seats' not in manual-series.json" } ], "indicatorStatus": "no-data", "chainEvents": [ { "seq": 228, "ts": "2026-09-07", "kind": "repriced", "hash": "664aacc06f90505454c24d107bca51e0a8c1cacebadd693dedf2e6368e32feae" }, { "seq": 229, "ts": "2026-09-07", "kind": "repriced", "hash": "7f908cba2193c84072d4d2973c13cd451f29e01fcefb935aa1d0c88dad591801" }, { "seq": 273, "ts": "2026-09-08", "kind": "registered", "hash": "cc8bffe630e97464f5412427d67264bfcad5c6822b7ae4388df989d381d87e2a" } ], "url": "/predictions/enterprise-absorption/RT-19" }, { "id": "RT-20", "title": "IT services payrolls fall below 2.3 million by 2029", "arena": "enterprise-absorption", "tier": "T2", "tag": "WARNING", "direction": "down", "status": "registered", "statement": "BLS payroll employment in computer systems design and related services (NAICS 5415, series CES6054150001, seasonally adjusted) prints at or below 2,300 thousand in at least one month with a reference period on or before 2029-11-30, about 6% below its 2024 peak, while BLS projects the industry to grow 15.8% over the decade.", "resolutionRule": "Score against BLS Current Employment Statistics series CES6054150001 (all employees, computer systems design and related services, seasonally adjusted, thousands). Hit if any monthly observation with a reference month of November 2029 or earlier is at or below 2,300.0 as first published in the Employment Situation release for that month (use the first-print value, not later benchmark revisions). Score on releases published on or before 2029-12-31. If BLS discontinues or redefines the series so that no NAICS 5415 seasonally adjusted payroll series is published, resolve against the forecaster (miss).", "resolutionSource": { "name": "BLS Current Employment Statistics, series CES6054150001", "url": "https://data.bls.gov/timeseries/CES6054150001" }, "ambiguity": "resolves-against-forecaster", "horizon": "2029-12-31", "p": 0.7, "ci80": [ 0.55, 0.82 ], "consensus": { "value": "BLS Employment Projections 2024 to 2034 show computer systems design and related services growing 15.8% (2,444.8 thousand to 2,831.6 thousand), the second-largest numeric gain of any industry. The actual series has fallen 25 consecutive months from 2,442.6 thousand (mid-2024) to 2,362.7 thousand (August 2026 first print).", "impliedP": 0.25, "impliedBy": "The official projection has the industry adding about 39 thousand jobs a year; a further 63 thousand decline by 2029 is the opposite sign. Allowing for the projection's own error band and the drift already in the data, the projection implies roughly one chance in four of a 2,300 print.", "source": "BLS Employment Projections, industries with the largest wage and salary employment growth, 2024-34", "url": "https://www.bls.gov/emp/tables/industries-large-grow-employment.htm", "asOf": "2025-08-28", "note": "No prediction market covers NAICS 5415 employment. The BLS projection is the most defensible public baseline and is the number this thesis argues against." }, "leadingIndicators": [ { "id": "li-1", "label": "Employment, computer systems design and related services", "connector": "bls", "seriesId": "CES6054150001", "unit": "thousands", "onTrack": { "op": "<=", "value": 2340 }, "offTrack": { "op": ">=", "value": 2380 }, "url": "https://data.bls.gov/timeseries/CES6054150001", "note": "BLS Current Employment Statistics, all employees, NAICS 5415, seasonally adjusted, thousands, monthly. 2,362.7 in August 2026. Thresholds apply to the June 2027 print; the thesis needs roughly 20 thousand of decline a year." }, { "id": "li-2", "label": "Software development job postings on Indeed (US)", "connector": "fred", "seriesId": "IHLIDXUSTPSOFTDEVE", "unit": "index-feb-2020-100", "onTrack": { "op": "<", "value": 65 }, "offTrack": { "op": ">=", "value": 85 }, "url": "https://fred.stlouisfed.org/series/IHLIDXUSTPSOFTDEVE", "note": "Indeed Hiring Lab index of US software development job postings, February 1, 2020 = 100, daily (use the monthly average). A leading read on IT services hiring intent; a rebound above 85 would signal demand returning ahead of headcount." }, { "id": "li-3", "label": "US nonfarm business labor productivity, quarterly growth", "connector": "bls", "seriesId": "PRS85006092", "unit": "percent-annualized", "onTrack": { "op": ">=", "value": 2.5 }, "offTrack": { "op": "<", "value": 1.5 }, "url": "https://data.bls.gov/timeseries/PRS85006092", "note": "BLS Productivity and Costs, nonfarm business output per hour, percent change from previous quarter at an annual rate. Sustained above-trend gains are consistent with AI substituting for IT services labor." } ], "milestones": [ { "id": "m1", "date": "2028-01-31", "statement": "CES6054150001 prints at or below 2,340 thousand for any month through December 2027.", "p": 0.7 }, { "id": "m2", "date": "2029-01-31", "statement": "CES6054150001 prints at or below 2,320 thousand for any month through December 2028.", "p": 0.65 } ], "falsifier": "CES6054150001 stays above 2,300 thousand in every month through November 2029, or turns up to regain its 2024 peak of about 2,443 thousand.", "whyItMatters": "IT services is the first large white-collar industry where AI does the work the workers were hired to do: writing, testing, and maintaining code and systems. BLS still projects it as the second-fastest source of new jobs in the country. If payrolls keep falling instead, the labor-market effect of coding agents shows up in a monthly federal series two years before it shows up in any company's 10-K.", "whatWouldRaise": [ "Three consecutive monthly declines of 5 thousand or more in CES6054150001.", "Large IT outsourcers reporting revenue growth with flat or falling headcount for two fiscal years.", "Indeed software development postings falling below 60 on the February 2020 index." ], "whatWouldCut": [ "Two consecutive months of net gains above 5 thousand in CES6054150001.", "An annual BLS benchmark revision that lifts the 2025-2026 level by more than 30 thousand.", "Enterprise AI deployment creating net new integration and systems work that outsourcers hire for." ], "namedEntities": [ "US Bureau of Labor Statistics" ], "marketCall": false, "recusalCheck": { "passed": true, "keysChecked": [ "employer-leasing", "named-competitor-bookings", "customer-pipeline", "ticker" ] }, "signedBy": "Brian Letort", "registeredAt": "2026-09-08", "amendments": [], "ruleHash": "7344ae0e0ea0471314d2118d37ed777575eb36cea20ac8ead91448daa316684d", "distance": 1.9459101490553132, "distanceLabel": "far above consensus", "engine": { "id": "RT-20", "prior": { "p": 0.98, "method": "drift-gbm-terminal:li-1", "note": "P(indicator li-1 satisfies onTrack <= 2340.0 at 2029-12-31); drift -0.0136/yr, vol 0.00589/sqrt(yr) from 32 points; proxy for the thesis, not its rule" }, "council": { "p": 0.6746454279721772, "votes": [ { "persona": "grid-and-power-planner", "model": "openai/gpt-5.6-sol", "p": 0.74, "rationale": "From 2,362.7, the series needs only another 62.7 thousand decline across 39 eligible reference months. The 25-month contraction and depressed software-development postings make a threshold touch plausible, although March–August 2026 lost just 6.9 thousand, indicating deceleration. My reference class is U.S. postwar recession entry, roughly 0.15 annual hazard; compounded over 3.25 years that is about 0.41, and a downturn would likely produce the needed drop. Without recession, persistent weak hiring still provides a material crossing chance, but BLS’s long-run growth projection supports eventual mean reversion. I discount the 0.98 GBM prior because industry payrolls are smooth and regime-switching, while resolution uses first prints." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.71, "rationale": "The threshold is only 62.7 thousand (2.7%) below August 2026 after a 79.9 thousand, 25-month contraction. Recent decline has slowed materially: March-August lost 6.9 thousand, a pace that would reach 2,300 slightly after the horizon. My reference class is prolonged industry-payroll contractions plus the compounded roughly three-year chance of a cyclical downturn; either renewed trend losses or one recessionary downswing could produce a qualifying monthly print. The weak software-job-posting index supports downside, while 1.4% productivity growth and BLS’s secular growth projection argue against assuming rapid displacement. I heavily discount the 0.98 drift-model prior because it extrapolates a nonstationary employment trend and underweights mean reversion." }, { "persona": "enterprise-cio", "model": "openai/gpt-5.6-sol", "p": 0.67, "rationale": "The series is only 62.7 thousand above the threshold after losing 79.9 thousand over 25 months. Crossing by November 2029 requires an average decline near 1.6 thousand monthly; the March-August 2026 pace was slightly slower, but the longer contraction was faster. My reference class is cyclical drawdowns in detailed professional-services payroll series: I use roughly a 0.25 annual crossing hazard conditional on an established contraction, compounded over 3.25 years, then adjust upward for proximity and the any-month rule. BLS’s decade growth projection supports eventual mean reversion but does not preclude an intervening trough. Slow enterprise AI absorption limits confidence in a sustained displacement story." }, { "persona": "credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.64, "rationale": "At 2,362.7, the series needs only a 62.7 thousand (2.7%) decline over 39 reference months. The 25-month fall totaled 79.9 thousand, while March–August 2026 still fell 6.9 thousand despite an interim plateau; weak software-development postings support continued softness. My reference class is the roughly 15% annual postwar U.S. recession-onset hazard, about 41% compounded over 3.25 years, augmented by the existing industry downtrend and the easier any-month barrier. Against that, modest productivity growth and BLS’s long-run expansion projection make stabilization plausible. I heavily discount the 0.98 drift prior because it extrapolates a finite downswing; I land below the forecaster’s 0.70." }, { "persona": "geopolitics-policy-analyst", "model": "openai/gpt-5.6-sol", "p": 0.64, "rationale": "Reference class: persistent CES industry contractions plus ordinary U.S. business-cycle downturns. I use an annual barrier-hit hazard near 0.24 over the remaining 3.25 years, about 0.59 compounded, then raise it because payrolls have already fallen 79.9 thousand and sit only 62.7 thousand above the threshold. Depressed software-development postings and renewed July-August declines support persistence. Against that, employment was nearly flat from March through June, measured productivity growth is modest, and BLS’s long-run projection implies substantial recovery. The any-month test helps, while the first-print requirement and series-discontinuation clause slightly reduce the chance. The engine’s 0.98 appears to over-extrapolate drift and underweight mean reversion." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.64, "rationale": "Reference-class base rate: about 0.45 for a detailed U.S. industry payroll series already contracting for two years to experience another 2.7% drawdown within roughly 39 months. The unusually persistent 25-month decline and depressed software-development postings raise the estimate; only 62.7 thousand additional losses are required, and any single-month breach counts. Against that, employment has fallen just 6.9 thousand since March 2026, productivity growth is modest, and the industry’s secular-growth baseline makes stabilization or recovery plausible. The engine’s 0.98 excessively extrapolates drift and understates mean reversion. Scheduled monthly publication makes disclosure failure a small risk." } ], "aggregation": "log-odds-mean", "extremization": 1, "dissent": null, "spread": 0.1, "warnings": [ "all 6 votes came from one vendor (openai); extremization disabled, persona diversity only" ] }, "final": { "p": 0.9097466953051365, "ci80": [ 0.8904333722332904, 0.9231223077583208 ], "method": "stacked-fixed-weights (radar: prior 0.5, council 0.5; no resolved radar items to learn from)", "weights": { "council": 0.5, "prior": 0.5 } }, "stated": 0.7, "history": [ { "asOf": "2026-09-07", "p": 0.9090085239058443, "source": "radar-engine", "runId": "radar-2026-09-07T23-12-09Z-06a58f" }, { "asOf": "2026-09-07", "p": 0.9097466953051365, "source": "radar-engine", "runId": "radar-2026-09-07T23-49-04Z-de8abf" } ], "indicatorStatus": "off-track", "milestones": [ { "id": "m1", "enginePrior": 0.65, "councilP": null }, { "id": "m2", "enginePrior": null, "councilP": null } ], "arena": "enterprise-absorption", "status": "draft", "tier": "T2" }, "indicators": [ { "thesisId": "RT-20", "indicatorId": "li-1", "connector": "bls", "seriesId": "CES6054150001", "unit": "thousands", "asOf": "2026-08-31", "current": 2362.7, "status": "between", "history": [ { "asOf": "2024-01-31", "value": 2446.9 }, { "asOf": "2024-02-29", "value": 2444.6 }, { "asOf": "2024-03-31", "value": 2439.8 }, { "asOf": "2024-04-30", "value": 2441 }, { "asOf": "2024-05-31", "value": 2440.6 }, { "asOf": "2024-06-30", "value": 2444.7 }, { "asOf": "2024-07-31", "value": 2442.6 }, { "asOf": "2024-08-31", "value": 2437.2 }, { "asOf": "2024-09-30", "value": 2437.6 }, { "asOf": "2024-10-31", "value": 2433.4 }, { "asOf": "2024-11-30", "value": 2429.2 }, { "asOf": "2024-12-31", "value": 2431.4 }, { "asOf": "2025-01-31", "value": 2426.3 }, { "asOf": "2025-02-28", "value": 2416.7 }, { "asOf": "2025-03-31", "value": 2408.9 }, { "asOf": "2025-04-30", "value": 2411.3 }, { "asOf": "2025-05-31", "value": 2410.3 }, { "asOf": "2025-06-30", "value": 2399.9 }, { "asOf": "2025-07-31", "value": 2395.7 }, { "asOf": "2025-08-31", "value": 2394 }, { "asOf": "2025-09-30", "value": 2387.6 }, { "asOf": "2025-10-31", "value": 2379.7 }, { "asOf": "2025-11-30", "value": 2381.5 }, { "asOf": "2025-12-31", "value": 2380.6 }, { "asOf": "2026-01-31", "value": 2374.2 }, { "asOf": "2026-02-28", "value": 2379.1 }, { "asOf": "2026-03-31", "value": 2369.6 }, { "asOf": "2026-04-30", "value": 2369.5 }, { "asOf": "2026-05-31", "value": 2370 }, { "asOf": "2026-06-30", "value": 2368.3 }, { "asOf": "2026-07-31", "value": 2363.6 }, { "asOf": "2026-08-31", "value": 2362.7 } ], "url": "https://data.bls.gov/timeseries/CES6054150001", "note": "BLS v1, last 10 years" }, { "thesisId": "RT-20", "indicatorId": "li-2", "connector": "fred", "seriesId": "IHLIDXUSTPSOFTDEVE", "unit": "index-feb-2020-100", "asOf": "2026-08-28", "current": 74.61, "status": "between", "history": [ { "asOf": "2026-06-30", "value": 73.45 }, { "asOf": "2026-07-01", "value": 73.48 }, { "asOf": "2026-07-02", "value": 73.56 }, { "asOf": "2026-07-03", "value": 73.8 }, { "asOf": "2026-07-04", "value": 73.87 }, { "asOf": "2026-07-05", "value": 74 }, { "asOf": "2026-07-06", "value": 74.12 }, { "asOf": "2026-07-07", "value": 74.15 }, { "asOf": "2026-07-08", "value": 74.37 }, { "asOf": "2026-07-09", "value": 74.48 }, { "asOf": "2026-07-10", "value": 74.55 }, { "asOf": "2026-07-11", "value": 74.79 }, { "asOf": "2026-07-12", "value": 74.98 }, { "asOf": "2026-07-13", "value": 75.22 }, { "asOf": "2026-07-14", "value": 75.47 }, { "asOf": "2026-07-15", "value": 75.55 }, { "asOf": "2026-07-16", "value": 75.65 }, { "asOf": "2026-07-17", "value": 75.6 }, { "asOf": "2026-07-18", "value": 75.5 }, { "asOf": "2026-07-19", "value": 75.41 }, { "asOf": "2026-07-20", "value": 75.4 }, { "asOf": "2026-07-21", "value": 75.31 }, { "asOf": "2026-07-22", "value": 75.35 }, { "asOf": "2026-07-23", "value": 75.36 }, { "asOf": "2026-07-24", "value": 75.49 }, { "asOf": "2026-07-25", "value": 75.56 }, { "asOf": "2026-07-26", "value": 75.65 }, { "asOf": "2026-07-27", "value": 75.75 }, { "asOf": "2026-07-28", "value": 75.91 }, { "asOf": "2026-07-29", "value": 75.94 }, { "asOf": "2026-07-30", "value": 75.74 }, { "asOf": "2026-07-31", "value": 75.45 }, { "asOf": "2026-08-01", "value": 75.16 }, { "asOf": "2026-08-02", "value": 74.87 }, { "asOf": "2026-08-03", "value": 74.45 }, { "asOf": "2026-08-04", "value": 74.06 }, { "asOf": "2026-08-05", "value": 73.74 }, { "asOf": "2026-08-06", "value": 73.67 }, { "asOf": "2026-08-07", "value": 73.65 }, { "asOf": "2026-08-08", "value": 73.7 }, { "asOf": "2026-08-09", "value": 73.75 }, { "asOf": "2026-08-10", "value": 73.84 }, { "asOf": "2026-08-11", "value": 73.94 }, { "asOf": "2026-08-12", "value": 74.1 }, { "asOf": "2026-08-13", "value": 74.25 }, { "asOf": "2026-08-14", "value": 74.37 }, { "asOf": "2026-08-15", "value": 74.41 }, { "asOf": "2026-08-16", "value": 74.46 }, { "asOf": "2026-08-17", "value": 74.47 }, { "asOf": "2026-08-18", "value": 74.54 }, { "asOf": "2026-08-19", "value": 74.49 }, { "asOf": "2026-08-20", "value": 74.49 }, { "asOf": "2026-08-21", "value": 74.57 }, { "asOf": "2026-08-22", "value": 74.65 }, { "asOf": "2026-08-23", "value": 74.79 }, { "asOf": "2026-08-24", "value": 74.82 }, { "asOf": "2026-08-25", "value": 74.82 }, { "asOf": "2026-08-26", "value": 74.79 }, { "asOf": "2026-08-27", "value": 74.72 }, { "asOf": "2026-08-28", "value": 74.61 } ], "url": "https://fred.stlouisfed.org/series/IHLIDXUSTPSOFTDEVE", "note": null }, { "thesisId": "RT-20", "indicatorId": "li-3", "connector": "bls", "seriesId": "PRS85006092", "unit": "percent-annualized", "asOf": "2026-06-30", "current": 1.4, "status": "off-track", "history": [ { "asOf": "2024-03-31", "value": 0.2 }, { "asOf": "2024-06-30", "value": 3.7 }, { "asOf": "2024-09-30", "value": 3.7 }, { "asOf": "2024-12-31", "value": 1.4 }, { "asOf": "2025-03-31", "value": -0.9 }, { "asOf": "2025-06-30", "value": 4.2 }, { "asOf": "2025-09-30", "value": 5.2 }, { "asOf": "2025-12-31", "value": 1.6 }, { "asOf": "2026-03-31", "value": 0.8 }, { "asOf": "2026-06-30", "value": 1.4 } ], "url": "https://data.bls.gov/timeseries/PRS85006092", "note": "BLS v1, last 10 years" } ], "indicatorStatus": "off-track", "chainEvents": [ { "seq": 230, "ts": "2026-09-07", "kind": "repriced", "hash": "f3cffa8426fa91c2c027918686e5d5434bb5fcf625f0492a6886ab12ba6a35e8" }, { "seq": 231, "ts": "2026-09-07", "kind": "repriced", "hash": "a29af5cf1ab52b2df2aed01d6b456587d47f9bb84e1ba08c2e528a35c38a9b60" }, { "seq": 274, "ts": "2026-09-08", "kind": "registered", "hash": "4cd5fb72f8196641e9534880aa48919589e6ab47d79c8265a25d3a72fe45b914" } ], "url": "/predictions/enterprise-absorption/RT-20" }, { "id": "RT-21", "title": "Customer service jobs fall 30% by 2032", "arena": "enterprise-absorption", "tier": "T3", "tag": "THESIS", "direction": "down", "status": "registered", "statement": "BLS national employment of customer service representatives (SOC 43-4051) in the May 2032 Occupational Employment and Wage Statistics is at least 30% below the May 2025 level of 2,595,750, that is below 1,817,000.", "resolutionRule": "Score against the national cross-industry employment estimate for SOC 43-4051 in the May 2032 OEWS release (published by BLS in spring 2033). Hit if the estimate is 1,817,000 or lower. If BLS revises the SOC code so that 43-4051 no longer exists, use the successor occupation BLS maps it to in its crosswalk; if no single successor exists, or the May 2032 estimate is not published by 2033-06-30, resolve against the forecaster (miss).", "resolutionSource": { "name": "BLS Occupational Employment and Wage Statistics, 43-4051 Customer Service Representatives", "url": "https://www.bls.gov/oes/current/oes434051.htm" }, "ambiguity": "resolves-against-forecaster", "horizon": "2033-06-30", "p": 0.45, "ci80": [ 0.3, 0.6 ], "consensus": { "value": "BLS Employment Projections 2024 to 2034 show customer service representative employment declining 5.5% (about 153,700 jobs) over the decade, with about 341,700 openings a year from replacement demand.", "impliedP": 0.1, "impliedBy": "The official path lands near 2.45 million in 2032, about 630 thousand above this thesis's 1.817 million bar. A 30% fall is roughly six times the projected rate and outside any BLS revision in the occupation's history, so the projection implies about one chance in ten.", "source": "BLS Occupational Outlook Handbook, Customer Service Representatives", "url": "https://www.bls.gov/ooh/office-and-administrative-support/customer-service-representatives.htm", "asOf": "2026-09-07", "note": "The official projection is the most defensible public baseline; no prediction market covers this occupation. A 30% decline is roughly six times the BLS projected rate." }, "leadingIndicators": [ { "id": "li-1", "label": "OEWS national employment, customer service representatives", "connector": "bls", "seriesId": "OEUN000000000000043405101", "unit": "employees", "onTrack": { "op": "<=", "value": 2400000 }, "offTrack": { "op": ">=", "value": 2600000 }, "url": "https://www.bls.gov/oes/current/oes434051.htm", "note": "BLS OEWS national cross-industry employment for SOC 43-4051, annual May reference period, published the following spring. Series id follows the OEWS convention (OEU, N for national, all-industry, occupation 434051, data type 01 employment); confirm the id on data.bls.gov before wiring." }, { "id": "li-2", "label": "US nonfarm business labor productivity, quarterly growth", "connector": "bls", "seriesId": "PRS85006092", "unit": "percent-annualized", "onTrack": { "op": ">=", "value": 2.5 }, "offTrack": { "op": "<", "value": 1.5 }, "url": "https://data.bls.gov/timeseries/PRS85006092", "note": "Nonfarm business output per hour, percent change from prior quarter at an annual rate. Sustained gains above trend are consistent with substitution in high-volume service occupations." } ], "milestones": [ { "id": "m1", "date": "2028-04-30", "statement": "May 2027 OEWS national employment for 43-4051 is below 2,350,000 (a decline of at least 9.5% from May 2025).", "p": 0.6 }, { "id": "m2", "date": "2030-04-30", "statement": "May 2029 OEWS national employment for 43-4051 is below 2,100,000 (a decline of at least 19% from May 2025).", "p": 0.55 }, { "id": "m3", "date": "2032-04-30", "statement": "May 2031 OEWS national employment for 43-4051 is below 1,950,000 (a decline of at least 25% from May 2025).", "p": 0.5 } ], "falsifier": "May 2029 OEWS employment for 43-4051 at or above 2,400,000, which would put the occupation on the BLS projected path rather than this one.", "whyItMatters": "Customer service is the largest occupation where current agents already do the whole task rather than assist a human. It is also the cleanest public test of whether cheap intelligence removes jobs or just changes them. A 30% decline in one occupation of 2.6 million people would be the first labor-market fact large enough to move policy.", "whatWouldRaise": [ "May 2027 OEWS showing a decline of more than 12% from May 2025.", "Two or more Fortune 500 filings tying contact-center headcount cuts to AI deployment.", "Call-center outsourcers reporting falling seat counts with rising revenue per seat." ], "whatWouldCut": [ "May 2027 OEWS flat or up versus May 2025.", "Regulation requiring a human option for customer service in large sectors (banking, health, utilities).", "Evidence that AI agents raise contact volume enough to hold headcount flat (a Jevons effect in service)." ], "namedEntities": [ "US Bureau of Labor Statistics" ], "marketCall": false, "recusalCheck": { "passed": true, "keysChecked": [ "employer-leasing", "named-competitor-bookings", "customer-pipeline", "ticker" ] }, "signedBy": "Brian Letort", "registeredAt": "2026-09-08", "amendments": [], "ruleHash": "cacb7e75af623416173a3120081e4046c325d866d915cb898cb4890fc4d64774", "distance": 1.9965538818740678, "distanceLabel": "far above consensus", "engine": { "id": "RT-21", "prior": { "p": 0.5700899571293079, "method": "drift-bm-terminal:li-2", "note": "P(indicator li-2 satisfies onTrack >= 2.5 at 2033-06-30); drift +0.534/yr, vol 5.64/sqrt(yr) from 10 points; proxy for the thesis, not its rule" }, "council": { "p": 0.2315314991635887, "votes": [ { "persona": "grid-and-power-planner", "model": "openai/gpt-5.6-sol", "p": 0.24, "rationale": "The threshold requires roughly a 5.0% compounded annual decline for seven years, versus about 0.6% annually in BLS’s 2024–34 projection. My reference class is large, broadly defined service occupations exposed to automation: sustained 5% annual contraction is uncommon without major task elimination, offshoring, or reclassification. Generative AI creates a meaningful severe-decline tail, but customer-service demand growth, human escalation work, and slow enterprise adoption should preserve many positions. The supplied 1.4% aggregate productivity reading is off-track and weak occupation-specific evidence. I also apply a small resolution penalty for publication or SOC-crosswalk failure. This is materially below the stated 0.45." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.29, "rationale": "The threshold requires a 30% decline in seven years, about 5.0% compounded annually. My reference class is BLS occupational projections and historically gradual adjustment in large, broadly distributed administrative occupations. BLS projects only a 5.5% decade decline, while the supplied 1.4% productivity reading is not occupation-specific and gives no evidence of the required displacement pace. Generative-AI adoption creates a meaningful downside tail, and the 2025 level may already reflect some contraction, but hitting every implied milestone requires sustained rapid cuts despite replacement demand. SOC-successor ambiguity and publication timing modestly reduce resolution probability. This is below the forecaster’s 0.45." }, { "persona": "enterprise-cio", "model": "openai/gpt-5.6-sol", "p": 0.17, "rationale": "I use the BLS 2024–34 occupational projection as the reference-class baseline. Reaching 1.817 million requires about a 5.0% compounded annual decline over seven years, versus roughly 0.6% annually in that baseline. The supplied 2025 OEWS level shows no decline yet, and 1.4% productivity growth does not indicate the economy-wide discontinuity needed. Generative AI creates a meaningful downside tail, especially for routine contacts, but procurement, integration, accuracy controls, labor turnover and demand growth usually delay realized headcount reductions. The 2027 and 2029 milestones require unusually rapid early displacement. I also apply a small penalty for publication and occupational-crosswalk failure under the adverse ambiguity rule." }, { "persona": "credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.24, "rationale": "Using BLS’s occupation-level projection as the reference class, the required decline is unusually steep: employment must contract about 5.0% annually from May 2025 through May 2032, versus roughly 0.6% annually in the 2024–34 projection. Generative-AI substitution makes customer service more exposed than the typical occupation, but adoption, workflow redesign, regulation, and demand growth usually diffuse over multiple replacement cycles. The supplied OEWS indicator establishes only the baseline, while the 1.4 productivity reading does not demonstrate disruptive labor displacement. A 30% outcome therefore requires a sustained adverse tail path. Resolution-against-forecaster treatment for missing or fragmented successor data slightly reduces the probability." }, { "persona": "geopolitics-policy-analyst", "model": "openai/gpt-5.6-sol", "p": 0.29, "rationale": "Using BLS occupational projections and historical occupation declines as the reference class, the threshold is a tail outcome. Falling from 2,595,750 to 1,817,000 in seven years requires a compounded decline near 0.05 annually, versus about 0.0056 annually in BLS’s decade projection. Generative-AI substitution is unusually well targeted to this occupation and could accelerate after enterprise adoption. However, replacement demand, human escalation, demand expansion, and implementation frictions make sustained net losses of roughly 111,000 each year difficult. The supplied productivity reading of 0.014 annualized does not corroborate broad acceleration. Publication or successor-code ambiguity slightly reduces literal-resolution probability." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.18, "rationale": "Base rate: the official BLS 2024–34 projection is a 0.055 decade decline; large occupations projected for only modest contraction rarely lose 0.30 within seven years. The threshold requires roughly 0.050 compounded annual shrinkage from May 2025, versus about 0.006 implied by BLS. Generative-AI substitution justifies raising the tail probability, but the supplied indicators show no post-2025 occupational decline, and aggregate productivity is an indirect proxy. The 2027 and 2029 milestones require unusually rapid early losses. I also discount for the resolution’s adverse code-successor and publication-deadline conditions. The stated 0.45 substantially overweights the automation narrative." } ], "aggregation": "log-odds-mean", "extremization": 1, "dissent": null, "spread": 0.12, "warnings": [ "all 6 votes came from one vendor (openai); extremization disabled, persona diversity only" ] }, "final": { "p": 0.38728649845822594, "ci80": [ 0.06091897672800104, 0.8228348825047006 ], "method": "stacked-fixed-weights (radar: prior 0.5, council 0.5; no resolved radar items to learn from)", "weights": { "council": 0.5, "prior": 0.5 } }, "stated": 0.45, "history": [ { "asOf": "2026-09-07", "p": 0.3977616248336444, "source": "radar-engine", "runId": "radar-2026-09-07T23-12-09Z-06a58f" }, { "asOf": "2026-09-07", "p": 0.38728649845822594, "source": "radar-engine", "runId": "radar-2026-09-07T23-49-04Z-de8abf" } ], "indicatorStatus": "off-track", "milestones": [ { "id": "m1", "enginePrior": 0.65, "councilP": null }, { "id": "m2", "enginePrior": null, "councilP": null }, { "id": "m3", "enginePrior": null, "councilP": null } ], "arena": "enterprise-absorption", "status": "draft", "tier": "T3" }, "indicators": [ { "thesisId": "RT-21", "indicatorId": "li-1", "connector": "bls", "seriesId": "OEUN000000000000043405101", "unit": "employees", "asOf": "2025-12-31", "current": 2595750, "status": "between", "history": [ { "asOf": "2025-12-31", "value": 2595750 } ], "url": "https://data.bls.gov/timeseries/OEUN000000000000043405101", "note": "BLS v1, last 10 years" }, { "thesisId": "RT-21", "indicatorId": "li-2", "connector": "bls", "seriesId": "PRS85006092", "unit": "percent-annualized", "asOf": "2026-06-30", "current": 1.4, "status": "off-track", "history": [ { "asOf": "2024-03-31", "value": 0.2 }, { "asOf": "2024-06-30", "value": 3.7 }, { "asOf": "2024-09-30", "value": 3.7 }, { "asOf": "2024-12-31", "value": 1.4 }, { "asOf": "2025-03-31", "value": -0.9 }, { "asOf": "2025-06-30", "value": 4.2 }, { "asOf": "2025-09-30", "value": 5.2 }, { "asOf": "2025-12-31", "value": 1.6 }, { "asOf": "2026-03-31", "value": 0.8 }, { "asOf": "2026-06-30", "value": 1.4 } ], "url": "https://data.bls.gov/timeseries/PRS85006092", "note": "BLS v1, last 10 years" } ], "indicatorStatus": "off-track", "chainEvents": [ { "seq": 232, "ts": "2026-09-07", "kind": "repriced", "hash": "2a36332fca1a0d3fb80bed8396174f58549b90db3e155dcb97800490125c1e60" }, { "seq": 233, "ts": "2026-09-07", "kind": "repriced", "hash": "62f92880283a5f0522363150d4fdf5a737ce5c718bfec8d303de3d7d83d45e51" }, { "seq": 275, "ts": "2026-09-08", "kind": "registered", "hash": "9a8394c1b42f351ef5f59abb3ca407c9873a6552a1be75ac1435107bc9a08d73" } ], "url": "/predictions/enterprise-absorption/RT-21" }, { "id": "RT-22", "title": "OpenAI weekly users stay under 1.5 billion", "arena": "consumer-commercial", "tier": "T1", "tag": "WARNING", "direction": "down", "status": "registered", "statement": "OpenAI does not publicly state a weekly active user count of 1.5 billion or more for ChatGPT or its combined products in any first-party disclosure dated on or before 2027-12-31.", "resolutionRule": "Score against first-party OpenAI statements only: openai.com blog posts and newsroom items, official OpenAI social accounts, named OpenAI executives quoted on the record, and any SEC filing by OpenAI or a successor issuer. The thesis hits if none of these, dated on or before 2027-12-31, states weekly active users (ChatGPT alone or all OpenAI products combined) of 1,500,000,000 or more. Any such statement resolves the thesis as a miss. If a statement gives an \"active users\" figure without specifying weekly or monthly, treat it as weekly (against the forecaster). Third-party estimates (Sensor Tower, Similarweb, press reports citing internal data) do not count either way.", "resolutionSource": { "name": "OpenAI newsroom and official announcements", "url": "https://openai.com/news/" }, "ambiguity": "resolves-against-forecaster", "horizon": "2027-12-31", "p": 0.7, "ci80": [ 0.55, 0.82 ], "consensus": { "value": "OpenAI reported 900 million weekly actives in late February 2026 and passed 1 billion on 2026-07-31, about seven months later than its own internal target. Continuing the 2026 pace (roughly 20 to 25 million net weekly actives per month) reaches about 1.35 to 1.45 billion by December 2027; the growth narrative in coverage still assumes acceleration.", "impliedP": 0.45, "impliedBy": "OpenAI's own plan (2.75 billion weekly actives by 2030) interpolates to about 1.45 billion at end-2027, right at the bar, and coverage assumes acceleration from here. Taken at face value the plan gives the thesis (no 1.5 billion statement through 2027) a little under even odds.", "source": "The Verge, reporting OpenAI's 1 billion weekly active users announcement", "url": "https://www.theverge.com/ai-artificial-intelligence/973791/openai-says-its-models-now-reach-more-than-1-billion-users", "asOf": "2026-07-31", "note": "No Metaculus, Polymarket, or Kalshi market exists on a 1.5 billion threshold; Manifold's WAU markets resolved at end-2025. The baseline is OpenAI's own disclosed trajectory." }, "leadingIndicators": [ { "id": "li-1", "label": "OpenAI stated weekly active users (latest first-party figure)", "connector": "manual", "seriesId": "openai-stated-weekly-active-users", "unit": "billions", "onTrack": { "op": "<", "value": 1.25 }, "offTrack": { "op": ">=", "value": 1.4 }, "url": "https://openai.com/news/", "note": "Most recent weekly active user figure stated by OpenAI in a blog post, newsroom item, or executive statement (0.9B Feb 2026, 1.0B Jul 2026). Thresholds are set for a mid-2027 reading." }, { "id": "li-2", "label": "Share of US adults who have used ChatGPT (Pew)", "connector": "manual", "seriesId": "pew-us-adults-ever-used-chatgpt", "unit": "percent", "onTrack": { "op": "<=", "value": 52 }, "offTrack": { "op": ">=", "value": 60 }, "url": "https://www.pewresearch.org/internet/2026/06/17/americans-and-ai-2026-chatbots-smart-devices-and-views-on-impact/", "note": "Pew Research Center annual Americans and AI survey, share of US adults who say they ever use ChatGPT (18% 2023, 23% 2024, 34% 2025, 44% 2026). Thresholds apply to the survey fielded in early 2027." } ], "milestones": [], "falsifier": "Any first-party OpenAI statement dated on or before 2027-12-31 of 1.5 billion or more weekly active users.", "whyItMatters": "Consumer AI is already mass market; the open question is whether one product keeps compounding or the category fragments across Gemini, Meta AI, Copilot, and Claude. OpenAI's own user curve is the single most cited input to its IPO valuation and to inference-capacity plans. Reaching 1 billion seven months late is the first public sign the curve is bending.", "whatWouldRaise": [ "No new first-party weekly active user figure for two consecutive quarters after the 1 billion announcement.", "Pew's 2027 survey showing US adult ChatGPT use below 50%.", "Gemini or Meta AI monthly actives growing faster than ChatGPT for two consecutive Alphabet or Meta earnings calls." ], "whatWouldCut": [ "A first-party figure of 1.2 billion or more weekly actives before 2027-03-31.", "A default-assistant distribution deal on a major phone or PC platform.", "OpenAI folding Codex, ChatGPT Work, and device users into one reported weekly figure that jumps the count." ], "namedEntities": [ "OpenAI" ], "marketCall": false, "recusalCheck": { "passed": true, "keysChecked": [ "employer-leasing", "named-competitor-bookings", "customer-pipeline", "ticker" ] }, "signedBy": "Brian Letort", "registeredAt": "2026-09-08", "amendments": [], "ruleHash": "e29ab5bf31d0f2307794af2b23e00ed3c2ef42d7c9bcdfeb0a47e936a667d101", "distance": 1.0479685558493548, "distanceLabel": "clearly above consensus", "engine": { "id": "RT-22", "prior": { "p": 0.5, "method": "reference-class:qtype:disclosure", "note": "ledger base rate, n=29, horizon 480d; the ledger has no multi-year history" }, "council": { "p": 0.6723102083216342, "votes": [ { "persona": "grid-and-power-planner", "model": "openai/gpt-5.6-sol", "p": 0.61, "rationale": "The reference class is billion-user consumer platforms, where adoption usually decelerates as the denominator expands. From 1 billion on 2026-07-31, OpenAI needs roughly 29 million net additions monthly through 2027, above the supplied 2026 pace of 20–25 million. However, the required acceleration is modest, combined products count, and a crossed round-number milestone would likely be disclosed. I assign an annual hazard near 0.30 for a qualifying statement; compounded over roughly 17 months, that implies about 0.39 miss risk and 0.61 probability the thesis holds." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.69, "rationale": "From the supplied 1.0 billion level on 2026-07-31, reaching 1.5 billion by the horizon requires roughly 29 million net additions monthly, versus the cited 2026 pace of 20–25 million. My reference class is consumer platforms already near 1 billion users, where absolute growth usually plateaus or decelerates as penetration rises. New products, geographic expansion, or looser combined-product measurement could still produce acceleration. The adverse ambiguity rule also matters: an unspecified “active users” claim would falsify the thesis. If OpenAI genuinely crosses 1.5 billion, it is likely to publicize that milestone, so disclosure risk closely tracks underlying growth." }, { "persona": "enterprise-cio", "model": "openai/gpt-5.6-sol", "p": 0.66, "rationale": "From 1.0 billion on 2026-07-31, OpenAI must add roughly 29 million weekly users monthly through December 2027 to reach 1.5 billion. The supplied recent pace was about 20 million monthly, implying approximately 1.34 billion without acceleration. My reference class is consumer platforms already near one billion users: absolute growth can remain large, but percentage growth usually decelerates as penetration, competition, and measurement quality bind. This favors no qualifying disclosure. Risk remains substantial because only modest acceleration is required, combined-product counting broadens the metric, and the ambiguity rule treats cadence-unspecified “active users” against the thesis." }, { "persona": "credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.69, "rationale": "From the supplied 1.0 billion level on 2026-07-31, OpenAI needs roughly 29 million net weekly actives per month through December 2027 to reach 1.5 billion. The cited 2026 pace of 20–25 million implies only 1.34–1.43 billion, while the missed internal timetable argues against assuming acceleration. My reference class is maturing global consumer platforms, where absolute additions can remain large but penetration and retention constrain acceleration. I still assign meaningful falsification risk because the rule includes combined products, unspecified-period “active users,” and executive remarks, creating multiple disclosure paths if adoption strengthens." }, { "persona": "geopolitics-policy-analyst", "model": "openai/gpt-5.6-sol", "p": 0.7, "rationale": "Reference class: mature global consumer platforms after reaching roughly 1 billion active users, where penetration usually slows growth and milestone disclosures remain selective. From the supplied 1.0 billion on 2026-07-31, OpenAI needs another 500 million in 17 months—about 29 million monthly, above the supplied 2026 pace of 20–25 million. I assign roughly a 0.22 annual hazard of a qualifying first-party statement; compounded through the horizon, that implies about 0.30 miss risk. The adverse treatment of combined products and unspecified “active users” increases that risk, but actual threshold attainment and public disclosure must still occur by the deadline." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.68, "rationale": "Base rate: 0.50 from the supplied reference class for hard-dated disclosure questions. Starting from the reported 1.0 billion level in July 2026, reaching 1.5 billion by year-end 2027 requires roughly 29 million net additions monthly. The cited 2026 pace of 20–25 million implies only about 1.34–1.43 billion, while the seven-month target delay argues against assuming acceleration. Saturation at billion-user scale also favors deceleration. The thesis additionally wins if the threshold is reached but not disclosed by the deadline. Against this, OpenAI has publicized milestones, and unspecified “active users” counts are interpreted adversely. No fresh indicator data supports changing the trajectory." } ], "aggregation": "log-odds-mean", "extremization": 1, "dissent": null, "spread": 0.09, "warnings": [ "all 6 votes came from one vendor (openai); extremization disabled, persona diversity only" ] }, "final": { "p": 0.5888773223200136, "ci80": [ 0.4868290016600418, 0.6838125581208002 ], "method": "stacked-fixed-weights (radar: prior 0.5, council 0.5; no resolved radar items to learn from)", "weights": { "council": 0.5, "prior": 0.5 } }, "stated": 0.7, "history": [ { "asOf": "2026-09-07", "p": 0.585903334817376, "source": "radar-engine", "runId": "radar-2026-09-07T23-12-09Z-06a58f" }, { "asOf": "2026-09-07", "p": 0.5888773223200136, "source": "radar-engine", "runId": "radar-2026-09-07T23-49-04Z-de8abf" } ], "indicatorStatus": "no-data", "milestones": [], "arena": "consumer-commercial", "status": "draft", "tier": "T1" }, "indicators": [ { "thesisId": "RT-22", "indicatorId": "li-1", "connector": "manual", "seriesId": "openai-stated-weekly-active-users", "unit": "billions", "asOf": "2026-09-07", "current": null, "status": "no-data", "history": [], "url": "https://openai.com/news/", "note": "manual: 'openai-stated-weekly-active-users' not in manual-series.json" }, { "thesisId": "RT-22", "indicatorId": "li-2", "connector": "manual", "seriesId": "pew-us-adults-ever-used-chatgpt", "unit": "percent", "asOf": "2026-09-07", "current": null, "status": "no-data", "history": [], "url": "https://www.pewresearch.org/internet/2026/06/17/americans-and-ai-2026-chatbots-smart-devices-and-views-on-impact/", "note": "manual: 'pew-us-adults-ever-used-chatgpt' not in manual-series.json" } ], "indicatorStatus": "no-data", "chainEvents": [ { "seq": 234, "ts": "2026-09-07", "kind": "repriced", "hash": "f9c30fd9051ca248be97300a140044e84b5cf70997c8d12e13052b7c6d8a2c80" }, { "seq": 235, "ts": "2026-09-07", "kind": "repriced", "hash": "336146ea32b2054fab8e24c1cdccb6b2bfdcb5cb3f95e1878a612497ce4f46bc" }, { "seq": 276, "ts": "2026-09-08", "kind": "registered", "hash": "5482485343377cf8cc17eb12c3d68e0fe19f94d112293c376e5fe9e6b8d7fbb8" } ], "url": "/predictions/consumer-commercial/RT-22" }, { "id": "RT-23", "title": "Google Search revenue never shrinks through 2032", "arena": "consumer-commercial", "tier": "T3", "tag": "THESIS", "direction": "up", "status": "registered", "statement": "Alphabet's reported Google Search & other revenue grows year over year in every fiscal year from 2026 through 2032, with no annual decline despite AI assistants taking query share.", "resolutionRule": "Score against the \"Google Search & other\" revenue line in Alphabet's Form 10-K for each fiscal year 2026 through 2032, as filed on EDGAR (CIK 0001652044). The thesis hits if each fiscal year's figure exceeds the prior fiscal year's figure as reported in that same 10-K (using the restated prior-year comparative if Alphabet restates). A single fiscal year with a year-over-year decline resolves the thesis as a miss at the time that 10-K is filed. If Alphabet stops reporting a Search & other line and no equivalent search-advertising line is disclosed, resolve against the forecaster (miss).", "resolutionSource": { "name": "Alphabet Inc. Form 10-K filings on SEC EDGAR (CIK 0001652044)", "url": "https://www.sec.gov/cgi-bin/browse-edgar?action=getcompany&CIK=0001652044&type=10-K&dateb=&owner=include&count=40" }, "ambiguity": "resolves-against-forecaster", "horizon": "2033-03-31", "p": 0.7, "ci80": [ 0.55, 0.82 ], "consensus": { "value": "Search & other grew 17% year over year in Q2 2026 ($63.3 billion), the first deceleration after four accelerating quarters. eMarketer (February 2026) projects Google's share of US search ad spend falls below 50% in 2026 for the first time in two decades; Semrush projects AI-search visitors overtake traditional search visitors by 2028. Sell-side models still assume positive Search growth through 2028 but not beyond.", "impliedP": 0.4, "impliedBy": "Sell-side models carry positive Search growth through 2028 but the disruption forecasts (Google under 50% of US search ad spend in 2026, AI search visitors overtaking classic search by 2028) imply at least one down year before 2032 is the modal outcome. Read together they give seven straight growth years about a 40% chance.", "source": "LumiRank summary of eMarketer, WordStream, and Alphabet Q2 2026 disclosures", "url": "https://lumirank.ca/journal/google-search-ads-cpc-inflation-2026", "asOf": "2026-08-01", "note": "No prediction market prices a multi-year Search revenue decline. The consensus here is the disruption narrative in trade coverage versus the company's reported growth; the two disagree, which is the point of the thesis." }, "leadingIndicators": [ { "id": "li-1", "label": "Google Search & other revenue, year-over-year growth (latest quarter)", "connector": "manual", "seriesId": "googl-search-and-other-yoy-growth", "unit": "percent", "onTrack": { "op": ">=", "value": 8 }, "offTrack": { "op": "<", "value": 3 }, "url": "https://abc.xyz/investor/", "note": "Year-over-year growth of the \"Google Search & other\" line in Alphabet's quarterly earnings release (Exhibit 99.1 to the 8-K on EDGAR). Q1 2026 19%, Q2 2026 17%." }, { "id": "li-2", "label": "Google global search engine market share (StatCounter)", "connector": "manual", "seriesId": "statcounter-google-global-search-share", "unit": "percent", "onTrack": { "op": ">=", "value": 85 }, "offTrack": { "op": "<", "value": 80 }, "url": "https://gs.statcounter.com/search-engine-market-share", "note": "StatCounter Global Stats, all platforms, worldwide, monthly. About 89.5% in July 2026, down from a 92.9% peak in 2023. Revenue can grow with falling share; a break below 80% would test that." } ], "milestones": [ { "id": "m1", "date": "2028-02-15", "statement": "Alphabet's FY2027 10-K shows Google Search & other revenue up at least 5% over FY2026.", "p": 0.8 }, { "id": "m2", "date": "2030-02-15", "statement": "Alphabet's FY2028 and FY2029 10-Ks each show Google Search & other revenue above the prior year.", "p": 0.75 }, { "id": "m3", "date": "2032-02-15", "statement": "Alphabet's FY2030 and FY2031 10-Ks each show Google Search & other revenue above the prior year.", "p": 0.7 } ], "falsifier": "Any Alphabet 10-K for fiscal 2026 through 2032 reporting Google Search & other revenue below the prior fiscal year.", "whyItMatters": "The largest single advertising line in the world is the test case for whether AI answers destroy or absorb the commercial intent they sit on top of. If Search keeps growing while its query share falls, AI monetization accrues to whoever owns distribution, and the ad-funded model for consumer AI is real. If it shrinks, the revenue case for consumer inference gets much harder for everyone, including the labs.", "whatWouldRaise": [ "Search & other growth holding at or above 10% through the tougher Q3 and Q4 2026 comparisons.", "Alphabet disclosing AI Mode or AI Overviews monetization at or above classic search CPCs.", "Google's global query volume continuing to set records while StatCounter share drifts lower." ], "whatWouldCut": [ "Two consecutive quarters of Search & other growth below 5%.", "A US or EU remedy that unbundles default search placement on Android or Chrome.", "Alphabet reclassifying AI subscription revenue into Search & other, which would make the line uninterpretable." ], "namedEntities": [ "Alphabet", "Google" ], "marketCall": false, "recusalCheck": { "passed": true, "keysChecked": [ "employer-leasing", "named-competitor-bookings", "customer-pipeline", "ticker" ] }, "signedBy": "Brian Letort", "registeredAt": "2026-09-08", "amendments": [], "ruleHash": "c43e08d6f0aec263c9d3c11b226009d6931418d00256f0936ac64dfb22babb7d", "distance": 1.2527629684953676, "distanceLabel": "clearly above consensus", "engine": { "id": "RT-23", "prior": null, "council": { "p": 0.523621671155865, "votes": [ { "persona": "grid-and-power-planner", "model": "openai/gpt-5.6-sol", "p": 0.57, "rationale": "FY2026 appears highly likely to clear given the supplied 0.17 Q2 growth. My reference class is Alphabet’s own Search annual series, supplemented by mature digital-ad franchises over six-year windows: nominal revenue can rise despite share erosion, but recessions and platform shifts occasionally produce down years. I assign FY2026 about 0.98, then an average 0.08 annual miss hazard across FY2027–2032, rising later as AI-substitution uncertainty accumulates. Compounding, plus a small chance Alphabet removes or redefines the required line—which explicitly counts as failure—yields about 0.57. The missing indicators add no evidence. The strict seven-year conjunction makes 0.70 too high." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.58, "rationale": "I use Alphabet’s historical annual Search-revenue resilience and mature digital-ad platforms as the reference class. Reported 17% Q2 2026 growth makes an FY2026 decline unlikely, but the thesis then requires six more increases. I assign conditional annual decline hazards rising from roughly 4% to 11% through 2032, reflecting recession exposure, query substitution, and uncertain AI monetization. Compounding those hazards produces about 0.63 survival before accounting for the separate risk that Alphabet reorganizes disclosure and provides no equivalent line; the resolution rule makes that an automatic miss. This long conjunction therefore warrants materially less than 0.7 despite strong near-term momentum." }, { "persona": "enterprise-cio", "model": "openai/gpt-5.6-sol", "p": 0.59, "rationale": "Reference class: mature, dominant global advertising franchises and Alphabet’s own Search line, which has historically remained positive through sharp advertising slowdowns. The supplied 17% Q2 2026 growth makes FY2026 very likely to pass, while nominal market growth, pricing, and monetization of Google’s own AI traffic can offset query-share losses. But this is a seven-year conjunction: I estimate annual failure risk rising from roughly 5% in 2027 to 10% by 2032 as AI substitution, recession, regulation, and reclassification risk accumulate. Conditional compounding, including the adverse line-discontinuation rule, produces 0.59, below the stated 0.70." }, { "persona": "credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.48, "rationale": "The reference class is mature, dominant digital-advertising revenue lines exposed to both recessions and platform shifts. The supplied 17% Q2 2026 growth makes FY2026 decline unlikely, but the claim requires seven consecutive increases. I assign annual survival probabilities tapering from about 0.98 in 2026 to 0.84 by 2032 as AI substitution, pricing pressure, and cyclical advertising risk accumulate. Compounding produces roughly 0.49, then a small penalty for the explicit reporting-line-discontinuation rule. Nominal ad-market expansion and improved monetization can offset query-share losses, but one recession, monetization gap, or disclosure reorganization causes failure. The forecaster’s 0.70 insufficiently penalizes this conjunction." }, { "persona": "geopolitics-policy-analyst", "model": "openai/gpt-5.6-sol", "p": 0.48, "rationale": "I use the reference class of mature, advertising-funded technology platforms: nominal revenue usually rises, but recessionary ad contractions and product shifts create meaningful annual downside risk. FY2026 appears highly likely to clear the test given the supplied 17% Q2 growth, while risk rises as AI interfaces mature. Across seven conjunctive fiscal years, I estimate roughly a 9% average annual hazard of either a reported decline or loss of an equivalent disclosed line; compounding implies about 0.52 survival before allowing for correlated upside from successful AI monetization, yielding 0.48. Alphabet’s historical resilience and pricing power help, but one weak macro year or cannibalization episode is enough to fail. This is below the forecaster’s 0.7." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.44, "rationale": "Base rate: roughly 0.45 for a mature, dominant advertising segment to deliver seven consecutive nominal annual increases while retaining a comparable disclosed revenue line. Reported 17% Q2 2026 growth makes the first leg highly likely, and continued digital-ad expansion plus AI-result monetization provide upside. But this is a seven-leg conjunction: six later years remain exposed to recessions, search-share erosion, pricing pressure, and AI cannibalization. Even an approximately 0.89 annual success rate compounds materially below 0.7. The rule also counts removal of the Search & other line without an equivalent as failure, adding classification and disclosure risk over a long horizon." } ], "aggregation": "log-odds-mean", "extremization": 1, "dissent": null, "spread": 0.15, "warnings": [ "all 6 votes came from one vendor (openai); extremization disabled, persona diversity only" ] }, "final": { "p": 0.523621671155865, "ci80": [ 0.43648426627119025, 0.6093443531584097 ], "method": "council-only:log-odds-mean", "weights": { "council": 1 } }, "stated": 0.7, "history": [ { "asOf": "2026-09-07", "p": 0.493296809786029, "source": "radar-engine", "runId": "radar-2026-09-07T23-12-09Z-06a58f" }, { "asOf": "2026-09-07", "p": 0.523621671155865, "source": "radar-engine", "runId": "radar-2026-09-07T23-49-04Z-de8abf" } ], "indicatorStatus": "no-data", "milestones": [ { "id": "m1", "enginePrior": 0.5, "councilP": null }, { "id": "m2", "enginePrior": null, "councilP": null }, { "id": "m3", "enginePrior": null, "councilP": null } ], "arena": "consumer-commercial", "status": "draft", "tier": "T3" }, "indicators": [ { "thesisId": "RT-23", "indicatorId": "li-1", "connector": "manual", "seriesId": "googl-search-and-other-yoy-growth", "unit": "percent", "asOf": "2026-09-07", "current": null, "status": "no-data", "history": [], "url": "https://abc.xyz/investor/", "note": "manual: 'googl-search-and-other-yoy-growth' not in manual-series.json" }, { "thesisId": "RT-23", "indicatorId": "li-2", "connector": "manual", "seriesId": "statcounter-google-global-search-share", "unit": "percent", "asOf": "2026-09-07", "current": null, "status": "no-data", "history": [], "url": "https://gs.statcounter.com/search-engine-market-share", "note": "manual: 'statcounter-google-global-search-share' not in manual-series.json" } ], "indicatorStatus": "no-data", "chainEvents": [ { "seq": 236, "ts": "2026-09-07", "kind": "repriced", "hash": "f775ef12b40a80a739e0194f9e02fb1880bb162f987b19161702f7b028fef287" }, { "seq": 237, "ts": "2026-09-07", "kind": "repriced", "hash": "4f0aa5c0ff7eb28923121337715dde0b4da4d0994057196b0ff624c63f8da3fa" }, { "seq": 277, "ts": "2026-09-08", "kind": "registered", "hash": "24e5248dd7a76cb89f80906d604ec7957490db9ac5a4f8ebd8ca9d791b5f85f4" } ], "url": "/predictions/consumer-commercial/RT-23" }, { "id": "RT-24", "title": "OpenAI states $100 billion run rate by 2028", "arena": "consumer-commercial", "tier": "T2", "tag": "CALL", "direction": "up", "status": "registered", "statement": "OpenAI publicly states an annualized revenue run rate of at least $100 billion, or reports a quarter with revenue of at least $25 billion, in a first-party disclosure dated on or before 2028-12-31.", "resolutionRule": "Hit if, on or before 2028-12-31, (a) an openai.com post, official OpenAI statement, or named OpenAI executive on the record states annualized revenue or annualized run-rate revenue of $100 billion or more, or (b) an SEC filing by OpenAI or a successor issuer reports revenue of $25 billion or more for a single fiscal quarter. Figures attributed to \"people familiar with the matter\" or leaked investor decks do not count. If OpenAI reports in a currency other than USD, convert at the Federal Reserve H.10 rate on the statement date. If OpenAI makes no revenue statement after 2027-06-30, resolve against the forecaster (miss).", "resolutionSource": { "name": "OpenAI newsroom and official announcements (or SEC filings after listing)", "url": "https://openai.com/news/" }, "ambiguity": "resolves-against-forecaster", "horizon": "2028-12-31", "p": 0.6, "ci80": [ 0.45, 0.74 ], "consensus": { "p": 0.79, "value": "Manifold traders price OpenAI reaching $100 billion revenue in 2028 (the market accepts an annualized run rate of $100 billion at any point in 2028) at 79%. FutureSearch's median for the run rate at end of June 2027 is $64.5 billion; OpenAI's own plan reaches $100 billion in 2029. The run rate topped $40 billion in August 2026.", "source": "Manifold, OpenAI reaches $100B revenue in 2028?", "url": "https://manifold.markets/dreev/openai-reaches-100b-revenue-in-2028", "asOf": "2026-09-07", "note": "The Manifold question is a near match for this statement (run rate at any time in 2028 counts as YES). I sit below the crowd: a 2028 crossing needs about 50% compound growth from the August 2026 run rate, a year ahead of the company's own plan." }, "leadingIndicators": [ { "id": "li-1", "label": "OpenAI stated annualized revenue run rate", "connector": "manual", "seriesId": "openai-stated-annualized-revenue", "unit": "usd-billions", "onTrack": { "op": ">=", "value": 60 }, "offTrack": { "op": "<", "value": 45 }, "url": "https://openai.com/news/", "note": "Latest first-party annualized revenue figure ($20B end-2025, $40B+ August 2026). Thresholds apply to the mid-2027 reading; the thesis needs roughly a 60% annual growth rate from August 2026." }, { "id": "li-2", "label": "OpenAI stated paid subscribers", "connector": "manual", "seriesId": "openai-stated-paid-subscribers", "unit": "millions", "onTrack": { "op": ">=", "value": 75 }, "offTrack": { "op": "<", "value": 55 }, "url": "https://openai.com/news/", "note": "Paid ChatGPT subscribers as stated by OpenAI (over 50 million in March 2026). Enterprise now exceeds half of revenue, so subscriber count is the consumer leg only." } ], "milestones": [ { "id": "m1", "date": "2027-06-30", "statement": "OpenAI states an annualized revenue run rate of at least $60 billion.", "p": 0.6 }, { "id": "m2", "date": "2027-12-31", "statement": "OpenAI states an annualized revenue run rate of at least $75 billion.", "p": 0.5 } ], "falsifier": "OpenAI's latest stated run rate at 2027-12-31 is below $60 billion, which would require an implausible 2028 to reach $100 billion.", "whyItMatters": "OpenAI's revenue is the first real test of whether consumer and enterprise AI can pay for frontier-scale compute rather than borrow against it. A $100 billion run rate by 2028 would put one lab at roughly Meta's 2021 revenue after six years of product, and would validate the compute commitments already signed. Missing it by a wide margin would reprice every neocloud contract that leans on OpenAI as tenant.", "whatWouldRaise": [ "A first-party run rate at or above $55 billion by 2027-03-31.", "Advertising run rate above $5 billion, adding a second engine to subscriptions and API.", "An S-1 that shows GAAP quarterly revenue within 15% of the annualized run-rate claim." ], "whatWouldCut": [ "Two consecutive quarters of GAAP revenue growth below 10% sequentially.", "Enterprise share of revenue falling as Anthropic, Google, or Microsoft win coding and agent workloads.", "Repeated per-token price cuts on flagship models without offsetting volume growth." ], "namedEntities": [ "OpenAI" ], "marketCall": false, "recusalCheck": { "passed": true, "keysChecked": [ "employer-leasing", "named-competitor-bookings", "customer-pipeline", "ticker" ] }, "signedBy": "Brian Letort", "registeredAt": "2026-09-08", "amendments": [], "ruleHash": "fdbc2db10ac2566753dfd75cf3887c69085f4ecffc0bdf929aca4794278d8338", "distance": -0.9194603066354345, "distanceLabel": "clearly below consensus", "engine": { "id": "RT-24", "prior": null, "council": { "p": 0.5083443815917237, "votes": [ { "persona": "grid-and-power-planner", "model": "openai/gpt-5.6-sol", "p": 0.54, "rationale": "Reference class: private hypergrowth technology companies already at tens-of-billions revenue scale, where growth commonly decelerates as the denominator and infrastructure constraints rise. The supplied trajectory requires roughly 50% compound growth through 2028 and is reportedly one year ahead of plan. I assign about 0.64 to reaching the economic threshold, then 0.84 conditional probability of a qualifying first-party statement or SEC disclosure by the deadline, including the explicit silence risk; their product is about 0.54. The $60B and $75B milestones remain unresolved, and both indicator feeds are empty. Achievement alone does not satisfy the literal disclosure rule." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.52, "rationale": "My reference class is high-growth software/platform companies already at tens of billions in annualized revenue: sustaining roughly 50% compound growth for another two-plus years is uncommon as capacity, customer budgets, and inference economics constrain expansion. OpenAI could still outperform through enterprise adoption, API usage, and new products. The literal disclosure hurdle lowers the estimate: reaching an internal run rate is insufficient unless OpenAI or a named executive states it on the record, or an SEC filing prints a $25 billion quarter. The 2027 milestones imply substantial execution risk, and absent indicator data provides no basis to raise the estimate toward consensus." }, { "persona": "enterprise-cio", "model": "openai/gpt-5.6-sol", "p": 0.46, "rationale": "My reference class is high-growth software/platform businesses already operating at tens of billions in annual revenue: growth usually decelerates as enterprise procurement, security, integration, capacity, and adoption constraints compound. The supplied trajectory requires roughly 50% compound growth through 2028, materially above typical scale-stage persistence. OpenAI could outperform through consumer subscriptions, API usage, and enterprise contracts, but reaching the economics is insufficient: the resolution also requires a qualifying first-party statement or SEC-reported $25 billion quarter. Selective private-company disclosure and the resolves-against-forecaster wording reduce the probability further. With no current indicator readings, I place this below the forecaster and well below the quoted consensus." }, { "persona": "credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.52, "rationale": "Reference class: hypergrowth private software/platform companies attempting to scale from tens of billions to $100 billion of annualized revenue. The supplied trajectory requires roughly 1.5× annual growth through 2028—possible given AI adoption, but unusually difficult at this scale and dependent on compute availability, pricing, and enterprise conversion. I assign about 0.65 to economically reaching the threshold, conditional on the indicated 2026 run rate, and 0.80 to producing a qualifying first-party statement by the deadline; their conjunction is about 0.52. The disclosure requirement and adverse ambiguity rule make this materially harder than merely reaching $100 billion during 2028." }, { "persona": "geopolitics-policy-analyst", "model": "openai/gpt-5.6-sol", "p": 0.54, "rationale": "Using the reference class of private hypergrowth software platforms attempting to pull forward management revenue plans, I estimate a 0.62 probability of economically reaching a $100B run rate by end-2028. Sustaining roughly 50% growth for over two years is feasible but exposed to compute constraints, pricing compression, and enterprise-adoption delays. I apply about 0.85 for a qualifying first-party disclosure after June 2027; leaks and unattributed reporting do not count. The SEC-quarter route adds little because it requires both a listing and a $25B quarter. Accounting for overlap and the strict ambiguity rule yields 0.54, below the 0.79 near-match consensus." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.47, "rationale": "Base rate: 0.30 for already-large, late-stage technology firms sustaining roughly 1.5-times annual revenue growth for about two years and publicly documenting the threshold by a fixed date. OpenAI’s exceptional commercial momentum raises my economic-crossing estimate to roughly 0.60, but both supplied indicators are no-data, leaving the trajectory unverified. I apply about 0.78 for a qualifying first-party disclosure by the deadline: private-company reporting is selective, and leaks, anonymous sourcing, or later confirmation fail. Combined, that yields about 0.47. The 0.79 near-match consensus likely underweights the strict disclosure condition; the forecaster’s 0.60 also looks high without milestone evidence." } ], "aggregation": "log-odds-mean", "extremization": 1, "dissent": null, "spread": 0.08, "warnings": [ "all 6 votes came from one vendor (openai); extremization disabled, persona diversity only" ] }, "final": { "p": 0.5083443815917237, "ci80": [ 0.4214998524266013, 0.5946882092713627 ], "method": "council-only:log-odds-mean", "weights": { "council": 1 } }, "stated": 0.6, "history": [ { "asOf": "2026-09-07", "p": 0.4631917097155392, "source": "radar-engine", "runId": "radar-2026-09-07T23-12-09Z-06a58f" }, { "asOf": "2026-09-07", "p": 0.5083443815917237, "source": "radar-engine", "runId": "radar-2026-09-07T23-49-04Z-de8abf" } ], "indicatorStatus": "no-data", "milestones": [ { "id": "m1", "enginePrior": 0.5, "councilP": null }, { "id": "m2", "enginePrior": 0.5, "councilP": null } ], "arena": "consumer-commercial", "status": "draft", "tier": "T2" }, "indicators": [ { "thesisId": "RT-24", "indicatorId": "li-1", "connector": "manual", "seriesId": "openai-stated-annualized-revenue", "unit": "usd-billions", "asOf": "2026-09-07", "current": null, "status": "no-data", "history": [], "url": "https://openai.com/news/", "note": "manual: 'openai-stated-annualized-revenue' not in manual-series.json" }, { "thesisId": "RT-24", "indicatorId": "li-2", "connector": "manual", "seriesId": "openai-stated-paid-subscribers", "unit": "millions", "asOf": "2026-09-07", "current": null, "status": "no-data", "history": [], "url": "https://openai.com/news/", "note": "manual: 'openai-stated-paid-subscribers' not in manual-series.json" } ], "indicatorStatus": "no-data", "chainEvents": [ { "seq": 238, "ts": "2026-09-07", "kind": "repriced", "hash": "cf137ab40b354a36d93dfdea04c154266891f195d44b09523c1f6d6a929375bf" }, { "seq": 239, "ts": "2026-09-07", "kind": "repriced", "hash": "27dabcf04a00ed8a53e0d1feb928e66266671ace12df15d96429d08618bcb129" }, { "seq": 278, "ts": "2026-09-08", "kind": "registered", "hash": "fa727b0226d6babd36defe315320fa9e41e2b9942ddaa79f2a0973786d9625ee" } ], "url": "/predictions/consumer-commercial/RT-24" }, { "id": "RT-25", "title": "Hyperscaler capex growth slows below 30% in 2027", "arena": "capital-credit", "tier": "T1", "tag": "WARNING", "direction": "down", "status": "registered", "statement": "Combined purchases of property and equipment by Alphabet, Amazon, and Meta for the first nine months of 2027 grow less than 30% over the first nine months of 2026, as reported in their Q3 2027 Forms 10-Q.", "resolutionRule": "Sum the XBRL \"purchases of property and equipment\" fact (USD, duration 2027-01-01 to 2027-09-30) from the Q3 2027 Form 10-Q of Alphabet (CIK 0001652044, concept PaymentsToAcquirePropertyPlantAndEquipment), Amazon (CIK 0001018724, concept PaymentsToAcquireProductiveAssets, which is how Amazon tags the line since 2017), and Meta (CIK 0001326801, concept PaymentsToAcquirePropertyPlantAndEquipment), and divide by the sum of the same facts for 2026-01-01 to 2026-09-30 as reported in the same filings (comparative column). Hit if the ratio is below 1.30. Amazon's figure includes finance-lease principal repayments only if Amazon reports them inside this concept; use the concept as tagged, not adjusted \"cash capex\". If any of the three has not filed its Q3 2027 10-Q by 2027-12-31, or stops tagging the named concept, resolve against the forecaster (miss).", "resolutionSource": { "name": "SEC XBRL company-concept API, purchases of property and equipment (Alphabet and Meta: PaymentsToAcquirePropertyPlantAndEquipment; Amazon: PaymentsToAcquireProductiveAssets)", "url": "https://data.sec.gov/api/xbrl/companyconcept/CIK0001652044/us-gaap/PaymentsToAcquirePropertyPlantAndEquipment.json" }, "ambiguity": "resolves-against-forecaster", "horizon": "2027-12-31", "p": 0.4, "ci80": [ 0.27, 0.55 ], "consensus": { "value": "Goldman Sachs (August 2026) says Street consensus implies hyperscaler capex of about $920 billion in 2027, up 22% from 2026, and argues the true figure is closer to $1.1 trillion (+45%). S&P Global (September 2026) models six hyperscalers at $1.3 trillion in 2027, roughly +50%, with Alphabet at $357 billion and Amazon at $319 billion.", "impliedP": 0.3, "impliedBy": "The Street's $920 billion (+22%) sits below the 30% bar, but every house that has published since (Goldman +45%, S&P +50%, Evercore and Bank of America above $1 trillion) sits above it, and 2027 estimates have only been revised up. Weighting the published range gives growth under 30% about a 30% chance.", "source": "Goldman Sachs on 2027 hyperscaler capex consensus, via Investing.com on Yahoo Finance", "url": "https://finance.yahoo.com/sectors/technology/articles/goldman-says-consensus-2027-hyperscaler-140152065.html", "asOf": "2026-08-20", "note": "No prediction market prices 2027 capex growth (Manifold's 17% on a quarterly decline before 2027 is a different question). This one is close to a coin flip on my side too: the exit run rate of about $165 billion a quarter already puts 2027 near +20% with no further growth, so under 30% means roughly flat spending from here. I lean slightly toward flat, the published estimates lean toward growth." }, "leadingIndicators": [ { "id": "li-1", "label": "Alphabet purchases of property and equipment (quarterly)", "connector": "sec-xbrl", "seriesId": "0001652044:PaymentsToAcquirePropertyPlantAndEquipment:USD", "unit": "usd", "onTrack": { "op": "<", "value": 56000000000 }, "offTrack": { "op": ">=", "value": 64000000000 }, "url": "https://data.sec.gov/api/xbrl/companyconcept/CIK0001652044/us-gaap/PaymentsToAcquirePropertyPlantAndEquipment.json", "note": "Quarterly (three-month) value derived from the year-to-date cash flow facts in Alphabet's 10-Q/10-K. Q2 2026 was $44.9 billion. Thresholds apply to the Q1 2027 print; below $56 billion keeps the sub-30% path alive." }, { "id": "li-2", "label": "Amazon purchases of property and equipment (quarterly)", "connector": "sec-xbrl", "seriesId": "0001018724:PaymentsToAcquireProductiveAssets:USD", "unit": "usd", "onTrack": { "op": "<", "value": 65000000000 }, "offTrack": { "op": ">=", "value": 75000000000 }, "url": "https://data.sec.gov/api/xbrl/companyconcept/CIK0001018724/us-gaap/PaymentsToAcquireProductiveAssets.json", "note": "Quarterly value derived from Amazon's year-to-date cash flow facts. Q2 2026 was $53.1 billion. Thresholds apply to the Q1 2027 print." }, { "id": "li-3", "label": "Meta purchases of property and equipment (quarterly)", "connector": "sec-xbrl", "seriesId": "0001326801:PaymentsToAcquirePropertyPlantAndEquipment:USD", "unit": "usd", "onTrack": { "op": "<", "value": 40000000000 }, "offTrack": { "op": ">=", "value": 46000000000 }, "url": "https://data.sec.gov/api/xbrl/companyconcept/CIK0001326801/us-gaap/PaymentsToAcquirePropertyPlantAndEquipment.json", "note": "Quarterly value derived from Meta's year-to-date cash flow facts. Q2 2026 was $31.1 billion. Thresholds apply to the Q1 2027 print." } ], "milestones": [], "falsifier": "The three companies' combined Q3 2027 year-to-date purchases of property and equipment are at least 1.30 times the 2026 comparative.", "whyItMatters": "Three companies now account for most of the incremental AI capex in the world, and every forecast on the Street assumes they keep growing it by a third or more in 2027. Alphabet's first negative free cash flow quarter and Meta's \"highly dynamic\" language are the first signs boards are looking at the bill. A deceleration below 30% would not be a bust, but it would reprice every supplier and lender who modeled 45%.", "whatWouldRaise": [ "Any of the three guiding 2027 capex to less than 20% growth on the Q4 2026 call.", "Two of the three reporting negative trailing-twelve-month free cash flow at the same time.", "Component (memory, power equipment) price deflation that lets the same capacity ship for fewer dollars." ], "whatWouldCut": [ "Combined Q1 2027 purchases above $185 billion, which would put the nine-month figure on a 40%-plus path.", "A new multi-year compute commitment from a frontier lab that a hyperscaler must build to serve.", "Alphabet or Amazon raising 2027 guidance mid-year, as all three did in 2026." ], "namedEntities": [ "Alphabet", "Amazon", "Meta" ], "marketCall": false, "recusalCheck": { "passed": true, "keysChecked": [ "employer-leasing", "named-competitor-bookings", "customer-pipeline", "ticker" ] }, "signedBy": "Brian Letort", "registeredAt": "2026-09-08", "amendments": [], "ruleHash": "0ce1d3d2a2e10ca4ff8fcc35996e4e1cf7aeed066b2a4aa47c76ef1dde2cf812", "distance": 0.4418327522790393, "distanceLabel": "leans above consensus", "engine": { "id": "RT-25", "prior": { "p": 0.7633333333333333, "method": "sec-xbrl-loglinear-bootstrap:li-1", "note": "log-linear growth +0.189/yr on the last 16 points, extrapolated from 2026-03-31; P(li-1 satisfies onTrack < 56000000000.0 at 2027-12-31); proxy for the thesis, not its rule" }, "council": { "p": 0.4799111474261619, "votes": [ { "persona": "grid-and-power-planner", "model": "openai/gpt-5.6-sol", "p": 0.44, "rationale": "My reference class is megacap infrastructure booms following a denominator year with exceptionally rapid growth: deceleration is common, but another year above 1.30 is not rare while compute and construction pipelines are ramping. The supplied frames show broad 2026 acceleration, so flat spending from the indicated exit rate is not yet the modal path. Conversely, the enlarged 2026 denominator, equipment and power constraints, and quoted consensus implying less than 1.30 make a hit plausible. I heavily discount the 0.763 structural prior because it extrapolates only Alphabet’s noisy Q1 series, not the specified combined nine-month facts. I also apply a small filing/tag-continuity penalty." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.46, "rationale": "My reference class is second-year growth following unusually large megacap investment surges; growth usually decelerates, but sustained AI buildouts can produce consecutive years above 30%. The supplied 2026 readings are 47%-107% above their 2025 counterparts, creating a demanding denominator. Conversely, long-lead power, accelerator and construction commitments support the cited 45%-50% growth scenarios. I discount the 0.763 structural prior because it extrapolates only Alphabet and does not model the combined, lumpy cash-purchase facts. A small deduction reflects the literal filing-and-tagging requirement." }, { "persona": "enterprise-cio", "model": "openai/gpt-5.6-sol", "p": 0.46, "rationale": "Reference class: one-year growth following a greater-than-50% infrastructure-capex ramp during a platform buildout. Such growth usually decelerates as the denominator rises, favoring a sub-30% result. However, the supplied 2026 facts still show roughly 47%-107% year-over-year increases, and long procurement and construction lead times make abrupt flattening less likely. Amazon and Alphabet dominate the combined sum, so continued expansion at either can push the ratio above 1.30. The cited forecasts straddle the threshold but use broader capex definitions and full-year periods, limiting their weight. I also allow a small miss probability for the filing-and-tagging condition." }, { "persona": "credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.53, "rationale": "The reference class is mega-cap investment surges: growth above 30% becomes difficult to sustain after an exceptional base year, but AI infrastructure remains an unusually strong cycle. Momentum still favors a miss—Alphabet Q1 purchases rose about 107% year over year, Meta about 47%, and Amazon Q2 about 68%—so hitting requires marked deceleration by early 2027. Conversely, the elevated 2026 denominator means flat-to-moderate sequential spending would qualify, and the cited forecasts straddle the 30% threshold. I discount the 0.7633 structural prior because it uses Alphabet alone and assign a small filing/tagging failure penalty. Net probability is modestly above the forecaster’s 0.4." }, { "persona": "geopolitics-policy-analyst", "model": "openai/gpt-5.6-sol", "p": 0.43, "rationale": "Reference class: mega-cap investment booms after two to three years of rapid expansion; growth commonly decelerates, but spending rarely flattens while capacity commitments are still ramping. The supplied 2026 facts create a high comparison base: Alphabet and Meta Q1 purchases reached $35.7bn and $19.0bn, while Amazon reached $44.2bn in Q1 and $54.2bn in Q2. That makes the 1.30 hurdle plausible without a downturn. Conversely, the quoted 2027 estimates span about 1.22 to 1.50 growth, with higher cases implying a miss. I discount the 0.763 structural prior because it uses Alphabet alone. Filing and exact-tag continuity add a small downside." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.56, "rationale": "I use a 0.65 base rate for mature megacap capex aggregates growing under 0.30 after an exceptional above-0.50 expansion. The supplied 2026 frames show unusually high bases—Alphabet 35.7bn and Meta 19.0bn in Q1, Amazon 54.2bn in Q2—favoring deceleration through base effects. However, quoted forecasts split sharply: 0.22 growth supports a hit, while 0.45–0.50 scenarios imply a miss, and multiyear infrastructure commitments can keep purchases elevated. I discount the 0.763 structural prior because it uses only Alphabet, not the required three-company sum. A small conjunctive penalty covers three timely filings and exact concept continuity. No calibration table was supplied; overall this is modestly above the forecaster’s 0.4." } ], "aggregation": "log-odds-mean", "extremization": 1, "dissent": null, "spread": 0.13, "warnings": [ "all 6 votes came from one vendor (openai); extremization disabled, persona diversity only" ] }, "final": { "p": 0.633049514322021, "ci80": [ 0.5335005519183268, 0.7383187570805271 ], "method": "stacked-fixed-weights (radar: prior 0.5, council 0.5; no resolved radar items to learn from)", "weights": { "council": 0.5, "prior": 0.5 } }, "stated": 0.4, "history": [ { "asOf": "2026-09-07", "p": 0.4634109101572922, "source": "radar-engine", "runId": "radar-2026-09-07T23-12-09Z-06a58f" }, { "asOf": "2026-09-07", "p": 0.633049514322021, "source": "radar-engine", "runId": "radar-2026-09-07T23-49-04Z-de8abf" } ], "indicatorStatus": "on-track", "milestones": [], "arena": "capital-credit", "status": "draft", "tier": "T1" }, "indicators": [ { "thesisId": "RT-25", "indicatorId": "li-1", "connector": "sec-xbrl", "seriesId": "0001652044:PaymentsToAcquirePropertyPlantAndEquipment:USD", "unit": "usd", "asOf": "2026-03-31", "current": 35674000000, "status": "on-track", "history": [ { "asOf": "2015-03-31", "value": 2927000000 }, { "asOf": "2015-06-30", "value": 2515000000 }, { "asOf": "2015-09-30", "value": 2383000000 }, { "asOf": "2016-03-31", "value": 2444000000 }, { "asOf": "2016-06-30", "value": 2136000000 }, { "asOf": "2016-09-30", "value": 2554000000 }, { "asOf": "2017-03-31", "value": 2508000000 }, { "asOf": "2017-06-30", "value": 2831000000 }, { "asOf": "2017-09-30", "value": 3538000000 }, { "asOf": "2018-03-31", "value": 7299000000 }, { "asOf": "2018-06-30", "value": 5477000000 }, { "asOf": "2018-09-30", "value": 5282000000 }, { "asOf": "2019-03-31", "value": 4638000000 }, { "asOf": "2019-06-30", "value": 6126000000 }, { "asOf": "2019-09-30", "value": 6732000000 }, { "asOf": "2020-03-31", "value": 6005000000 }, { "asOf": "2020-06-30", "value": 5391000000 }, { "asOf": "2020-09-30", "value": 5406000000 }, { "asOf": "2021-03-31", "value": 5942000000 }, { "asOf": "2022-03-31", "value": 9786000000 }, { "asOf": "2023-03-31", "value": 6289000000 }, { "asOf": "2024-03-31", "value": 12012000000 }, { "asOf": "2025-03-31", "value": 17197000000 }, { "asOf": "2026-03-31", "value": 35674000000 } ], "url": "https://data.sec.gov/api/xbrl/companyconcept/CIK0001652044/us-gaap/PaymentsToAcquirePropertyPlantAndEquipment.json", "note": "quarterly frames" }, { "thesisId": "RT-25", "indicatorId": "li-2", "connector": "sec-xbrl", "seriesId": "0001018724:PaymentsToAcquireProductiveAssets:USD", "unit": "usd", "asOf": "2026-06-30", "current": 54208000000, "status": "on-track", "history": [ { "asOf": "2017-09-30", "value": 3074000000 }, { "asOf": "2018-03-31", "value": 3098000000 }, { "asOf": "2018-06-30", "value": 3243000000 }, { "asOf": "2018-09-30", "value": 3352000000 }, { "asOf": "2019-03-31", "value": 3290000000 }, { "asOf": "2019-06-30", "value": 3562000000 }, { "asOf": "2019-09-30", "value": 4697000000 }, { "asOf": "2020-03-31", "value": 6795000000 }, { "asOf": "2020-06-30", "value": 7459000000 }, { "asOf": "2020-09-30", "value": 11063000000 }, { "asOf": "2021-03-31", "value": 12082000000 }, { "asOf": "2021-06-30", "value": 14288000000 }, { "asOf": "2021-09-30", "value": 15748000000 }, { "asOf": "2022-03-31", "value": 14951000000 }, { "asOf": "2022-06-30", "value": 15724000000 }, { "asOf": "2022-09-30", "value": 16378000000 }, { "asOf": "2023-03-31", "value": 14207000000 }, { "asOf": "2023-06-30", "value": 11455000000 }, { "asOf": "2023-09-30", "value": 12479000000 }, { "asOf": "2024-03-31", "value": 14925000000 }, { "asOf": "2024-06-30", "value": 17620000000 }, { "asOf": "2024-09-30", "value": 22620000000 }, { "asOf": "2025-03-31", "value": 25019000000 }, { "asOf": "2025-06-30", "value": 32183000000 }, { "asOf": "2025-09-30", "value": 35095000000 }, { "asOf": "2026-03-31", "value": 44203000000 }, { "asOf": "2026-06-30", "value": 54208000000 } ], "url": "https://data.sec.gov/api/xbrl/companyconcept/CIK0001018724/us-gaap/PaymentsToAcquireProductiveAssets.json", "note": "quarterly frames" }, { "thesisId": "RT-25", "indicatorId": "li-3", "connector": "sec-xbrl", "seriesId": "0001326801:PaymentsToAcquirePropertyPlantAndEquipment:USD", "unit": "usd", "asOf": "2026-03-31", "current": 18997000000, "status": "on-track", "history": [ { "asOf": "2012-03-31", "value": 453000000 }, { "asOf": "2013-03-31", "value": 327000000 }, { "asOf": "2014-03-31", "value": 363000000 }, { "asOf": "2015-03-31", "value": 502000000 }, { "asOf": "2016-03-31", "value": 1132000000 }, { "asOf": "2017-03-31", "value": 1271000000 }, { "asOf": "2018-03-31", "value": 2812000000 }, { "asOf": "2019-03-31", "value": 3837000000 }, { "asOf": "2020-03-31", "value": 3558000000 }, { "asOf": "2021-03-31", "value": 4303000000 }, { "asOf": "2022-03-31", "value": 5441000000 }, { "asOf": "2023-03-31", "value": 6823000000 }, { "asOf": "2024-03-31", "value": 6400000000 }, { "asOf": "2025-03-31", "value": 12941000000 }, { "asOf": "2026-03-31", "value": 18997000000 } ], "url": "https://data.sec.gov/api/xbrl/companyconcept/CIK0001326801/us-gaap/PaymentsToAcquirePropertyPlantAndEquipment.json", "note": "quarterly frames" } ], "indicatorStatus": "on-track", "chainEvents": [ { "seq": 240, "ts": "2026-09-07", "kind": "repriced", "hash": "c2ee2ad74a975c4f4b01fd0e067d3ef10f2c5e88d324693225a6877bf504771a" }, { "seq": 241, "ts": "2026-09-07", "kind": "repriced", "hash": "cef83f0229222dc2e9f73be7d94a25b9beee32948cc3f7f54b87e23cf48854ad" }, { "seq": 279, "ts": "2026-09-08", "kind": "registered", "hash": "b91e1af80f711970dd84589312fba73681f998c7dea50d736887ff856f8b330b" } ], "url": "/predictions/capital-credit/RT-25" }, { "id": "RT-26", "title": "A listed neocloud hits a credit event by 2029", "arena": "capital-credit", "tier": "T2", "tag": "WARNING", "direction": "down", "status": "registered", "statement": "By 2029-12-31 at least one US-listed GPU cloud provider (CoreWeave, Nebius, or another SEC registrant whose primary business is renting AI accelerators) discloses a payment default, covenant breach, distressed debt exchange, or bankruptcy filing.", "resolutionRule": "Hit if an SEC filing (8-K, 10-Q, 10-K, or 6-K) on EDGAR dated on or before 2029-12-31 by a registrant whose most recent 10-K or 20-F describes its principal business as providing GPU or AI accelerator compute as a service contains any of the following: (a) a missed scheduled interest or principal payment on debt of $100 million or more, (b) a statement that the registrant is in breach of a financial covenant (a waiver obtained before the breach date does not count), (c) an exchange offer that a nationally recognized rating agency designates a distressed exchange or selective default in a public release, or (d) a petition under Chapter 11 or Chapter 7. Covenant amendments negotiated without a disclosed breach, rating downgrades alone, and equity dilution do not count. Score by EDGAR full-text search.", "resolutionSource": { "name": "SEC EDGAR full-text search", "url": "https://www.sec.gov/edgar/search/" }, "ambiguity": "resolves-against-forecaster", "horizon": "2029-12-31", "p": 0.7, "ci80": [ 0.55, 0.82 ], "consensus": { "value": "In late July 2026 CoreWeave's five-year credit default swap traded near 855 basis points, which standard pricing implied as roughly a 50% probability of default within five years; spreads narrowed in August after an oversubscribed loan cleared at a 9% all-in cost with a cash lockbox and maintenance covenant. Sell-side equity targets ($139 to $165) imply no credit event.", "impliedP": 0.45, "impliedBy": "A 40 to 50% five-year default probability for one issuer scales to about 30 to 35% over this thesis's three and a quarter years; adding covenant breaches and distressed exchanges (which precede most defaults) and a second listed issuer lifts the market-implied figure to roughly 45%.", "source": "MarketWise, citing Bloomberg CDS data on CoreWeave", "url": "https://marketwise.com/investing/coreweave-stock-the-2-6-billion-debt-signal-the-ai-bubble-is-ignoring/", "asOf": "2026-08-25", "note": "Polymarket's company-bankruptcy markets do not list a neocloud. The CDS implied probability is the closest public market price and covers only one issuer over five years, so the thesis (any listed neocloud, three and a half years) should sit near or slightly above it." }, "leadingIndicators": [ { "id": "li-1", "label": "CoreWeave total debt (long-term debt plus current portion)", "connector": "sec-xbrl", "seriesId": "0001769628:LongTermDebt:USD", "unit": "usd", "onTrack": { "op": ">=", "value": 45000000000 }, "offTrack": { "op": "<", "value": 30000000000 }, "url": "https://data.sec.gov/api/xbrl/companyconcept/CIK0001769628/us-gaap/LongTermDebt.json", "note": "Total debt reported by CoreWeave in its 10-Q balance sheet ($35 billion at 2026-06-30). If CoreWeave tags a different concept (for example DebtInstrumentCarryingAmount), use the concept that reconciles to the \"total debt\" figure in the MD&A. Thresholds apply to the 2027-06-30 balance sheet." }, { "id": "li-2", "label": "ICE BofA US High Yield Index option-adjusted spread", "connector": "fred", "seriesId": "BAMLH0A0HYM2", "unit": "percent", "onTrack": { "op": ">=", "value": 4.5 }, "offTrack": { "op": "<", "value": 3 }, "url": "https://fred.stlouisfed.org/series/BAMLH0A0HYM2", "note": "Daily high-yield spread over Treasuries. Neocloud refinancing between 2026 and 2028 depends on open credit markets; a wide spread makes a distressed outcome more likely." }, { "id": "li-3", "label": "CoreWeave quarterly interest expense", "connector": "sec-xbrl", "seriesId": "0001769628:InterestExpenseDebt:USD", "unit": "usd", "onTrack": { "op": ">=", "value": 900000000 }, "offTrack": { "op": "<", "value": 600000000 }, "url": "https://data.sec.gov/api/xbrl/companyconcept/CIK0001769628/us-gaap/InterestExpenseDebt.json", "note": "Three-month interest expense from the 10-Q ($640 million net in Q2 2026). Thresholds apply to the Q2 2027 filing. Interest growing faster than revenue is the path to a covenant breach." } ], "milestones": [ { "id": "m1", "date": "2027-12-31", "statement": "A listed neocloud discloses a covenant waiver, amendment, or forbearance agreement on debt of $500 million or more.", "p": 0.6 }, { "id": "m2", "date": "2028-12-31", "statement": "A nationally recognized rating agency rates a listed neocloud's senior debt at CCC+ or lower.", "p": 0.5 } ], "falsifier": "No listed neocloud discloses a qualifying event by 2029-12-31 and all 2026-2028 maturities are refinanced on schedule.", "whyItMatters": "Neoclouds are the leveraged middle of the AI capital stack: they borrow against GPUs with five-year useful lives to serve contracts that often run shorter, and they lean on a handful of tenants. The first disclosed credit event tests whether GPU collateral holds value and whether the vendor and hyperscaler backstops are real. Lenders to every private data center deal will reprice off that answer.", "whatWouldRaise": [ "A neocloud's largest customer disclosed at more than 50% of backlog while that customer's own funding slips.", "Secondary H100 rental prices falling below the implied debt-service cost per GPU-hour.", "A new neocloud loan priced above 10% all-in, or a failed syndication reported in an 8-K." ], "whatWouldCut": [ "Backlog conversion to revenue above 40% a year with interest coverage improving for two quarters.", "A hyperscaler or chip vendor acquiring or guaranteeing a neocloud's debt outright.", "Refinancing of 2027 maturities completed by mid-2027 at spreads below 400 basis points." ], "namedEntities": [ "CoreWeave", "Nebius" ], "marketCall": false, "recusalCheck": { "passed": true, "keysChecked": [ "employer-leasing", "named-competitor-bookings", "customer-pipeline", "ticker" ] }, "signedBy": "Brian Letort", "registeredAt": "2026-09-08", "amendments": [], "ruleHash": "5f5c40039e0d9bda35eeda5f0351e95b07c46f2bdd23f3db8982ef592f80bbf8", "distance": 1.0479685558493548, "distanceLabel": "clearly above consensus", "engine": { "id": "RT-26", "prior": { "p": 0.98, "method": "drift-gbm-terminal:li-1", "note": "P(indicator li-1 satisfies onTrack >= 45000000000.0 at 2029-12-31); drift +0.918/yr, vol 0.243/sqrt(yr) from 6 points; proxy for the thesis, not its rule" }, "council": { "p": 0.5083807451095246, "votes": [ { "persona": "grid-and-power-planner", "model": "openai/gpt-5.6-sol", "p": 0.48, "rationale": "My reference class is deeply leveraged single-B/CCC growth borrowers: roughly 0.10–0.15 annual default hazard, implying 0.30–0.41 over 3.3 years before covenant events. The supplied debt series rose from $7.9 billion to $24.9 billion in five quarters and quarterly interest expense to $558 million, making a maintenance-covenant breach plausible. However, the cited five-year CDS probability scales to about 0.37 over this horizon under constant hazard; successful refinancing and a 2.65-point broad high-yield spread indicate funding remains available. Additional registrants and covenant breaches raise basket risk, but pre-breach waivers, downgrades, and dilution do not count. I discount the GBM prior because debt growth alone is not a credit-event model." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.52, "rationale": "Reference class: one issuer with a five-year market-implied default probability of 0.50. A flat hazard converts that to about 0.37 over the remaining 3.3 years. I raise this for multiple eligible issuers and because covenant breach is broader than payment default: the supplied filer’s long-term debt rose from $7.9 billion to $24.9 billion in five quarters, while quarterly interest expense reached $558 million. I reject the 0.98 drift prior because mechanically extrapolating debt is not a credit-event model. Offsets include successful refinancing, a cash lockbox, tight broad high-yield spreads, and strict EDGAR wording and principal-business tests. Correlated AI-demand and refinancing risks limit the multi-issuer uplift. The stated 0.70 is too aggressive." }, { "persona": "enterprise-cio", "model": "openai/gpt-5.6-sol", "p": 0.53, "rationale": "I use stressed speculative-grade borrowers as the reference class, anchored by the supplied CDS indication. A five-year default probability near 0.50 implies roughly 0.13 annual hazard; over 3.3 years that compounds to about 0.37 for CoreWeave. The resolution also counts disclosed covenant breaches, while additional qualifying issuers create portfolio optionality, lifting the estimate. CoreWeave’s debt rose from $7.9 billion to $24.9 billion in five quarters and quarterly interest expense reached $558 million, materially increasing refinancing and execution risk. Offsetting this, broad high-yield spreads remain benign, recent financing cleared, and waivers obtained before breach do not count. The strict EDGAR wording and principal-business test also exclude many near-distress outcomes. I discount the mechanical 0.98 prior because debt growth alone is not..." }, { "persona": "credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.56, "rationale": "Reference class: a highly leveraged neocloud priced near a 0.50 five-year default probability. Constant-hazard compounding converts that to roughly 0.37 through the 3.3-year horizon. I raise this for the broader covenant-breach trigger and additional listed issuers, but not independently because their refinancing risks are highly correlated. Debt increased from $7.9 billion to $24.9 billion in five quarters, while quarterly interest expense reached $558 million, materially increasing refinancing and maintenance-covenant risk. Offsetting this, the recent loan cleared, broad high-yield spreads remain only 2.65%, and pre-breach waivers or amendments do not qualify. The 0.98 structural prior extrapolates debt growth rather than the filing-defined event and is not decision-useful." }, { "persona": "geopolitics-policy-analyst", "model": "openai/gpt-5.6-sol", "p": 0.49, "rationale": "Reference class: single-B distressed issuers with rapid debt growth, adjusted from the supplied five-year CDS-implied default risk. A constant-hazard conversion of 0.50 over five years gives roughly 0.37 through the 3.3-year horizon for CoreWeave. The rule’s covenant-breach trigger is broader than conventional default, and debt plus quarterly interest expense are rising rapidly, lifting risk. Adding other eligible registrants raises portfolio probability only modestly because their debt risks are correlated and the qualifying universe is narrow. The oversubscribed financing, lockbox, maintenance covenant, and benign broad high-yield spread reduce near-term refinancing hazard. The mechanical 0.98 debt-drift prior ignores liquidity, revenue growth, refinancing, and strict EDGAR criteria." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.47, "rationale": "Base rate: a 5–8% annual default hazard for deeply speculative-grade US corporate issuers, implying roughly 16–24% over the remaining 3.3 years for one issuer. CoreWeave-specific evidence warrants a large uplift: $24.9 billion of long-term debt, rapidly rising interest expense, and the quoted 855 bp CDS. But converting its cited five-year 50% default probability to 3.3 years gives about 37%, not 50%. Successful 9% refinancing and tight aggregate high-yield spreads reduce near-term risk. Other eligible registrants add portfolio hazard, while covenant breach broadens the trigger. Conversely, eligibility, hard-date EDGAR disclosure, and adverse ambiguity narrow it. I reject the 0.98 drift prior because debt growth alone does not model credit events." } ], "aggregation": "log-odds-mean", "extremization": 1, "dissent": null, "spread": 0.09, "warnings": [ "all 6 votes came from one vendor (openai); extremization disabled, persona diversity only" ] }, "final": { "p": 0.8768219639094311, "ci80": [ 0.8566417681956147, 0.8945112539319943 ], "method": "stacked-fixed-weights (radar: prior 0.5, council 0.5; no resolved radar items to learn from)", "weights": { "council": 0.5, "prior": 0.5 } }, "stated": 0.7, "history": [ { "asOf": "2026-09-07", "p": 0.8746278147865417, "source": "radar-engine", "runId": "radar-2026-09-07T23-12-09Z-06a58f" }, { "asOf": "2026-09-07", "p": 0.8768219639094311, "source": "radar-engine", "runId": "radar-2026-09-07T23-49-04Z-de8abf" } ], "indicatorStatus": "off-track", "milestones": [ { "id": "m1", "enginePrior": 0.5, "councilP": null }, { "id": "m2", "enginePrior": null, "councilP": null } ], "arena": "capital-credit", "status": "draft", "tier": "T2" }, "indicators": [ { "thesisId": "RT-26", "indicatorId": "li-1", "connector": "sec-xbrl", "seriesId": "0001769628:LongTermDebt:USD", "unit": "usd", "asOf": "2026-03-31", "current": 24859000000, "status": "off-track", "history": [ { "asOf": "2024-12-31", "value": 7926000000 }, { "asOf": "2025-03-31", "value": 8711666000 }, { "asOf": "2025-06-30", "value": 11051501000 }, { "asOf": "2025-09-30", "value": 14034934000 }, { "asOf": "2025-12-31", "value": 21373000000 }, { "asOf": "2026-03-31", "value": 24859000000 } ], "url": "https://data.sec.gov/api/xbrl/companyconcept/CIK0001769628/us-gaap/LongTermDebt.json", "note": "quarterly frames" }, { "thesisId": "RT-26", "indicatorId": "li-2", "connector": "fred", "seriesId": "BAMLH0A0HYM2", "unit": "percent", "asOf": "2026-09-03", "current": 2.65, "status": "off-track", "history": [ { "asOf": "2026-06-12", "value": 2.71 }, { "asOf": "2026-06-15", "value": 2.66 }, { "asOf": "2026-06-16", "value": 2.71 }, { "asOf": "2026-06-17", "value": 2.63 }, { "asOf": "2026-06-18", "value": 2.66 }, { "asOf": "2026-06-19", "value": 2.66 }, { "asOf": "2026-06-22", "value": 2.65 }, { "asOf": "2026-06-23", "value": 2.71 }, { "asOf": "2026-06-24", "value": 2.76 }, { "asOf": "2026-06-25", "value": 2.78 }, { "asOf": "2026-06-26", "value": 2.83 }, { "asOf": "2026-06-29", "value": 2.8 }, { "asOf": "2026-06-30", "value": 2.75 }, { "asOf": "2026-07-01", "value": 2.74 }, { "asOf": "2026-07-02", "value": 2.75 }, { "asOf": "2026-07-03", "value": 2.74 }, { "asOf": "2026-07-06", "value": 2.72 }, { "asOf": "2026-07-07", "value": 2.67 }, { "asOf": "2026-07-08", "value": 2.7 }, { "asOf": "2026-07-09", "value": 2.7 }, { "asOf": "2026-07-10", "value": 2.69 }, { "asOf": "2026-07-13", "value": 2.69 }, { "asOf": "2026-07-14", "value": 2.72 }, { "asOf": "2026-07-15", "value": 2.71 }, { "asOf": "2026-07-16", "value": 2.71 }, { "asOf": "2026-07-17", "value": 2.73 }, { "asOf": "2026-07-20", "value": 2.69 }, { "asOf": "2026-07-21", "value": 2.69 }, { "asOf": "2026-07-22", "value": 2.68 }, { "asOf": "2026-07-23", "value": 2.77 }, { "asOf": "2026-07-24", "value": 2.79 }, { "asOf": "2026-07-27", "value": 2.81 }, { "asOf": "2026-07-28", "value": 2.84 }, { "asOf": "2026-07-29", "value": 2.87 }, { "asOf": "2026-07-30", "value": 2.84 }, { "asOf": "2026-07-31", "value": 2.85 }, { "asOf": "2026-08-03", "value": 2.78 }, { "asOf": "2026-08-04", "value": 2.73 }, { "asOf": "2026-08-05", "value": 2.75 }, { "asOf": "2026-08-06", "value": 2.71 }, { "asOf": "2026-08-07", "value": 2.7 }, { "asOf": "2026-08-10", "value": 2.7 }, { "asOf": "2026-08-11", "value": 2.72 }, { "asOf": "2026-08-12", "value": 2.71 }, { "asOf": "2026-08-13", "value": 2.71 }, { "asOf": "2026-08-14", "value": 2.67 }, { "asOf": "2026-08-17", "value": 2.7 }, { "asOf": "2026-08-18", "value": 2.75 }, { "asOf": "2026-08-19", "value": 2.73 }, { "asOf": "2026-08-20", "value": 2.75 }, { "asOf": "2026-08-21", "value": 2.7 }, { "asOf": "2026-08-24", "value": 2.69 }, { "asOf": "2026-08-25", "value": 2.7 }, { "asOf": "2026-08-26", "value": 2.67 }, { "asOf": "2026-08-27", "value": 2.63 }, { "asOf": "2026-08-28", "value": 2.6 }, { "asOf": "2026-08-31", "value": 2.63 }, { "asOf": "2026-09-01", "value": 2.65 }, { "asOf": "2026-09-02", "value": 2.66 }, { "asOf": "2026-09-03", "value": 2.65 } ], "url": "https://fred.stlouisfed.org/series/BAMLH0A0HYM2", "note": null }, { "thesisId": "RT-26", "indicatorId": "li-3", "connector": "sec-xbrl", "seriesId": "0001769628:InterestExpenseDebt:USD", "unit": "usd", "asOf": "2026-06-30", "current": 558000000, "status": "off-track", "history": [ { "asOf": "2024-03-31", "value": 40503000 }, { "asOf": "2024-06-30", "value": 67097000 }, { "asOf": "2024-09-30", "value": 102328000 }, { "asOf": "2025-03-31", "value": 257000000 }, { "asOf": "2025-06-30", "value": 256000000 }, { "asOf": "2025-09-30", "value": 283416000 }, { "asOf": "2026-03-31", "value": 427000000 }, { "asOf": "2026-06-30", "value": 558000000 } ], "url": "https://data.sec.gov/api/xbrl/companyconcept/CIK0001769628/us-gaap/InterestExpenseDebt.json", "note": "quarterly frames" } ], "indicatorStatus": "off-track", "chainEvents": [ { "seq": 242, "ts": "2026-09-07", "kind": "repriced", "hash": "4db362b35af9009f5beaf4962520989b50b9c1ccd1afc59be3c2f7e903b48407" }, { "seq": 243, "ts": "2026-09-07", "kind": "repriced", "hash": "c1f4d24ea7af904b66899335205d545b689ab601fd1cb0b5198c82da053954fb" }, { "seq": 280, "ts": "2026-09-08", "kind": "registered", "hash": "6fe954ae20560f63d68e2b813c9f247117cadc04874694829060ce44ff85e6e2" } ], "url": "/predictions/capital-credit/RT-26" }, { "id": "RT-27", "title": "A big-three cloud cuts server life to four years", "arena": "capital-credit", "tier": "T2", "tag": "THESIS", "direction": "down", "status": "registered", "statement": "By 2029-12-31 at least one of Alphabet, Microsoft, or Amazon discloses in a Form 10-K or 10-Q that the estimated useful life of servers, or of a separately identified class of AI accelerators, has been reduced to four years or less.", "resolutionRule": "Hit if a 10-K or 10-Q filed on EDGAR on or before 2029-12-31 by Alphabet (CIK 0001652044), Microsoft (CIK 0000789019), or Amazon (CIK 0001018724) states in the property and equipment or significant accounting policies note that the estimated useful life for servers, networking equipment, GPUs, or AI accelerators (or any named subset of them) is four years or fewer, where the prior filing stated a longer life. A stated range whose upper bound is four years or less counts; a range like \"two to six years\" does not. Impairments, write-downs, and accelerated depreciation on specific retired assets do not count. If all three companies stop disclosing useful lives, resolve against the forecaster (miss).", "resolutionSource": { "name": "SEC EDGAR full-text search", "url": "https://www.sec.gov/edgar/search/" }, "ambiguity": "resolves-against-forecaster", "horizon": "2029-12-31", "p": 0.45, "ci80": [ 0.3, 0.62 ], "consensus": { "value": "Alphabet and Microsoft depreciate servers over six years (Microsoft discloses \"two to six years\" and, from fiscal 2027, extended data center buildings from 15 to 25 years). Amazon moved a subset of servers and networking equipment from six to five years effective 2025-01-01. No hyperscaler has signaled a move below five years; the direction of accounting change since 2020 has been longer, not shorter.", "impliedP": 0.15, "impliedBy": "Every useful-life change by the three since 2020 except one has been longer, the one shortening was one year on a subset, and none of the three has signalled a four-year life. The disclosed policies imply a cut to four years or less by 2029 at roughly 15%.", "source": "Amazon FY2025 Form 10-K property and equipment note, via edgar.tools", "url": "https://app.edgar.tools/companies/AMZN/disclosures/property-plant-equipment", "asOf": "2026-02-06", "note": "No prediction market covers depreciation policy. The baseline is the current disclosed policies of the three companies." }, "leadingIndicators": [ { "id": "li-1", "label": "Shortest disclosed server useful life, big three clouds", "connector": "manual", "seriesId": "big3-cloud-min-server-useful-life-years", "unit": "years", "onTrack": { "op": "<=", "value": 5 }, "offTrack": { "op": ">=", "value": 6 }, "url": "https://www.sec.gov/edgar/search/#/q=%22useful%20lives%22%20servers&forms=10-K", "note": "Minimum stated useful life (years) for servers or AI accelerators across the latest 10-K of Alphabet, Microsoft, and Amazon. Read the property and equipment note. Currently 5 (Amazon subset)." }, { "id": "li-2", "label": "Amazon depreciation and amortization (quarterly)", "connector": "sec-xbrl", "seriesId": "0001018724:DepreciationDepletionAndAmortization:USD", "unit": "usd", "onTrack": { "op": ">=", "value": 22000000000 }, "offTrack": { "op": "<", "value": 17000000000 }, "url": "https://data.sec.gov/api/xbrl/companyconcept/CIK0001018724/us-gaap/DepreciationDepletionAndAmortization.json", "note": "Three-month depreciation and amortization from the cash flow statement. A step-up faster than capex growth indicates shorter lives or accelerated depreciation being applied. Thresholds apply to Q2 2027." } ], "milestones": [ { "id": "m1", "date": "2027-12-31", "statement": "One of the three discloses AI accelerators or GPUs as a separately depreciated asset class with its own useful life.", "p": 0.5 }, { "id": "m2", "date": "2028-12-31", "statement": "A second reduction in stated server useful life (any of the three, any subset) appears in a 10-K or 10-Q.", "p": 0.5 } ], "falsifier": "All three companies' fiscal 2029 10-Ks still state a minimum server useful life of five years or more.", "whyItMatters": "Useful life is the accounting hinge of the whole buildout. At six years, $200 billion of servers costs $33 billion a year in depreciation; at four, $50 billion. A cut to four years by any of the big three would confirm that annual accelerator generations make older fleets uneconomic for inference, and would flow straight into cloud pricing, margins, and the collateral math behind GPU-backed debt.", "whatWouldRaise": [ "A hyperscaler taking an impairment or accelerated depreciation charge on a named GPU generation.", "Auditor critical audit matters citing server useful-life estimates in a 10-K.", "Secondary market prices for three-year-old accelerators below 20% of original cost." ], "whatWouldCut": [ "Any of the three lengthening server lives again, as Microsoft did for buildings.", "Disclosed fleet data showing older accelerators still running at high utilization for inference.", "A slowdown in accelerator generation cadence to every two years or more." ], "namedEntities": [ "Alphabet", "Microsoft", "Amazon" ], "marketCall": false, "recusalCheck": { "passed": true, "keysChecked": [ "employer-leasing", "named-competitor-bookings", "customer-pipeline", "ticker" ] }, "signedBy": "Brian Letort", "registeredAt": "2026-09-08", "amendments": [], "ruleHash": "497457f24ef1e3ea3c1fe17e4112bd4b3ceee72953957631e6110ba28edce743", "distance": 1.533930359925955, "distanceLabel": "far above consensus", "engine": { "id": "RT-27", "prior": { "p": 0.98, "method": "sec-xbrl-loglinear-bootstrap:li-2", "note": "log-linear growth +0.164/yr on the last 16 points, extrapolated from 2026-06-30; P(li-2 satisfies onTrack >= 22000000000.0 at 2029-12-31); proxy for the thesis, not its rule" }, "council": { "p": 0.2941159095811337, "votes": [ { "persona": "grid-and-power-planner", "model": "openai/gpt-5.6-sol", "p": 0.31, "rationale": "Reference class: disclosed useful-life revisions by these three hyperscalers over roughly the past decade. Lengthenings have dominated; a qualifying shortening to four years or less is rare. Using an approximately 0.03 annual company-level hazard across about 11 remaining 10-K cycles gives a low-0.3 cumulative probability, adjusted for correlated AI-obsolescence pressure. Amazon’s six-to-five-year change makes another step plausible, but not likely. A newly disclosed accelerator class may fail the requirement that a prior filing stated a longer life. Rising aggregate depreciation chiefly reflects the growing asset base and does not identify useful-life policy, so the 0.98 structural prior is not probative." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.33, "rationale": "Reference class: Big Three useful-life estimate changes over the past decade. Changes occur, but qualifying reversals to four years or less are rarer than extensions. I use an approximately 0.11 annual group-level hazard over the 3.3-year window, compounding to about 0.32. Amazon’s five-year subset is the clearest one-step path, while rapid AI hardware turnover adds some pressure. Offsets are correlated policies, the requirement that a prior filing show a longer life, and exclusion of retirement charges. Rising Amazon D&A is confounded by asset growth and does not support the 0.98 structural prior. The cutoff also excludes 2029 annual reports filed in 2030." }, { "persona": "enterprise-cio", "model": "openai/gpt-5.6-sol", "p": 0.28, "rationale": "I use useful-life estimate changes among megacap technology firms as the reference class: qualifying changes are infrequent, and recent hyperscaler revisions have generally lengthened lives. A rough 0.035 annual qualifying hazard per company gives about 0.29 across three firms over 3.3 years before correlation. Amazon’s move from six to five years supports some downward hazard, but reaching four is a larger accounting step. Rapid accelerator obsolescence does not necessarily shorten physical economic life because equipment can be redeployed. A newly disclosed GPU class may also fail the strict prior-longer-life requirement. Rising depreciation is driven by asset growth and is not evidence of a policy change, so the structural prior is not informative." }, { "persona": "credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.3, "rationale": "I use a low-single-digit annual issuer hazard for a qualifying depreciation-policy reversal, compounded across three issuers and roughly 3.3 years, with Amazon carrying the highest hazard because one subset is already at five years. The reference class is disclosed useful-life changes by mature hyperscalers: these are infrequent, and recent changes have predominantly lengthened lives. Rising Amazon depreciation is weak evidence because rapid capex and asset-base growth can produce the same pattern. The rule additionally requires an EDGAR filing, a prior longer life, and an upper bound no greater than four; ambiguity and ranges such as Microsoft’s two-to-six years count against resolution. The XBRL structural prior does not model those requirements." }, { "persona": "geopolitics-policy-analyst", "model": "openai/gpt-5.6-sol", "p": 0.31, "rationale": "Using disclosed useful-life revisions at large technology firms as the reference class, I estimate annual qualifying hazards near 0.06 for Amazon and 0.03 for each other company; three-year compounding, adjusted for correlated accounting practice, gives about 0.30. Amazon’s prior six-to-five-year subset reduction raises its hazard, and AI obsolescence creates pressure. Against that, the recent accounting direction is longer lives, reuse for inference can support five-plus years, and auditors require observed retirement evidence. A newly disclosed four-year accelerator class may fail because the rule requires a prior longer life. Rising depreciation mostly reflects capex, not shorter lives. The 0.98 XBRL prior is therefore non-diagnostic." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.24, "rationale": "Base rate: I use a 0.025 annual hazard per company for a qualifying useful-life reduction among large cloud filers, reflecting the rarity of such reversals and no indicated move to four years or less since 2020. Across roughly ten company-years, that compounds to about 0.22. Amazon’s prior six-to-five-year reduction and rapid AI hardware turnover modestly raise the estimate. Offsetting this, recent accounting changes have favored longer lives, rising depreciation is mostly an asset-base signal, and resolution requires both an actual policy reversal and timely, explicit EDGAR disclosure. A newly separated accelerator class may also fail the prior-longer-life test under adverse ambiguity." } ], "aggregation": "log-odds-mean", "extremization": 1, "dissent": null, "spread": 0.09, "warnings": [ "all 6 votes came from one vendor (openai); extremization disabled, persona diversity only" ] }, "final": { "p": 0.8187900793723636, "ci80": [ 0.6615090566926156, 0.8433240460966045 ], "method": "stacked-fixed-weights (radar: prior 0.5, council 0.5; no resolved radar items to learn from)", "weights": { "council": 0.5, "prior": 0.5 } }, "stated": 0.45, "history": [ { "asOf": "2026-09-07", "p": 0.8255699972915652, "source": "radar-engine", "runId": "radar-2026-09-07T23-12-09Z-06a58f" }, { "asOf": "2026-09-07", "p": 0.8187900793723636, "source": "radar-engine", "runId": "radar-2026-09-07T23-49-04Z-de8abf" } ], "indicatorStatus": "mixed", "milestones": [ { "id": "m1", "enginePrior": 0.5, "councilP": null }, { "id": "m2", "enginePrior": null, "councilP": null } ], "arena": "capital-credit", "status": "draft", "tier": "T2" }, "indicators": [ { "thesisId": "RT-27", "indicatorId": "li-1", "connector": "manual", "seriesId": "big3-cloud-min-server-useful-life-years", "unit": "years", "asOf": "2026-09-07", "current": null, "status": "no-data", "history": [], "url": "https://www.sec.gov/edgar/search/#/q=%22useful%20lives%22%20servers&forms=10-K", "note": "manual: 'big3-cloud-min-server-useful-life-years' not in manual-series.json" }, { "thesisId": "RT-27", "indicatorId": "li-2", "connector": "sec-xbrl", "seriesId": "0001018724:DepreciationDepletionAndAmortization:USD", "unit": "usd", "asOf": "2026-06-30", "current": 19988000000, "status": "between", "history": [ { "asOf": "2016-06-30", "value": 1909000000 }, { "asOf": "2016-09-30", "value": 2084000000 }, { "asOf": "2017-03-31", "value": 2435000000 }, { "asOf": "2017-06-30", "value": 2633000000 }, { "asOf": "2017-09-30", "value": 2912000000 }, { "asOf": "2018-03-31", "value": 3671000000 }, { "asOf": "2018-06-30", "value": 3630000000 }, { "asOf": "2018-09-30", "value": 3778000000 }, { "asOf": "2019-03-31", "value": 4854000000 }, { "asOf": "2019-06-30", "value": 5202000000 }, { "asOf": "2019-09-30", "value": 5563000000 }, { "asOf": "2020-03-31", "value": 5362000000 }, { "asOf": "2020-06-30", "value": 5748000000 }, { "asOf": "2020-09-30", "value": 6523000000 }, { "asOf": "2021-03-31", "value": 7508000000 }, { "asOf": "2021-06-30", "value": 8038000000 }, { "asOf": "2021-09-30", "value": 8948000000 }, { "asOf": "2022-03-31", "value": 9193000000 }, { "asOf": "2022-06-30", "value": 9716000000 }, { "asOf": "2022-09-30", "value": 10327000000 }, { "asOf": "2023-03-31", "value": 11123000000 }, { "asOf": "2023-06-30", "value": 11589000000 }, { "asOf": "2023-09-30", "value": 12131000000 }, { "asOf": "2024-03-31", "value": 11684000000 }, { "asOf": "2024-06-30", "value": 12038000000 }, { "asOf": "2024-09-30", "value": 13442000000 }, { "asOf": "2025-03-31", "value": 14262000000 }, { "asOf": "2025-06-30", "value": 15227000000 }, { "asOf": "2025-09-30", "value": 16796000000 }, { "asOf": "2026-03-31", "value": 18945000000 }, { "asOf": "2026-06-30", "value": 19988000000 } ], "url": "https://data.sec.gov/api/xbrl/companyconcept/CIK0001018724/us-gaap/DepreciationDepletionAndAmortization.json", "note": "quarterly frames" } ], "indicatorStatus": "mixed", "chainEvents": [ { "seq": 244, "ts": "2026-09-07", "kind": "repriced", "hash": "4afe5c25803226a2ec1e3de3cbd87592839e398cdc3de24c4295fa8ed70f022e" }, { "seq": 245, "ts": "2026-09-07", "kind": "repriced", "hash": "6e46d6a840b3508a35c630ff063b2b2903272eaab778f80872d3f0be6623821b" }, { "seq": 281, "ts": "2026-09-08", "kind": "registered", "hash": "61e1f350a2a1816a96c16668823fb2e1194c0f0c7d6591b7152798d319729e2e" } ], "url": "/predictions/capital-credit/RT-27" }, { "id": "RT-28", "title": "Grid equipment ETF beats S&P 500 by 15 points", "arena": "capital-credit", "tier": "T1", "tag": "CALL", "direction": "up", "status": "registered", "statement": "The First Trust NASDAQ Clean Edge Smart Grid Infrastructure Index Fund (GRID) delivers a cumulative NAV total return for calendar years 2026 and 2027 at least 15 percentage points above the S&P 500 Index total return over the same period.", "resolutionRule": "Compute GRID's cumulative return by compounding its calendar-year 2026 and 2027 NAV total returns as published in the \"Calendar Year Total Returns\" table on the First Trust GRID fund page after 2027-12-31. Compute the S&P 500 Index cumulative return by compounding the calendar-year 2026 and 2027 \"Benchmark S&P 500 Index\" total returns published on the SSGA SPY fund page. Hit if GRID cumulative minus S&P 500 cumulative is 15.0 percentage points or more. Score using figures as published on or before 2028-01-31. If GRID is liquidated or merged before 2027-12-31, resolve against the forecaster (miss).", "resolutionSource": { "name": "First Trust GRID fund page, Calendar Year Total Returns", "url": "https://www.ftportfolios.com/etf/GRID" }, "ambiguity": "resolves-against-forecaster", "horizon": "2027-12-31", "p": 0.5, "ci80": [ 0.35, 0.65 ], "consensus": { "value": "Under an efficient-market baseline the expected excess return of any diversified equity ETF over the S&P 500 is zero; historically a two-year outperformance margin of 15 points or more for a sector fund with GRID's volatility occurs about one year in four. At 2026-07-31 GRID's NAV was up 17.76% year to date against 10.06% for SPY, a 7.7 point lead.", "impliedP": 0.33, "impliedBy": "With zero expected excess return from here and a 7.7 point lead already banked, the market-implied chance of finishing the two years at least 15 points ahead (another 7 points of relative gain over 17 months at GRID's tracking volatility) is about one in three.", "source": "First Trust GRID fund page (performance and holdings)", "url": "https://www.ftportfolios.com/etf/GRID", "asOf": "2026-09-04", "note": "No prediction market or published sell-side target exists for GRID relative to the S&P 500. The consensus is the zero-excess-return prior, which is what a market call must beat. My 50% is an honest coin flip on the 15-point margin: I expect GRID to finish ahead, but the extra 7 points of relative gain needed is about one year of the fund's tracking volatility, so the margin, not the direction, is the uncertain part." }, "leadingIndicators": [ { "id": "li-1", "label": "GRID year-to-date NAV total return minus S&P 500 total return", "connector": "manual", "seriesId": "grid-minus-sp500-ytd-total-return", "unit": "percentage-points", "onTrack": { "op": ">=", "value": 5 }, "offTrack": { "op": "<", "value": -5 }, "url": "https://www.ftportfolios.com/etf/GRID", "note": "GRID NAV year-to-date total return (First Trust page) minus S&P 500 Index year-to-date total return (SSGA SPY page, Benchmark row). Read at each quarter end; thresholds apply from 2026-12-31 onward as cumulative two-year gap." }, { "id": "li-2", "label": "Industrial production, electrical equipment, appliances, and components", "connector": "fred", "seriesId": "IPG335S", "unit": "index-2017-100", "onTrack": { "op": ">=", "value": 108 }, "offTrack": { "op": "<", "value": 100 }, "url": "https://fred.stlouisfed.org/series/IPG335S", "note": "Federal Reserve G.17 industrial production index for NAICS 335 (electrical equipment, appliance, and component manufacturing), seasonally adjusted, 2017=100, monthly. Real output of the switchgear and transformer makers GRID holds." } ], "milestones": [], "falsifier": "GRID's cumulative 2026-2027 NAV total return is less than 15 percentage points above the S&P 500 Index total return as published after year-end 2027.", "whyItMatters": "Power equipment (switchgear, transformers, cabling) is the part of the AI buildout with the longest lead times and the fewest suppliers, and it keeps selling even if GPU spending slows, because the grid work has to happen regardless. Public equity is the only place a stranger can score that view. This is the one market call in the book, and it is deliberately a basket, not a name.", "whatWouldRaise": [ "Utility capex plans for 2027 rising again in the fall 2026 rate-case cycle.", "Transformer and switchgear lead times still above 100 weeks in supplier disclosures.", "GRID's top holdings (Schneider, Eaton, ABB) guiding to double-digit orders growth for 2027." ], "whatWouldCut": [ "A hyperscaler capex deceleration that hits electrical orders before it hits GPUs.", "A broad equity rally led by the largest S&P 500 constituents that GRID does not hold.", "Tariff or rate shocks that hit European industrials, which are half of GRID's top ten." ], "namedEntities": [ "First Trust" ], "marketCall": true, "recusalCheck": { "passed": true, "keysChecked": [ "employer-leasing", "named-competitor-bookings", "customer-pipeline", "ticker" ] }, "signedBy": "Brian Letort", "registeredAt": "2026-09-08", "amendments": [], "ruleHash": "d9976b5a4f716cfa41c2fecfc81aef7582930d88b1c96a59910645a7ff0675bc", "distance": 0.7081850579244856, "distanceLabel": "leans above consensus", "engine": { "id": "RT-28", "prior": { "p": 0.17677115338084115, "method": "drift-gbm-terminal:li-2", "note": "P(indicator li-2 satisfies onTrack >= 108.0 at 2027-12-31); drift +0.00458/yr, vol 0.0492/sqrt(yr) from 655 points; proxy for the thesis, not its rule" }, "council": { "p": 0.35812013603618575, "votes": [ { "persona": "grid-and-power-planner", "model": "openai/gpt-5.6-sol", "p": 0.34, "rationale": "Reference class: diversified sector ETFs facing a two-calendar-year 15-point excess-return hurdle; the supplied historical base rate is about 0.25. By July 31, GRID had banked a 7.7-point YTD lead, materially improving the conditional odds. However, compounding leaves roughly another 6–8 points of relative performance needed through 2027, depending on market returns. That remains a meaningful move over about 17 months of sector tracking volatility, leaving the odds below even. Rising electrical-equipment industrial production supports the sector backdrop, but IPG335S is a weak equity-return predictor and says nothing about valuation. I also allow small downside for liquidation or untimely publication under the strict rule." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.38, "rationale": "My reference class is diversified sector ETFs over two-year windows: the supplied historical base rate is roughly 0.25 for beating the S&P 500 by 15 points. GRID’s 7.7-point lead through July 2026 materially raises the conditional probability, but approximately half the measurement window remains and the threshold still requires substantial relative gains. Electrical-equipment production rising from 98.3 in February to 101.4 in July is supportive, though it is a weak equity-return signal. A zero-excess-return diffusion conditioned on the current lead suggests roughly a one-in-three chance; modest grid-investment momentum lifts my estimate slightly. Liquidation and publication risks are minimal but nonzero." }, { "persona": "enterprise-cio", "model": "openai/gpt-5.6-sol", "p": 0.34, "rationale": "My reference class is two-year sector-ETF performance versus the S&P 500, with the supplied historical frequency near 0.25 for outperformance of at least 15 points. GRID’s reported 7.7-point lead through July 2026 materially improves the conditional odds, but roughly half the required margin remains and 17 months of equity exposure can readily reverse it. Rising electrical-equipment production is mildly supportive, not strongly predictive of shareholder returns because expectations and valuations are already priced. I discount the 0.50 call for the unusually wide required margin, benchmark strength, and minor publication or fund-continuity risk." }, { "persona": "credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.37, "rationale": "I use the supplied reference-class base rate: comparable sector funds exceed the S&P 500 by 15 points over two years about one time in four. GRID’s reported 7.7-point lead through July 2026 materially improves the conditional odds, leaving roughly 7.3 points to gain over 17 months. However, sector-relative returns are volatile and mean reversion is a substantial risk after early outperformance. Rising electrical-equipment production supports the theme but is a weak equity-return timing signal and may already be priced. Fund-liquidation and publication risks are small. The structural prior understates the known YTD lead, while 0.5 overweights thematic momentum." }, { "persona": "geopolitics-policy-analyst", "model": "openai/gpt-5.6-sol", "p": 0.38, "rationale": "Reference class: diversified sector ETFs over two-year windows, with roughly a 0.25 frequency of beating the S&P 500 by at least 15 points. The supplied July NAV figures bank a 7.7-point lead after seven months, materially lifting the conditional probability, but roughly 7 more points must be gained over 17 months of volatile relative returns. Rising electrical-equipment production is supportive, though it is a weak, lagging proxy for equity excess returns and likely partly priced. A zero-excess-return baseline still makes the positive threshold harder than a coin flip. Publication and liquidation risk is small but one-sided. Thus 0.5 looks high." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.34, "rationale": "Base rate: 0.25, using the supplied historical frequency that a sector ETF with GRID-like volatility beats the S&P 500 by at least 15 points over two years. The supplied 7.7-point lead through July 2026 raises the conditional odds, but 17 months remained at that reading and roughly another 7 points was still needed. Rising electrical-equipment production is modestly supportive but is not a direct equity-return signal. The stale relative-return indicator, exact publication deadline, ambiguity rule, and small liquidation or merger risk warrant discounts. The evidence does not support the forecaster’s 0.5." } ], "aggregation": "log-odds-mean", "extremization": 1, "dissent": null, "spread": 0.04, "warnings": [ "all 6 votes came from one vendor (openai); extremization disabled, persona diversity only" ] }, "final": { "p": 0.2571269091648988, "ci80": [ 0.19755279254222627, 0.3293521776574487 ], "method": "stacked-fixed-weights (radar: prior 0.5, council 0.5; no resolved radar items to learn from)", "weights": { "council": 0.5, "prior": 0.5 } }, "stated": 0.5, "history": [ { "asOf": "2026-09-07", "p": 0.2647963179158265, "source": "radar-engine", "runId": "radar-2026-09-07T23-12-09Z-06a58f" }, { "asOf": "2026-09-07", "p": 0.2571269091648988, "source": "radar-engine", "runId": "radar-2026-09-07T23-49-04Z-de8abf" } ], "indicatorStatus": "mixed", "milestones": [], "arena": "capital-credit", "status": "draft", "tier": "T1" }, "indicators": [ { "thesisId": "RT-28", "indicatorId": "li-1", "connector": "manual", "seriesId": "grid-minus-sp500-ytd-total-return", "unit": "percentage-points", "asOf": "2026-09-07", "current": null, "status": "no-data", "history": [], "url": "https://www.ftportfolios.com/etf/GRID", "note": "manual: 'grid-minus-sp500-ytd-total-return' not in manual-series.json" }, { "thesisId": "RT-28", "indicatorId": "li-2", "connector": "fred", "seriesId": "IPG335S", "unit": "index-2017-100", "asOf": "2026-07-01", "current": 101.424, "status": "between", "history": [ { "asOf": "2021-08-01", "value": 101.2307 }, { "asOf": "2021-09-01", "value": 101.2509 }, { "asOf": "2021-10-01", "value": 100.3728 }, { "asOf": "2021-11-01", "value": 99.6056 }, { "asOf": "2021-12-01", "value": 98.9178 }, { "asOf": "2022-01-01", "value": 97.9831 }, { "asOf": "2022-02-01", "value": 98.8539 }, { "asOf": "2022-03-01", "value": 98.5089 }, { "asOf": "2022-04-01", "value": 98.1876 }, { "asOf": "2022-05-01", "value": 96.3099 }, { "asOf": "2022-06-01", "value": 97.2326 }, { "asOf": "2022-07-01", "value": 95.8791 }, { "asOf": "2022-08-01", "value": 95.1732 }, { "asOf": "2022-09-01", "value": 96.1107 }, { "asOf": "2022-10-01", "value": 97.5861 }, { "asOf": "2022-11-01", "value": 95.3402 }, { "asOf": "2022-12-01", "value": 95.6987 }, { "asOf": "2023-01-01", "value": 96.7387 }, { "asOf": "2023-02-01", "value": 96.6048 }, { "asOf": "2023-03-01", "value": 96.3137 }, { "asOf": "2023-04-01", "value": 97.9152 }, { "asOf": "2023-05-01", "value": 98.0307 }, { "asOf": "2023-06-01", "value": 97.602 }, { "asOf": "2023-07-01", "value": 96.9484 }, { "asOf": "2023-08-01", "value": 96.6278 }, { "asOf": "2023-09-01", "value": 96.0037 }, { "asOf": "2023-10-01", "value": 98.2244 }, { "asOf": "2023-11-01", "value": 98.0359 }, { "asOf": "2023-12-01", "value": 98.0651 }, { "asOf": "2024-01-01", "value": 97.1513 }, { "asOf": "2024-02-01", "value": 97.0993 }, { "asOf": "2024-03-01", "value": 98.8427 }, { "asOf": "2024-04-01", "value": 97.4678 }, { "asOf": "2024-05-01", "value": 98.306 }, { "asOf": "2024-06-01", "value": 99.1503 }, { "asOf": "2024-07-01", "value": 99.2296 }, { "asOf": "2024-08-01", "value": 100.7553 }, { "asOf": "2024-09-01", "value": 98.3148 }, { "asOf": "2024-10-01", "value": 98.3476 }, { "asOf": "2024-11-01", "value": 97.1741 }, { "asOf": "2024-12-01", "value": 97.2505 }, { "asOf": "2025-01-01", "value": 98.6586 }, { "asOf": "2025-02-01", "value": 98.9181 }, { "asOf": "2025-03-01", "value": 98.3295 }, { "asOf": "2025-04-01", "value": 96.0371 }, { "asOf": "2025-05-01", "value": 97.8279 }, { "asOf": "2025-06-01", "value": 94.7302 }, { "asOf": "2025-07-01", "value": 95.514 }, { "asOf": "2025-08-01", "value": 95.1896 }, { "asOf": "2025-09-01", "value": 96.6708 }, { "asOf": "2025-10-01", "value": 96.5442 }, { "asOf": "2025-11-01", "value": 97.2376 }, { "asOf": "2025-12-01", "value": 98.2291 }, { "asOf": "2026-01-01", "value": 98.0444 }, { "asOf": "2026-02-01", "value": 98.2922 }, { "asOf": "2026-03-01", "value": 98.9972 }, { "asOf": "2026-04-01", "value": 100.13 }, { "asOf": "2026-05-01", "value": 100.8885 }, { "asOf": "2026-06-01", "value": 100.1522 }, { "asOf": "2026-07-01", "value": 101.424 } ], "url": "https://fred.stlouisfed.org/series/IPG335S", "note": null } ], "indicatorStatus": "mixed", "chainEvents": [ { "seq": 246, "ts": "2026-09-07", "kind": "repriced", "hash": "1a2d4a342f667fd43f6f9fe638d13337141e2ea858c4378e5cd746e38bb8d1e7" }, { "seq": 247, "ts": "2026-09-07", "kind": "repriced", "hash": "d0274d9eb885af659e58387a1d36d3427218bbcf4ebb9a403811271e842c54bd" }, { "seq": 282, "ts": "2026-09-08", "kind": "registered", "hash": "61823e4fc71595437fed3ba361c22c405feb1184d23ff1d8901635ade2961a57" } ], "url": "/predictions/capital-credit/RT-28" }, { "id": "RT-29", "title": "Congress preempts state AI laws by end-2027", "arena": "geopolitics-policy", "tier": "T1", "tag": "CALL", "direction": "shift", "status": "registered", "statement": "A US federal statute signed into law on or before 2027-12-31 expressly preempts, or conditions federal funds on the non-enforcement of, state laws that specifically regulate the development or deployment of AI models.", "resolutionRule": "Hit if a bill enacted into public law (signed by the President or enacted over a veto) on or before 2027-12-31, as shown by \"Became Public Law\" status on Congress.gov, contains text that (a) expressly preempts state or local laws that specifically regulate AI models or AI systems, or (b) withholds or conditions federal funding on a state not enacting or enforcing such laws. Partial preemption (for example, only laws regulating model development, or only laws applying to frontier developers) counts. Preemption limited to a single application area (for example, only political deepfakes or only children's online safety) does not count. Executive orders, agency rules, and litigation outcomes do not count.", "resolutionSource": { "name": "Congress.gov legislation search", "url": "https://www.congress.gov/search?q=%7B%22congress%22%3A%5B%22119%22%2C%22120%22%5D%2C%22source%22%3A%22legislation%22%2C%22search%22%3A%22artificial%20intelligence%20preemption%22%7D" }, "ambiguity": "resolves-against-forecaster", "horizon": "2027-12-31", "p": 0.4, "ci80": [ 0.25, 0.55 ], "consensus": { "value": "Manifold prices a federal AI preemption law passing by 2027-01-03 (end of the 119th Congress) at 14%, down from 24% earlier in 2026, with 13 traders. No Polymarket or Metaculus market covers preemption directly.", "impliedP": 0.25, "impliedBy": "This thesis adds the first year of the 120th Congress to the Manifold window. Treating the first post-midterm year as roughly as likely as the remainder of this Congress, 14% for the shorter window implies about 25% for the longer one.", "source": "Manifold, Federal AI preemption passes by January 3, 2027?", "url": "https://manifold.markets/EricNeyman/federal-ai-preemption-passes-by-jan", "asOf": "2026-09-07", "note": "The Manifold figure is a live market on a shorter horizon; the implied number extends it. My 40% reflects the administration's leverage over states and industry pressure for uniformity, against a midterm calendar." }, "leadingIndicators": [ { "id": "li-1", "label": "Manifold community probability, federal AI preemption by 2027-01-03", "connector": "manual", "seriesId": "manifold-federal-ai-preemption-2027", "unit": "probability", "onTrack": { "op": ">=", "value": 0.25 }, "offTrack": { "op": "<", "value": 0.08 }, "url": "https://manifold.markets/EricNeyman/federal-ai-preemption-passes-by-jan", "note": "Displayed market probability on the Manifold question page. Read weekly; after the market resolves, switch to any successor market for the 120th Congress." }, { "id": "li-2", "label": "States enacting AI laws in the calendar year (count)", "connector": "manual", "seriesId": "ncsl-states-enacting-ai-laws-ytd", "unit": "states", "onTrack": { "op": ">=", "value": 30 }, "offTrack": { "op": "<", "value": 20 }, "url": "https://www.ncsl.org/technology-and-communication/artificial-intelligence-2025-legislation", "note": "Number of states that have enacted at least one AI-specific law in the current calendar year, from the NCSL AI legislation tracker (29 states by mid-2026 per TechPolicy.Press). NCSL publishes one page per year; use the page for the current year. More state activity raises industry pressure for federal preemption." } ], "milestones": [], "falsifier": "No public law with qualifying preemption or funding-condition text is enacted on or before 2027-12-31.", "whyItMatters": "Whether AI is regulated by fifty legislatures or one decides compliance cost, where models get deployed first, and how much leverage the federal government has over the labs. Both parties want something from a preemption deal (industry wants uniformity, safety advocates want a federal floor), which is why a bargain is likelier than the 2025 moratorium fight suggested. The midterm calendar is the main reason to doubt it.", "whatWouldRaise": [ "The Obernolte-Trahan draft or a Senate counterpart formally introduced with bipartisan cosponsors.", "A preemption title attached to must-pass legislation (NDAA, appropriations) in either chamber.", "California or New York enacting a frontier-model law that triggers a federal response." ], "whatWouldCut": [ "A federal court upholding a state AI law against the administration's litigation task force.", "The 120th Congress opening with divided control and no AI bill on either leadership agenda.", "Industry splitting on preemption scope, with developers and deployers lobbying against each other." ], "namedEntities": [ "US Congress" ], "marketCall": false, "recusalCheck": { "passed": true, "keysChecked": [ "employer-leasing", "named-competitor-bookings", "customer-pipeline", "ticker" ] }, "signedBy": "Brian Letort", "registeredAt": "2026-09-08", "amendments": [], "ruleHash": "95291b7b22f19dce37e7c66b70bd5fe9b3f021a5a03a7088582d052e4e7ae59a", "distance": 0.6931471805599455, "distanceLabel": "leans above consensus", "engine": { "id": "RT-29", "prior": { "p": 0.3888888888888889, "method": "reference-class:qtype:timing", "note": "ledger base rate, n=8, horizon 480d; the ledger has no multi-year history" }, "council": { "p": 0.2613530176287422, "votes": [ { "persona": "grid-and-power-planner", "model": "openai/gpt-5.6-sol", "p": 0.27, "rationale": "I use the reference class of federal technology and privacy bills seeking broad state-law displacement: proposals recur, but enactment is uncommon without agreement on a federal regulatory substitute. The remaining 119th Congress provides little legislative time, while 2027 adds one first-session opportunity and several must-pass vehicles. Partial preemption and funding conditions broaden the pathways, but the rule excludes easier, application-specific provisions. Senate procedural and federalism objections are substantial, and enactment—not committee passage—is required. Compounding an approximately 0.11 hazard through January 2027 with an approximately 0.18 hazard during the remainder of 2027 yields about 0.27. The 0.14 shorter-horizon market is compatible with this estimate; 0.40 appears high." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.24, "rationale": "My reference class is contentious technology-policy preemption enacted within roughly 16 months: most proposals stall, especially where federalism concerns create bipartisan resistance. The qualifying event also requires presidential enactment and broad AI-model/system language; narrow deepfake, child-safety, or sector-specific provisions fail the rule. A must-pass funding bill offers a plausible vehicle, and industry demand for national uniformity raises the chance above a typical standalone bill. However, both indicators provide no data, and the shorter-horizon market at 0.14 indicates limited near-term momentum. Extending through 2027 adds another legislative session, but not enough to support the forecaster’s 0.4." }, { "persona": "enterprise-cio", "model": "openai/gpt-5.6-sol", "p": 0.27, "rationale": "My reference class is controversial federal technology preemption enacted through must-pass or budget legislation. I assign roughly 0.14 through the 119th Congress, consistent with the thin quoted market, then a conditional 0.15 during 2027; compounding gives about 0.27. The funding-condition route and partial, sector-broad preemption expand the paths, while proliferating state rules create business pressure for uniformity. Against that, federalism opposition is bipartisan, Senate passage is a high hurdle, and 2027 is only the first year of a new Congress, when committee work and coalition-building often outrun enactment. The generic 0.389 structural prior likely underweights the claim’s specific text and deadline." }, { "persona": "credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.27, "rationale": "Reference class: controversial federal preemption proposals requiring enactment within roughly 16 months. The alternative funding-condition route and eligibility of partial, sector-wide preemption create plausible paths through reconciliation or must-pass legislation. But the claim still requires qualifying statutory text to survive both chambers and presidential action; executive pressure, litigation, studies, or application-specific rules do not count. Federalism concerns and a midterm-disrupted calendar reduce the annual enactment hazard. With no current indicator data, I place more weight on the shorter-horizon 0.14 market signal than on the generic 0.3889 timing prior, while allowing meaningful 2027 opportunity." }, { "persona": "geopolitics-policy-analyst", "model": "openai/gpt-5.6-sol", "p": 0.28, "rationale": "I use the reference class of contentious federal technology/privacy preemption proposals: enactment is uncommon without bipartisan committee support or attachment to must-pass legislation. The shorter-horizon market implies roughly 0.14 through the 119th Congress. Conditional on failure there, I assign about a 0.16 hazard during 2027, yielding approximately 0.28 cumulatively. State-law proliferation and industry demand for uniformity increase pressure, but broad AI preemption faces federalism objections, Senate constraints, and possible divided government after the midterms. The literal rule also excludes application-specific provisions and requires enactment, not merely House passage, executive action, or litigation." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.24, "rationale": "Base rate: I use 0.18 for enactment within roughly 16 months of a controversial federal technology bill overriding state regulation; broad US privacy and technology uniformity bills commonly stall despite industry support. I adjust upward because direct preemption or a federal-funding condition qualifies, partial model-development preemption counts, and the horizon spans parts of two Congresses. I adjust downward for bicameral passage, presidential enactment, federalism opposition, the midterm transition, and exclusion of application-specific provisions. The supplied 0.14 shorter-horizon market supports a low near-term hazard, though it is thin. No leading-indicator data or qualifying enacted text is supplied. The 0.3889 structural prior is too broad for this statutory threshold." } ], "aggregation": "log-odds-mean", "extremization": 1, "dissent": null, "spread": 0.04, "warnings": [ "all 6 votes came from one vendor (openai); extremization disabled, persona diversity only" ] }, "final": { "p": 0.3218099092105298, "ci80": [ 0.19558716239552762, 0.46379928096139633 ], "method": "stacked-fixed-weights (radar: prior 0.5, council 0.5; no resolved radar items to learn from)", "weights": { "council": 0.5, "prior": 0.5 } }, "stated": 0.4, "history": [ { "asOf": "2026-09-07", "p": 0.3284656895477064, "source": "radar-engine", "runId": "radar-2026-09-07T23-12-09Z-06a58f" }, { "asOf": "2026-09-07", "p": 0.3218099092105298, "source": "radar-engine", "runId": "radar-2026-09-07T23-49-04Z-de8abf" } ], "indicatorStatus": "no-data", "milestones": [], "arena": "geopolitics-policy", "status": "draft", "tier": "T1" }, "indicators": [ { "thesisId": "RT-29", "indicatorId": "li-1", "connector": "manual", "seriesId": "manifold-federal-ai-preemption-2027", "unit": "probability", "asOf": "2026-09-07", "current": null, "status": "no-data", "history": [], "url": "https://manifold.markets/EricNeyman/federal-ai-preemption-passes-by-jan", "note": "manual: 'manifold-federal-ai-preemption-2027' not in manual-series.json" }, { "thesisId": "RT-29", "indicatorId": "li-2", "connector": "manual", "seriesId": "ncsl-states-enacting-ai-laws-ytd", "unit": "states", "asOf": "2026-09-07", "current": null, "status": "no-data", "history": [], "url": "https://www.ncsl.org/technology-and-communication/artificial-intelligence-2025-legislation", "note": "manual: 'ncsl-states-enacting-ai-laws-ytd' not in manual-series.json" } ], "indicatorStatus": "no-data", "chainEvents": [ { "seq": 248, "ts": "2026-09-07", "kind": "repriced", "hash": "ee45338619877422aba66b0b26a2e8f7e31611e6d443beed6ed93819bba3b287" }, { "seq": 249, "ts": "2026-09-07", "kind": "repriced", "hash": "36202394e2c9e0421892d0940f96472abf277d8545d61cd3a48c48af56ce858d" }, { "seq": 283, "ts": "2026-09-08", "kind": "registered", "hash": "76b6a907ffd6fb637345f7331497ac17fb833a089b60a58479e4d1f14dbf3fe5" } ], "url": "/predictions/geopolitics-policy/RT-29" }, { "id": "RT-30", "title": "NVIDIA China share stays under 10% through FY2029", "arena": "geopolitics-policy", "tier": "T2", "tag": "WARNING", "direction": "down", "status": "registered", "statement": "China (including Hong Kong) accounts for less than 10% of NVIDIA's total revenue in each of its fiscal 2027, 2028, and 2029 Forms 10-K, meaning US export controls permanently remove China as a material NVIDIA market.", "resolutionRule": "For each NVIDIA (CIK 0001045810) Form 10-K for fiscal years ending January 2027, January 2028, and January 2029, divide the \"China (including Hong Kong)\" line in the revenue-by-geography table (customer headquarters basis) by total revenue. Hit if all three ratios are below 10.0%. Any one fiscal year at or above 10.0% resolves the thesis as a miss when that 10-K is filed. If NVIDIA stops disclosing a China line or changes the basis so that China is no longer separately shown, resolve against the forecaster (miss). Score on filings on EDGAR dated on or before 2029-03-31.", "resolutionSource": { "name": "NVIDIA Form 10-K filings on SEC EDGAR (CIK 0001045810)", "url": "https://www.sec.gov/cgi-bin/browse-edgar?action=getcompany&CIK=0001045810&type=10-K&dateb=&owner=include&count=40" }, "ambiguity": "resolves-against-forecaster", "horizon": "2029-03-31", "p": 0.85, "ci80": [ 0.72, 0.93 ], "consensus": { "value": "China was 9.1% of NVIDIA revenue in fiscal 2026 ($19.7 billion of $215.9 billion), down from 13.1% in fiscal 2025. NVIDIA's outlook assumes zero China data center compute revenue, while Jensen Huang calls China a $50 billion annual accelerator market growing 50% a year and Commerce licensed H200 sales to about ten Chinese buyers in May 2026 with a theoretical ceiling near 750,000 units. Street models treat China as upside, not baseline.", "impliedP": 0.6, "impliedBy": "Manifold prices NVIDIA resuming top-GPU sales to China by April 2027 at 15% and B30A approval by 2027 at 19%. Even with a licensed return, China would need $60 billion-plus a year against a $600 billion-plus revenue base to reach 10%; the market's view of a partial reopening implies about 60% that China stays under 10% in all three years.", "source": "NVIDIA fiscal 2026 10-K revenue by geography, via edgar.tools", "url": "https://app.edgar.tools/companies/NVDA/disclosures/segments", "asOf": "2026-02-25", "note": "No prediction market prices NVIDIA's China share directly; the implied figure combines the Manifold export-license markets with the arithmetic of a revenue base growing 70% a year (fiscal 2028 guidance). I am at 85% because the denominator does most of the work regardless of policy." }, "leadingIndicators": [ { "id": "li-1", "label": "NVIDIA China (incl. Hong Kong) share of quarterly revenue", "connector": "manual", "seriesId": "nvda-china-share-of-revenue-quarterly", "unit": "percent", "onTrack": { "op": "<", "value": 8 }, "offTrack": { "op": ">=", "value": 11 }, "url": "https://investor.nvidia.com/financial-info/sec-filings/default.aspx", "note": "China (including Hong Kong) revenue divided by total revenue from the geographic table in each 10-Q or 10-K, customer headquarters basis. Compute from the filing; do not use press-release commentary." }, { "id": "li-2", "label": "NVIDIA total quarterly revenue", "connector": "sec-xbrl", "seriesId": "0001045810:Revenues:USD", "unit": "usd", "onTrack": { "op": ">=", "value": 90000000000 }, "offTrack": { "op": "<", "value": 70000000000 }, "url": "https://data.sec.gov/api/xbrl/companyconcept/CIK0001045810/us-gaap/Revenues.json", "note": "Three-month total revenue ($81.6 billion in Q1 FY2027). A growing denominator makes the 10% bar harder for China to clear even if some licensed sales resume. Thresholds apply to Q2 FY2027 and later." } ], "milestones": [ { "id": "m1", "date": "2027-03-31", "statement": "NVIDIA's fiscal 2027 10-K shows China (including Hong Kong) below 10% of total revenue.", "p": 0.8 }, { "id": "m2", "date": "2028-03-31", "statement": "NVIDIA's fiscal 2028 10-K shows China (including Hong Kong) below 10% of total revenue.", "p": 0.65 } ], "falsifier": "Any of the fiscal 2027, 2028, or 2029 10-Ks shows China (including Hong Kong) at 10.0% or more of total revenue.", "whyItMatters": "Export controls are usually debated as policy; this thesis scores them as revenue. If China stays under a tenth of NVIDIA's sales for three more years while the company doubles, the controls will have done what they were meant to do commercially, and Chinese demand will have moved to Huawei and domestic silicon for good. That is the fork that decides whether there is one global accelerator stack or two.", "whatWouldRaise": [ "H200 deliveries to China staying below 200,000 units through fiscal 2027 despite US licenses.", "Beijing directing state-linked buyers to domestic accelerators for new data centers.", "A US rule formalizing a revenue levy or per-buyer cap that makes China sales uneconomic." ], "whatWouldCut": [ "A Blackwell-class part (B30A or successor) licensed for China with volume shipments in a 10-Q.", "A US-China trade agreement that removes accelerator export licensing for named buyers.", "NVIDIA total revenue growth stalling, which would raise China's share arithmetically." ], "namedEntities": [ "NVIDIA", "China", "Huawei" ], "marketCall": false, "recusalCheck": { "passed": true, "keysChecked": [ "employer-leasing", "named-competitor-bookings", "customer-pipeline", "ticker" ] }, "signedBy": "Brian Letort", "registeredAt": "2026-09-08", "amendments": [], "ruleHash": "8287c9af41629ea9f1c37cfc02e84b8bc3c54dd5e5ede8142328b7f695a33671", "distance": 1.329135947279942, "distanceLabel": "clearly above consensus", "engine": { "id": "RT-30", "prior": { "p": 0.98, "method": "sec-xbrl-loglinear-bootstrap:li-2", "note": "log-linear growth +0.597/yr on the last 16 points, extrapolated from 2026-07-26; P(li-2 satisfies onTrack >= 90000000000.0 at 2029-03-31); proxy for the thesis, not its rule" }, "council": { "p": 0.5571599394487103, "votes": [ { "persona": "grid-and-power-planner", "model": "openai/gpt-5.6-sol", "p": 0.6, "rationale": "Reference class: three-year persistence of U.S. technology export restrictions and geographic revenue shares starting within one percentage point of a cutoff. FY2026’s 9.1% leaves little cushion. The latest $96.2 billion quarterly total supports denominator growth but does not measure China, so the 0.98 structural prior overweights a numerator-free trend. Licensed H200 sales or later policy relaxation could lift China above 10% in at least one year, especially as companywide growth decelerates. I estimate roughly 0.84, 0.80, and 0.78 conditional annual survival; policy correlation offsets some naive compounding, while the separate-disclosure requirement adds miss risk." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.56, "rationale": "My reference class is multi-year semiconductor export-control regimes, where restrictions persist but licensing and product redesign create recurring reopening risk. Fiscal 2026 starts only modestly below the threshold at 0.091. Rapid total-revenue growth strongly helps, and the supplied quarterly series supports that denominator thesis, but it provides no China numerator. Licensed H200 sales could lift China revenue materially, especially as overall growth decelerates in fiscal 2028-2029. The three-year conjunction compounds threshold-crossing risk. I also penalize the thesis because, after several sub-threshold years, NVIDIA could stop separately presenting China; the rule counts that as a miss. The denominator-only structural prior therefore overstates confidence." }, { "persona": "enterprise-cio", "model": "openai/gpt-5.6-sol", "p": 0.5, "rationale": "My reference class is three-year persistence of country revenue shares around a reporting threshold during changing trade controls, not revenue-growth extrapolations. FY2027 is likely below 10%: the reported FY2026 share was 9.1%, the FY2027 denominator is expanding rapidly, and baseline China data-center revenue is zero. I assign 0.88 to FY2027, then conditional survival of 0.78 and 0.76 for FY2028–29, plus 0.95 for continued separate disclosure and timely filing: about 0.50. Licensing creates a pathway for the numerator to recover. The structural prior captures denominator growth but not policy-driven China sales or the explicit disclosure-failure clause." }, { "persona": "credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.61, "rationale": "Reference class: three-year survival of a near-threshold geographic revenue share under a volatile trade-control regime. Fiscal 2026 began at 9.1%, while the supplied quarterly revenue run-rate makes the fiscal 2027 denominator protective. However, licensed H200 sales can lift the numerator, and policy easing or slower aggregate growth raises later breach risk. I assign 12%-18% annual breach hazards; allowing for their correlation gives roughly 0.64 for all three economic tests, then a 3-point penalty for the disclosure/basis failure clause. This is below 0.85 because the 0.98 structural prior extrapolates total revenue without modeling China revenue or the conjunctive filing requirement." }, { "persona": "geopolitics-policy-analyst", "model": "openai/gpt-5.6-sol", "p": 0.59, "rationale": "I use the reference class of export-control regimes over three years: restrictions usually persist, but licensing, product redesigns, diplomatic bargaining, and enforcement gaps create meaningful revenue recovery risk. FY2026 starts narrowly below the threshold at 9.1%. Rapid non-China growth makes FY2027 favorable, although licensed H200 sales could materially raise the numerator. By FY2028–FY2029, China’s large accelerator demand and multiple policy-reset opportunities increase the chance that at least one year reaches 10%. Because the test is conjunctive, correlated annual estimates around 0.75–0.82 do not justify 0.85 for all three. I also apply a small penalty for changed geographic disclosure, which automatically causes a miss." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.48, "rationale": "Base rate: I use 0.50 for a geography sitting within one percentage point of a cutoff to remain on the same side in three consecutive annual filings. Rapid total-revenue growth and export controls raise the estimate, but FY2026 was only 9.1%, licensed H200 sales create numerator upside, and policy can change over 2.5 years. This is a three-leg conjunction plus a named-line, named-basis disclosure requirement; correlated outcomes still compound failure risk, and any disclosure change loses. The 0.98 structural prior extrapolates total revenue while omitting China revenue. The stated 0.85 also exceeds its own 0.65 FY2028 milestone, which upper-bounds the full conjunction." } ], "aggregation": "log-odds-mean", "extremization": 1, "dissent": null, "spread": 0.13, "warnings": [ "all 6 votes came from one vendor (openai); extremization disabled, persona diversity only" ] }, "final": { "p": 0.8870275428712252, "ci80": [ 0.7725184452584984, 0.9034126797765866 ], "method": "stacked-fixed-weights (radar: prior 0.5, council 0.5; no resolved radar items to learn from)", "weights": { "council": 0.5, "prior": 0.5 } }, "stated": 0.85, "history": [ { "asOf": "2026-09-07", "p": 0.8962602094018436, "source": "radar-engine", "runId": "radar-2026-09-07T23-12-09Z-06a58f" }, { "asOf": "2026-09-07", "p": 0.8870275428712252, "source": "radar-engine", "runId": "radar-2026-09-07T23-49-04Z-de8abf" } ], "indicatorStatus": "on-track", "milestones": [ { "id": "m1", "enginePrior": 0.5, "councilP": 0.81852660647133 }, { "id": "m2", "enginePrior": 0.5, "councilP": null } ], "arena": "geopolitics-policy", "status": "draft", "tier": "T2" }, "indicators": [ { "thesisId": "RT-30", "indicatorId": "li-1", "connector": "manual", "seriesId": "nvda-china-share-of-revenue-quarterly", "unit": "percent", "asOf": "2026-09-07", "current": null, "status": "no-data", "history": [], "url": "https://investor.nvidia.com/financial-info/sec-filings/default.aspx", "note": "manual: 'nvda-china-share-of-revenue-quarterly' not in manual-series.json" }, { "thesisId": "RT-30", "indicatorId": "li-2", "connector": "sec-xbrl", "seriesId": "0001045810:Revenues:USD", "unit": "usd", "asOf": "2026-07-26", "current": 96221000000, "status": "on-track", "history": [ { "asOf": "2009-10-25", "value": 903206000 }, { "asOf": "2010-01-31", "value": 982488000 }, { "asOf": "2010-05-02", "value": 1001813000 }, { "asOf": "2010-07-31", "value": 811208000 }, { "asOf": "2010-10-31", "value": 843912000 }, { "asOf": "2011-01-30", "value": 886376000 }, { "asOf": "2011-05-01", "value": 962039000 }, { "asOf": "2011-07-31", "value": 1016517000 }, { "asOf": "2011-10-30", "value": 1066180000 }, { "asOf": "2012-01-29", "value": 953194000 }, { "asOf": "2012-04-29", "value": 924877000 }, { "asOf": "2012-07-29", "value": 1044270000 }, { "asOf": "2012-10-28", "value": 1204110000 }, { "asOf": "2013-01-27", "value": 1106902000 }, { "asOf": "2013-04-28", "value": 954739000 }, { "asOf": "2013-07-28", "value": 977238000 }, { "asOf": "2013-10-27", "value": 1053967000 }, { "asOf": "2014-01-26", "value": 1144218000 }, { "asOf": "2014-04-27", "value": 1103000000 }, { "asOf": "2014-07-27", "value": 1103000000 }, { "asOf": "2014-10-26", "value": 1225000000 }, { "asOf": "2015-01-25", "value": 1251000000 }, { "asOf": "2015-04-26", "value": 1151000000 }, { "asOf": "2015-07-26", "value": 1153000000 }, { "asOf": "2015-10-25", "value": 1305000000 }, { "asOf": "2016-01-31", "value": 1401000000 }, { "asOf": "2016-05-01", "value": 1305000000 }, { "asOf": "2016-07-31", "value": 1428000000 }, { "asOf": "2016-10-30", "value": 2004000000 }, { "asOf": "2017-01-29", "value": 2173000000 }, { "asOf": "2017-04-30", "value": 1937000000 }, { "asOf": "2017-07-30", "value": 2230000000 }, { "asOf": "2017-10-29", "value": 2636000000 }, { "asOf": "2018-01-28", "value": 2911000000 }, { "asOf": "2018-04-29", "value": 3207000000 }, { "asOf": "2018-07-29", "value": 3123000000 }, { "asOf": "2018-10-28", "value": 3181000000 }, { "asOf": "2019-04-28", "value": 2220000000 }, { "asOf": "2019-07-28", "value": 2579000000 }, { "asOf": "2019-10-27", "value": 3014000000 }, { "asOf": "2020-04-26", "value": 3080000000 }, { "asOf": "2020-07-26", "value": 3866000000 }, { "asOf": "2020-10-25", "value": 4726000000 }, { "asOf": "2021-05-02", "value": 5661000000 }, { "asOf": "2021-08-01", "value": 6507000000 }, { "asOf": "2021-10-31", "value": 7103000000 }, { "asOf": "2022-05-01", "value": 8288000000 }, { "asOf": "2022-07-31", "value": 6704000000 }, { "asOf": "2022-10-30", "value": 5931000000 }, { "asOf": "2023-04-30", "value": 7192000000 }, { "asOf": "2023-07-30", "value": 13507000000 }, { "asOf": "2023-10-29", "value": 18120000000 }, { "asOf": "2024-04-28", "value": 26044000000 }, { "asOf": "2024-07-28", "value": 30040000000 }, { "asOf": "2024-10-27", "value": 35082000000 }, { "asOf": "2025-04-27", "value": 44062000000 }, { "asOf": "2025-07-27", "value": 46743000000 }, { "asOf": "2025-10-26", "value": 57006000000 }, { "asOf": "2026-04-26", "value": 81615000000 }, { "asOf": "2026-07-26", "value": 96221000000 } ], "url": "https://data.sec.gov/api/xbrl/companyconcept/CIK0001045810/us-gaap/Revenues.json", "note": "quarterly frames" } ], "indicatorStatus": "on-track", "chainEvents": [ { "seq": 250, "ts": "2026-09-07", "kind": "repriced", "hash": "40a28e9fdf9febf4c6b226a66e44e1f5dd73a27ebfc128112ee5f32d2e46c842" }, { "seq": 251, "ts": "2026-09-07", "kind": "repriced", "hash": "81762a4c5b559f507b547e2ebcdeaaf31c8d6bbb13b4074cb414a8ecabadbfe3" }, { "seq": 284, "ts": "2026-09-08", "kind": "registered", "hash": "2cdd0d33cb4af0e74800412b45a395c193f960c7631e41de89403dee03794645" } ], "url": "/predictions/geopolitics-policy/RT-30" }, { "id": "RT-31", "title": "Brussels fines a US frontier lab under AI Act", "arena": "geopolitics-policy", "tier": "T2", "tag": "CALL", "direction": "shift", "status": "registered", "statement": "By 2028-12-31 the European Commission publishes a decision imposing a fine of at least EUR 10 million on a US-headquartered provider of a general-purpose AI model under Article 101 of the EU AI Act.", "resolutionRule": "Hit if, on or before 2028-12-31, the European Commission (including the European AI Office) publishes on the Commission press corner or the AI Office pages a decision imposing a fine under Article 101 of Regulation (EU) 2024/1689 on a general-purpose AI model provider whose ultimate parent is headquartered in the United States, with a stated amount of EUR 10,000,000 or more (a fine expressed as a turnover percentage counts if the Commission's release states a euro amount at or above the threshold). Fines by national market surveillance authorities under Article 99, periodic penalty payments, and settlements or commitments without a fine do not count. Fines on providers headquartered outside the United States do not count. The decision counts on the publication date even if the provider appeals to the Court of Justice.", "resolutionSource": { "name": "European Commission press corner", "url": "https://ec.europa.eu/commission/presscorner/home/en" }, "ambiguity": "resolves-against-forecaster", "horizon": "2028-12-31", "p": 0.55, "ci80": [ 0.38, 0.7 ], "consensus": { "value": "Article 101 fining powers activated 2026-08-02. As of late August 2026 no fine, formal investigation, or enforcement decision had been confirmed; the AI Office said technical compliance dialogues are its preferred first tool. On 2026-08-29 the Commission confirmed formal information requests to more than 30 general-purpose AI providers, the first use of the new powers.", "impliedP": 0.3, "impliedBy": "Commentators expect a first GPAI fine in the EUR 5 to 25 million range within 12 to 18 months, but rank xAI and Meta as the likeliest targets and note the Commission's preference for dialogue and its exposure to US trade retaliation. A EUR 10 million-plus fine specifically on a US-headquartered lab by end-2028 comes out near 30% on that reading.", "source": "EU Perspectives, on the Commission's first formal AI Act information requests", "url": "https://euperspectives.eu/2026/09/the-ai-act-gives-brussels-new-powers-frontier-labs-are-first-in-line/", "asOf": "2026-09-01", "note": "Manifold prices \"any formal EU AI Act enforcement action against a frontier lab in 2026\" at 54%, but that market counts information requests (already sent) and is a different question. The DMA precedent (powers March 2024, first fines on Apple and Meta April 2025) is why I sit above the implied number." }, "leadingIndicators": [ { "id": "li-1", "label": "Formal Article 101 proceedings opened against GPAI providers (cumulative)", "connector": "manual", "seriesId": "eu-ai-office-article101-proceedings-count", "unit": "count", "onTrack": { "op": ">=", "value": 1 }, "offTrack": { "op": "==", "value": 0 }, "url": "https://digital-strategy.ec.europa.eu/en/policies/ai-office", "note": "Count of general-purpose AI providers against which the Commission has publicly opened formal non-compliance proceedings (beyond information requests) under Articles 91 to 101. Read from AI Office news and the Commission press corner; thresholds apply from 2027-06-30." }, { "id": "li-2", "label": "Signatories to the GPAI Code of Practice", "connector": "manual", "seriesId": "eu-gpai-code-of-practice-signatories", "unit": "count", "onTrack": { "op": "<=", "value": 30 }, "offTrack": { "op": ">", "value": 40 }, "url": "https://digital-strategy.ec.europa.eu/en/policies/contents-code-gpai", "note": "Number of providers listed as signatories on the Commission's GPAI Code of Practice page. A provider that declines to sign is the likeliest first enforcement target; broad signature reduces fine probability." } ], "milestones": [ { "id": "m1", "date": "2027-06-30", "statement": "The Commission publicly opens a formal non-compliance proceeding against at least one general-purpose AI model provider.", "p": 0.65 }, { "id": "m2", "date": "2027-12-31", "statement": "A national authority or the Commission publishes any AI Act fine, of any size, against any provider or deployer.", "p": 0.6 } ], "falsifier": "No Commission decision imposing an Article 101 fine of EUR 10 million or more on a US-headquartered provider is published by 2028-12-31.", "whyItMatters": "The AI Act only matters to the labs if it is enforced against them. A fine on a US general-purpose model provider is the moment the regulation moves from paperwork to a cost of doing business in Europe, and it will set the template for how frontier labs treat every other jurisdiction's rules. It also tests whether the Commission will act against US companies while trade negotiations are live.", "whatWouldRaise": [ "A provider publicly refusing to answer the August 2026 information requests or answering incompletely.", "A frontier model incident in the EU attributed to a documented safety shortfall.", "The AI Office publishing a preliminary findings notice naming a specific provider." ], "whatWouldCut": [ "All major providers signing the GPAI Code of Practice and the Commission confirming compliance.", "A further Digital Omnibus deferring or softening GPAI obligations.", "A US-EU trade arrangement that explicitly pauses AI Act enforcement against US firms." ], "namedEntities": [ "European Commission" ], "marketCall": false, "recusalCheck": { "passed": true, "keysChecked": [ "employer-leasing", "named-competitor-bookings", "customer-pipeline", "ticker" ] }, "signedBy": "Brian Letort", "registeredAt": "2026-09-08", "amendments": [], "ruleHash": "d5dd4077132c077b3a55f5c31f95e0c5785dda4a36e130f107b68dd336e60ea0", "distance": 1.047968555849355, "distanceLabel": "clearly above consensus", "engine": { "id": "RT-31", "prior": null, "council": { "p": 0.38761360669410594, "votes": [ { "persona": "grid-and-power-planner", "model": "openai/gpt-5.6-sol", "p": 0.37, "rationale": "I use first-penalty timing under new EU digital regimes, especially the DMA and DSA, as the reference class: a roughly 0.20 annual hazard of a qualifying decision over the remaining 2.3 years gives about 0.40 cumulatively. The reported information requests establish an enforcement pipeline, and US parents are prominent among GPAI providers; conditional on a corporate fine, EUR 10 million is a relatively low threshold. However, no formal proceeding is reported, compliance dialogue is preferred, and the rule requires a Commission-published final fine under Article 101. Investigation, defense, and decision drafting can readily extend beyond 2028. Those conjunctive filters reduce my estimate below 0.55." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.42, "rationale": "I use first penalties under new EU digital regimes as the reference class: typically one to three years from activated powers to a published fine, with the DMA’s roughly 13-month precedent unusually fast because targets and obligations were already defined. The 28-month window, broad information requests, and concentration of frontier GPAI among US parents create meaningful hazard. But resolution is conjunctive: the Commission must escalate from dialogue, establish noncompliance, use Article 101, fine at least EUR 10 million, target a US parent, and publish by the deadline. No confirmed proceeding in the supplied snapshot leaves substantial procedural and timing risk, so I am below 0.55." }, { "persona": "enterprise-cio", "model": "openai/gpt-5.6-sol", "p": 0.38, "rationale": "I use first-enforcement timing for new Commission-run digital regimes, assigning conditional hazards of 0.14 through 2027 and 0.28 in 2028; compounding gives about 0.38. Information requests are a meaningful precursor, and US firms comprise many prominent general-purpose model providers. However, no formal proceeding is confirmed, technical dialogue is the preferred initial route, and complex investigations can exceed two years. The literal rule is demanding: requests, proceedings, commitments, periodic payments, national fines, and smaller sanctions all miss. If a qualifying US provider is fined, EUR 10 million is unlikely to be the main bottleneck; reaching and publishing a final Article 101 decision is." }, { "persona": "credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.43, "rationale": "Using first-penalty timing under new EU digital regimes as the reference class, I estimate a 0.16 qualifying-fine hazard through 2027 and 0.32 during 2028, implying about 0.43 cumulatively. The reported information requests establish an enforcement pipeline, and EUR 10 million is modest relative to Article 101’s ceiling for a major provider. However, the claim is conjunctive: the Commission must progress from dialogue to formal proceedings, establish non-compliance, choose a fine rather than remediation, target a US-parented provider, exceed the threshold, and publish through the specified channels. No confirmed proceeding and missing indicator data keep this below the forecaster’s 0.55." }, { "persona": "geopolitics-policy-analyst", "model": "openai/gpt-5.6-sol", "p": 0.41, "rationale": "I use the reference class of first enforcement decisions under new centralized EU digital regimes, especially the DMA, rather than information requests. Roughly 2.3 years remain. A 0.24–0.28 annual hazard of a published Article 101 fine gives about 0.47–0.53 cumulatively; discounts for a US parent, the EUR 10 million threshold, and exact publication requirements pull this lower, while the broad information-request wave supports 0.41. Dialogue-first practice, no confirmed proceeding, procedural rights, and the AI Office’s institutional novelty constrain timing. The forecaster’s 0.55 overweights the DMA analogy and underweights the conjunctive resolution filters." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.32, "rationale": "Base rate: about 0.50 for the European Commission publishing a first material fine under a new centralized digital regime within roughly 2.5 years. Formal information requests raise enforcement risk, and US providers are prominent likely targets. However, no formal proceeding is reported, compliance dialogues are preferred, and investigations plus defense rights can consume much of the horizon. The resolution is strongly conjunctive: Commission rather than national action, Article 101, a GPAI provider, US ultimate parent, at least EUR 10 million, and publication by the deadline. The DMA timing precedent helps, but transferability to technically complex GPAI enforcement is limited. These filters reduce the estimate well below the forecaster’s 0.55." } ], "aggregation": "log-odds-mean", "extremization": 1, "dissent": null, "spread": 0.11, "warnings": [ "all 6 votes came from one vendor (openai); extremization disabled, persona diversity only" ] }, "final": { "p": 0.38761360669410594, "ci80": [ 0.30845453045991605, 0.47318713140955543 ], "method": "council-only:log-odds-mean", "weights": { "council": 1 } }, "stated": 0.55, "history": [ { "asOf": "2026-09-07", "p": 0.3561538073365972, "source": "radar-engine", "runId": "radar-2026-09-07T23-12-09Z-06a58f" }, { "asOf": "2026-09-07", "p": 0.38761360669410594, "source": "radar-engine", "runId": "radar-2026-09-07T23-49-04Z-de8abf" } ], "indicatorStatus": "no-data", "milestones": [ { "id": "m1", "enginePrior": 0.65, "councilP": null }, { "id": "m2", "enginePrior": 0.3888888888888889, "councilP": null } ], "arena": "geopolitics-policy", "status": "draft", "tier": "T2" }, "indicators": [ { "thesisId": "RT-31", "indicatorId": "li-1", "connector": "manual", "seriesId": "eu-ai-office-article101-proceedings-count", "unit": "count", "asOf": "2026-09-07", "current": null, "status": "no-data", "history": [], "url": "https://digital-strategy.ec.europa.eu/en/policies/ai-office", "note": "manual: 'eu-ai-office-article101-proceedings-count' not in manual-series.json" }, { "thesisId": "RT-31", "indicatorId": "li-2", "connector": "manual", "seriesId": "eu-gpai-code-of-practice-signatories", "unit": "count", "asOf": "2026-09-07", "current": null, "status": "no-data", "history": [], "url": "https://digital-strategy.ec.europa.eu/en/policies/contents-code-gpai", "note": "manual: 'eu-gpai-code-of-practice-signatories' not in manual-series.json" } ], "indicatorStatus": "no-data", "chainEvents": [ { "seq": 252, "ts": "2026-09-07", "kind": "repriced", "hash": "f769e78cdff5894dd52b1fe9967b50c876c1e7644916424f34be4887ebef7e4f" }, { "seq": 253, "ts": "2026-09-07", "kind": "repriced", "hash": "728e186313f598392880232f4336737b42e768afb8d2a098771732a0359f89d4" }, { "seq": 285, "ts": "2026-09-08", "kind": "registered", "hash": "4d6432ecfa3c70d2249c69727e6f922ec64fe1aa2edddbf9faf2af30e4c9dc66" } ], "url": "/predictions/geopolitics-policy/RT-31" }, { "id": "RT-32", "title": "Fewer than three EU AI gigafactories run by 2030", "arena": "geopolitics-policy", "tier": "T3", "tag": "WARNING", "direction": "down", "status": "registered", "statement": "Fewer than three of the AI gigafactories procured under the EuroHPC Joint Undertaking's 2026 call are declared operational by EuroHPC on or before 2030-06-30, against a Commission plan for up to seven facilities running from mid-2028.", "resolutionRule": "Count AI gigafactories selected under EuroHPC call EUROHPC-2026-CEI-AIGF-01 (or its direct successor procurement) for which EuroHPC JU or the European Commission has published, on or before 2030-06-30, a press release or system page stating that the facility is operational, inaugurated, or has begun delivering compute access time under the framework contract. Hit if that count is zero, one, or two. Three or more resolves the thesis as a miss. Existing EuroHPC AI Factories (upgrades of pre-2026 supercomputers) do not count. If the programme is cancelled before any facility operates, the count is zero and the thesis hits.", "resolutionSource": { "name": "EuroHPC Joint Undertaking, AI Gigafactories", "url": "https://www.eurohpc-ju.europa.eu/eurohpc-joint-undertaking-launches-ai-gigafactories-call-2026-07-30_en" }, "ambiguity": "resolves-against-forecaster", "horizon": "2030-12-31", "p": 0.5, "ci80": [ 0.35, 0.65 ], "consensus": { "value": "EuroHPC's 30 July 2026 call targets up to seven AI gigafactories (four medium, three large) with awards in early 2027, construction starting in 2027, and operation within 18 months of contract signature, that is by mid-to-late 2028. Public funding is up to EUR 10 billion with more than EUR 20 billion of private investment expected; 76 consortia expressed preliminary interest.", "impliedP": 0.25, "impliedBy": "The official schedule has up to seven facilities operating by late 2028, two years before this thesis's cut-off; a plan that delivers even half of its sites on a two-year slip still clears three. Taken at face value with normal EU programme slippage, the plan implies about 25% that fewer than three are running by mid-2030.", "source": "EuroHPC JU press release, AI Gigafactories call launch", "url": "https://www.eurohpc-ju.europa.eu/eurohpc-joint-undertaking-launches-ai-gigafactories-call-2026-07-30_en", "asOf": "2026-07-30", "note": "No prediction market covers the programme. The consensus is the official schedule; this thesis expects grid connection, financing, and accelerator supply to hold it to at most two operating sites by mid-2030." }, "leadingIndicators": [ { "id": "li-1", "label": "AI gigafactory framework contracts signed (cumulative)", "connector": "manual", "seriesId": "eurohpc-aigf-contracts-signed", "unit": "count", "onTrack": { "op": "<=", "value": 3 }, "offTrack": { "op": ">=", "value": 5 }, "url": "https://www.eurohpc-ju.europa.eu/news_en", "note": "Number of consortia with a signed AI gigafactory framework contract, per EuroHPC JU news releases. Thresholds apply from 2027-09-30; fewer signatures early means fewer sites can be running by 2030." }, { "id": "li-2", "label": "AI gigafactories declared operational (cumulative)", "connector": "manual", "seriesId": "eurohpc-aigf-operational", "unit": "count", "onTrack": { "op": "<=", "value": 1 }, "offTrack": { "op": ">=", "value": 2 }, "url": "https://www.eurohpc-ju.europa.eu/news_en", "note": "Count of selected gigafactories with a EuroHPC or Commission statement of operation or first delivery of compute access time. Thresholds apply at 2029-06-30, one year before the resolution date." } ], "milestones": [ { "id": "m1", "date": "2027-06-30", "statement": "EuroHPC announces award decisions for at least four AI gigafactory consortia.", "p": 0.55 }, { "id": "m2", "date": "2028-12-31", "statement": "At least one selected AI gigafactory is declared operational by EuroHPC or the Commission.", "p": 0.35 }, { "id": "m3", "date": "2029-12-31", "statement": "Fewer than three selected AI gigafactories are operational at year-end 2029.", "p": 0.7 } ], "falsifier": "EuroHPC or the Commission declares three or more selected AI gigafactories operational on or before 2030-06-30.", "whyItMatters": "The gigafactories are Europe's answer to depending on US and Gulf compute, and the largest sovereign AI programme with published budget figures. If fewer than three are running two years after the target date, the sovereign compute model (public anchor customer, private balance sheet, national grid connection) will have failed its first real test, and European labs and enterprises will keep renting American capacity. The constraint is power and permitting, not money or intent.", "whatWouldRaise": [ "Award decisions slipping past mid-2027 or fewer than four consortia selected in the first phase.", "Selected sites without a confirmed high-voltage grid connection date in their public announcements.", "Private co-investment commitments below the EUR 20 billion the Commission expects." ], "whatWouldCut": [ "Two or more consortia announcing site construction start with accelerator supply agreements before 2027-12-31.", "Selected sites reusing existing powered data centre shells rather than greenfield builds.", "A member state fast-tracking permits and grid access for a gigafactory as critical infrastructure." ], "namedEntities": [ "EuroHPC Joint Undertaking", "European Commission" ], "marketCall": false, "recusalCheck": { "passed": true, "keysChecked": [ "employer-leasing", "named-competitor-bookings", "customer-pipeline", "ticker" ] }, "signedBy": "Brian Letort", "registeredAt": "2026-09-08", "amendments": [], "ruleHash": "3d11841cf59501c97f4f6a5fbfc13c88532e74cda0a36b34f86189b4812ba19a", "distance": 1.0986122886681098, "distanceLabel": "clearly above consensus", "engine": { "id": "RT-32", "prior": null, "council": { "p": 0.5100448674632688, "votes": [ { "persona": "grid-and-power-planner", "model": "openai/gpt-5.6-sol", "p": 0.53, "rationale": "Reference class: first-of-kind European HPC procurements and large-load campuses, where award-to-service schedules commonly slip one to three years. The cutoff leaves roughly three years after anticipated 2027 awards, but financing, permits, grid connection, transformers, cooling, and accelerator delivery are serial gates with correlated delay risk. That favors fewer than three. Conversely, 76 preliminary expressions, large public support, and selection of advanced consortia make cancellation or only two awards unlikely. Most importantly, inauguration or initial framework compute access counts, not full build-out; staged commissioning could let three sites qualify. With no contract or operational indicators yet, the thesis is slightly above even odds." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.52, "rationale": "Reference class: first-of-kind European HPC procurements, where contract award, site readiness, power connection and acceptance commonly produce multi-quarter to multi-year slippage. Awards expected in 2027 leave roughly three years to the strict 2030-06-30 cutoff. Reaching three operational sites is conjunctive: several awards must close, financing and grid works must proceed, accelerators must arrive, and EuroHPC must publish qualifying language. These risks are correlated across sites. Conversely, 76 expressions of interest, substantial public funding and the permissive inclusion of inauguration or initial compute access make three declarations achievable even before full-scale operation. With no contract or operational indicator data, I place the fewer-than-three outcome near even odds, slightly favoring a hit." }, { "persona": "enterprise-cio", "model": "openai/gpt-5.6-sol", "p": 0.56, "rationale": "Reference class: EU supercomputer procurements and grid-intensive public infrastructure. From planned early-2027 awards to the cutoff is about 39 months. I use roughly a 0.20 annual post-award operational hazard, compounded to about 0.52 per site, with strong shared schedule risk. Four-to-seven awards create portfolio redundancy, while “inaugurated” or initial-access language is easier than full production readiness. Still, financing closure, grid connection, accelerator procurement, security review, and consortium contracting can each consume years; “up to seven” is not seven funded projects. The supplied indicators contain no contract or operational data. Correlated megaproject delays make fewer than three slightly more likely than not." }, { "persona": "credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.44, "rationale": "My reference class is European public HPC procurements and multi-site energy-intensive infrastructure programs, which often suffer correlated delays from permitting, grid connections, financing, and technology redesign. However, the threshold is only three facilities from up to seven, the operational definition includes inauguration or initial compute access, and the cutoff provides roughly three years after expected awards—substantial recovery time beyond the planned 18-month delivery. Portfolio redundancy and up to EUR 10 billion of public funding make three declarations more likely than any individual site’s schedule suggests. With no contract or construction indicators yet, I assign modestly below-even odds to fewer than three." }, { "persona": "geopolitics-policy-analyst", "model": "openai/gpt-5.6-sol", "p": 0.43, "rationale": "Reference class: EU-backed first-of-kind compute and energy-intensive infrastructure procurements, where award and construction schedules commonly slip by one to three years. The thesis benefits from correlated grid, permitting, financing, accelerator-supply, and framework-contract delays, plus the strict requirement for an official operational declaration. However, the announced 2028 operating target leaves roughly 18–24 months of schedule buffer, and a seven-site portfolio needs only three successes. EUR 10 billion of public support and 76 preliminary expressions of interest also reduce cancellation and award-shortfall risk. Portfolio redundancy therefore makes three operational declarations modestly more likely than not by the cutoff." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.58, "rationale": "Base-rate prior: about 0.60 that fewer than three sites in a new, multi-site European public compute-infrastructure programme become operational and officially documented within roughly 3.5 years of expected awards. The plan has substantial funding and 76 preliminary expressions of interest, lowering the failure probability. However, reaching three requires multiple awards, financing, grid connections, accelerator delivery, construction, framework-contract readiness, and timely EuroHPC/Commission disclosure by a hard date. Correlated permitting and supply delays materially weaken the official 18-month schedule. Existing AI Factories are excluded, and no contract or operational indicator data are supplied. I therefore place the thesis slightly below its reference-class prior but above the forecaster’s 0.5." } ], "aggregation": "log-odds-mean", "extremization": 1, "dissent": null, "spread": 0.15, "warnings": [ "all 6 votes came from one vendor (openai); extremization disabled, persona diversity only" ] }, "final": { "p": 0.5100448674632688, "ci80": [ 0.4231598657917389, 0.5963272037836719 ], "method": "council-only:log-odds-mean", "weights": { "council": 1 } }, "stated": 0.5, "history": [ { "asOf": "2026-09-07", "p": 0.5354823694107796, "source": "radar-engine", "runId": "radar-2026-09-07T23-12-09Z-06a58f" }, { "asOf": "2026-09-07", "p": 0.5100448674632688, "source": "radar-engine", "runId": "radar-2026-09-07T23-49-04Z-de8abf" } ], "indicatorStatus": "no-data", "milestones": [ { "id": "m1", "enginePrior": 0.5, "councilP": null }, { "id": "m2", "enginePrior": null, "councilP": null }, { "id": "m3", "enginePrior": null, "councilP": null } ], "arena": "geopolitics-policy", "status": "draft", "tier": "T3" }, "indicators": [ { "thesisId": "RT-32", "indicatorId": "li-1", "connector": "manual", "seriesId": "eurohpc-aigf-contracts-signed", "unit": "count", "asOf": "2026-09-07", "current": null, "status": "no-data", "history": [], "url": "https://www.eurohpc-ju.europa.eu/news_en", "note": "manual: 'eurohpc-aigf-contracts-signed' not in manual-series.json" }, { "thesisId": "RT-32", "indicatorId": "li-2", "connector": "manual", "seriesId": "eurohpc-aigf-operational", "unit": "count", "asOf": "2026-09-07", "current": null, "status": "no-data", "history": [], "url": "https://www.eurohpc-ju.europa.eu/news_en", "note": "manual: 'eurohpc-aigf-operational' not in manual-series.json" } ], "indicatorStatus": "no-data", "chainEvents": [ { "seq": 254, "ts": "2026-09-07", "kind": "repriced", "hash": "97f22962b9bbd48622c558a43dc024af758da9d6d9f8ca3632ce6e0f776cd5e2" }, { "seq": 255, "ts": "2026-09-07", "kind": "repriced", "hash": "075be7b78944c900e4353841d5948182466b6d4abfe2f15cc93f7fdf37da92cb" }, { "seq": 286, "ts": "2026-09-08", "kind": "registered", "hash": "0d62da6058e6ecb5d29d3450441f88931d4b6d246923c759b33458128ffe2bbb" } ], "url": "/predictions/geopolitics-policy/RT-32" }, { "id": "RT-33", "title": "10,000 humanoids in paid deployment by 2031", "arena": "agi-capabilities", "tier": "T3", "tag": "CALL", "direction": "up", "status": "registered", "statement": "By 2031-12-31 an SEC-registered company discloses in a filing or earnings release that at least 10,000 humanoid or general-purpose mobile-manipulator robots are deployed under paid contracts at customers other than the robot maker.", "resolutionRule": "Resolves HIT if, on or before 2031-12-31, a company with securities registered with the SEC states in a 10-K, 10-Q, 8-K, 20-F, or an earnings release furnished as an 8-K exhibit that at least 10,000 humanoid or wheeled mobile-manipulator robots (general-purpose robots with manipulator arms; fixed industrial arms, autonomous mobile carts, and robotaxis do not count) are deployed under paid contracts at customers other than the maker or its affiliates. The disclosing company may be the maker (e.g. Tesla) or the customer (e.g. Amazon disclosing its own paid fleet of a third party's robots). Cumulative units produced or shipped do not count unless the filing states they are deployed at paying customers. Verify via SEC EDGAR full-text search. If EDGAR full-text search is discontinued, the filings themselves control. Baseline 2026-09-07: the largest disclosed third-party humanoid fleet is in the tens of units (Agility Robotics at Toyota and GXO).", "resolutionSource": { "name": "SEC EDGAR full-text search", "url": "https://efts.sec.gov/LATEST/search-index?q=%22humanoid%22" }, "ambiguity": "resolves-against-forecaster", "horizon": "2031-12-31", "p": 0.3, "ci80": [ 0.18, 0.45 ], "consensus": { "value": "Manifold's \"over one hundred thousand humanoid robots deployed in the real world by 2030\" market trades near 70%; Metaculus prices Tesla producing 1,000 Optimus units by end-2027 at roughly one in three; Hyundai has announced a 30,000 robots-per-year Boston Dynamics factory and Tesla has stated Optimus volume production targets.", "impliedP": 0.55, "impliedBy": "If 100,000 humanoids are deployed anywhere by 2030 at ~70% (Manifold), then a single SEC registrant disclosing 10,000 in paid third-party deployment a year later is roughly 80% likely conditional on that world and near zero otherwise, giving ~0.55 on the consensus path.", "source": "Manifold, \"2030 - Over one hundred thousand humanoid robots will be deployed in the real world\"; Metaculus, \"Will Tesla be able to mass produce humanoid robots by the end of 2027?\"", "url": "https://manifold.markets/HenriThunberg/2030-3-over-one-hundred-thousand-hu", "asOf": "2026-09-07", "note": "Rung 5 of the capability ladder. My 0.30 is half the implied odds because the largest disclosed paid fleet is under 100 units, the most likely 10,000-unit deployers (Tesla internal, Hyundai plants, Chinese OEMs) are either self-deployments, non-SEC registrants, or both, and the disclosure has to say \"paid\" and \"at customers\", which no filing has yet done at any scale." }, "leadingIndicators": [ { "id": "li-1", "label": "Largest disclosed third-party humanoid fleet (SEC filings)", "connector": "manual", "seriesId": "largest-disclosed-humanoid-fleet-units", "unit": "robots", "onTrack": { "op": ">=", "value": 500 }, "offTrack": { "op": "<", "value": 100 }, "url": "https://efts.sec.gov/LATEST/search-index?q=%22humanoid%22", "note": "Largest number of humanoid or mobile-manipulator robots any SEC registrant states are deployed at paying customers. Thresholds are for 2027-12-31." }, { "id": "li-2", "label": "Metaculus community, Tesla 1,000 humanoids by end-2027", "connector": "metaculus", "seriesId": "19879", "unit": "probability", "onTrack": { "op": ">=", "value": 0.6 }, "offTrack": { "op": "<", "value": 0.25 }, "url": "https://www.metaculus.com/questions/19879/", "note": "Community prediction on Metaculus question 19879. Thresholds are for 2027-06-30; mass production is a precondition for a paid fleet." }, { "id": "li-3", "label": "BLS employment, warehousing and storage", "connector": "bls", "seriesId": "CES4349300001", "unit": "thousands of employees", "onTrack": { "op": "<", "value": 1800 }, "offTrack": { "op": ">=", "value": 1950 }, "url": "https://data.bls.gov/timeseries/CES4349300001", "note": "Seasonally adjusted all-employees series for NAICS 4931. Falling warehouse headcount alongside rising volumes is where a 10,000-robot fleet would first show. Thresholds are for the December 2028 reading." } ], "milestones": [ { "id": "m1", "date": "2027-12-31", "statement": "Tesla is judged to have mass-produced 1,000 Optimus units (Metaculus 19879 resolves yes).", "p": 0.35 }, { "id": "m2", "date": "2028-12-31", "statement": "An SEC registrant discloses at least 1,000 humanoid or mobile-manipulator robots deployed at paying third-party customers.", "p": 0.45 }, { "id": "m3", "date": "2030-12-31", "statement": "An SEC registrant discloses at least 5,000 such robots deployed at paying third-party customers.", "p": 0.35 }, { "id": "m4", "date": "2031-12-31", "statement": "An SEC registrant discloses at least 10,000 such robots deployed at paying third-party customers.", "p": 0.3 } ], "falsifier": "On 2028-12-31 no SEC registrant has disclosed 1,000 or more humanoid or mobile-manipulator robots deployed at paying third-party customers; the thesis is then withdrawn as falsified regardless of the 2031 horizon.", "whyItMatters": "A 10,000-unit paid fleet is the smallest number at which physical-task automation stops being a pilot and starts showing up in warehouse and plant headcount. It is the top rung of the ladder and the one furthest from today's evidence. Reading it from filings rather than demo videos keeps the question honest.", "whatWouldRaise": [ "Tesla, Hyundai, or Figure disclose a paid third-party fleet above 1,000 units in 2027.", "A logistics or automotive customer discloses robot fleet size in its own filings.", "Unit costs disclosed below $50,000 with multi-year service contracts." ], "whatWouldCut": [ "Humanoid deployments remain internal to the makers' own factories through 2028.", "A public safety incident triggers OSHA restrictions on humanoids in shared workspaces.", "Makers report fleets in \"units produced\" only, with no paid-deployment disclosure." ], "namedEntities": [ "Tesla", "Agility Robotics", "Toyota", "GXO", "Hyundai", "Figure", "Amazon", "SEC" ], "marketCall": false, "recusalCheck": { "passed": true, "keysChecked": [ "employer-leasing", "named-competitor-bookings", "customer-pipeline", "ticker" ] }, "signedBy": "Brian Letort", "registeredAt": "2026-09-08", "amendments": [], "ruleHash": "09317202cee1aeaee78741e086094e80402a508f8abfcf30721db60ccd28897f", "distance": -1.047968555849355, "distanceLabel": "clearly below consensus", "engine": { "id": "RT-33", "prior": { "p": 0.16718936350025437, "method": "drift-gbm-terminal:li-3", "note": "P(indicator li-3 satisfies onTrack < 1800.0 at 2031-12-31); drift +0.000527/yr, vol 0.0105/sqrt(yr) from 32 points; proxy for the thesis, not its rule" }, "council": { "p": 0.1424123858674903, "votes": [ { "persona": "grid-and-power-planner", "model": "openai/gpt-5.6-sol", "p": 0.11, "rationale": "Using the reference class of newly commercialized warehouse and industrial-automation platforms, scaling from pilot fleets to 10,000 customer-deployed units within five years is uncommon. The binding hurdle is conjunctive: an SEC registrant must disclose 1,000 paid third-party deployments by 2028 or the thesis is falsified, then reach 10,000 by 2031, with qualifying wording in an enumerated filing. The supplied baseline remains only tens of units, while production targets and internal deployments do not qualify. The BLS series is not probative. My estimate is therefore well below the forecaster’s 0.30 and broad-market forecasts that use looser definitions of deployment." }, { "persona": "semiconductor-supply-chain-analyst", "model": "openai/gpt-5.6-sol", "p": 0.16, "rationale": "Using the reference class of novel industrial-automation platforms scaling from customer pilots to five-digit paid fleets, I estimate a 0.30 chance that an SEC registrant explicitly discloses 1,000 qualifying third-party deployments by the binding 2028 falsifier. Conditional on that unusually rapid step, I assign about 0.53 to reaching and disclosing 10,000 by 2031, yielding 0.16 jointly. Production targets are weak evidence because internal deployments, affiliates, shipments, carts, and non-SEC manufacturers do not qualify. The baseline is only tens of units, while the BLS series is not a direct adoption indicator. Strict wording and disclosure risk put this below the forecaster’s 0.30." }, { "persona": "enterprise-cio", "model": "openai/gpt-5.6-sol", "p": 0.14, "rationale": "The controlling constraint is the mandatory 2028 gate, not 2031 manufacturing capacity. Using enterprise industrial-robot rollouts as the reference class, I assign about 0.28 to any SEC registrant moving from today’s tens of paid third-party units to a filing-confirmed 1,000 by end-2028. Pilots, safety validation, procurement, and site integration usually consume several annual cycles. Conditional on clearing that gate, I assign about 0.50 to a tenfold paid-customer ramp and qualifying disclosure by 2031; compounding gives 0.14. Announced capacity, production, internal deployment, shipments, and Chinese deployments do not satisfy the rule. This is below 0.30 because exact SEC disclosure and third-party payment are additional conjunctive requirements." }, { "persona": "credit-analyst", "model": "openai/gpt-5.6-sol", "p": 0.14, "rationale": "I use the reference class of complex industrial hardware moving from pilot fleets to at least 10,000 paid customer deployments within roughly five years. The binding condition is conjunctive: an SEC registrant must disclose at least 1,000 qualifying external paid deployments by 2028, then reach 10,000 by 2031. From a baseline in the tens, I estimate 0.25 for the 2028 gate and 0.55 for reaching and explicitly disclosing 10,000 conditional on clearing it, yielding about 0.14. Announced production capacity and maker-affiliate deployments do not qualify. The BLS series is weakly relevant. The forecaster’s 0.30 underweights the early falsifier and disclosure specificity." }, { "persona": "geopolitics-policy-analyst", "model": "openai/gpt-5.6-sol", "p": 0.15, "rationale": "The binding hurdle is the falsifier: a qualifying SEC disclosure must jump from tens to 1,000 paid third-party deployments by end-2028, not merely production or internal use. Using first-generation industrial-automation fleet ramps as the reference class, I assign 0.25 to that milestone—roughly a 0.12 annual hazard. Conditional on clearing it, I assign 0.60 to a tenfold expansion with qualifying disclosure by 2031, yielding 0.15. Multiple vendors help, but internal fleets, non-SEC manufacturers, pilots, shipments, and vague deployment claims do not qualify. The stated 0.30 appears to overweight broader deployment forecasts that lack the paid-customer, SEC-disclosure, and 2028-knockout conjunction." }, { "persona": "superforecaster-statistician", "model": "openai/gpt-5.6-sol", "p": 0.16, "rationale": "I use a 0.15 base rate for pre-scale complex-robot platforms progressing from tens to 10,000 paid third-party deployments, with qualifying public-company disclosure, within roughly five years. Announced factory capacity modestly raises the technology-side outlook, but production is not deployment. The claim also requires a qualifying SEC registrant, paying unaffiliated customers, explicit deployed-unit language, and timely filing disclosure. Most highlighted pathways involve excluded internal deployments or uncertain disclosure. The mandatory 1,000-unit disclosure by 2028 is an especially difficult conjunctive gate. The BLS series is not informative here, and no relevant live fleet indicator or calibration table was supplied. Ambiguity and vendor-promise risk keep me near the base rate." } ], "aggregation": "log-odds-mean", "extremization": 1, "dissent": null, "spread": 0.05, "warnings": [ "all 6 votes came from one vendor (openai); extremization disabled, persona diversity only" ] }, "final": { "p": 0.15439503972705432, "ci80": [ 0.04659225522527269, 0.4727983084303339 ], "method": "stacked-fixed-weights (radar: prior 0.5, council 0.5; no resolved radar items to learn from)", "weights": { "council": 0.5, "prior": 0.5 } }, "stated": 0.3, "history": [ { "asOf": "2026-09-07", "p": 0.15187745065123454, "source": "radar-engine", "runId": "radar-2026-09-07T23-12-09Z-06a58f" }, { "asOf": "2026-09-07", "p": 0.15439503972705432, "source": "radar-engine", "runId": "radar-2026-09-07T23-49-04Z-de8abf" } ], "indicatorStatus": "mixed", "milestones": [ { "id": "m1", "enginePrior": 0.5, "councilP": null }, { "id": "m2", "enginePrior": null, "councilP": null }, { "id": "m3", "enginePrior": null, "councilP": null }, { "id": "m4", "enginePrior": null, "councilP": null } ], "arena": "agi-capabilities", "status": "draft", "tier": "T3" }, "indicators": [ { "thesisId": "RT-33", "indicatorId": "li-1", "connector": "manual", "seriesId": "largest-disclosed-humanoid-fleet-units", "unit": "robots", "asOf": "2026-09-07", "current": null, "status": "no-data", "history": [], "url": "https://efts.sec.gov/LATEST/search-index?q=%22humanoid%22", "note": "manual: 'largest-disclosed-humanoid-fleet-units' not in manual-series.json" }, { "thesisId": "RT-33", "indicatorId": "li-2", "connector": "metaculus", "seriesId": "19879", "unit": "probability", "asOf": "2026-09-07", "current": null, "status": "no-data", "history": [], "url": "https://www.metaculus.com/questions/19879/", "note": "METACULUS_TOKEN not set; the Metaculus API rejects anonymous requests (HTTP 403)" }, { "thesisId": "RT-33", "indicatorId": "li-3", "connector": "bls", "seriesId": "CES4349300001", "unit": "thousands of employees", "asOf": "2026-08-31", "current": 1837.4, "status": "between", "history": [ { "asOf": "2024-01-31", "value": 1834.9 }, { "asOf": "2024-02-29", "value": 1838.4 }, { "asOf": "2024-03-31", "value": 1845.7 }, { "asOf": "2024-04-30", "value": 1853.1 }, { "asOf": "2024-05-31", "value": 1860.4 }, { "asOf": "2024-06-30", "value": 1865.3 }, { "asOf": "2024-07-31", "value": 1871.1 }, { "asOf": "2024-08-31", "value": 1870.7 }, { "asOf": "2024-09-30", "value": 1871 }, { "asOf": "2024-10-31", "value": 1873.1 }, { "asOf": "2024-11-30", "value": 1877.6 }, { "asOf": "2024-12-31", "value": 1881.2 }, { "asOf": "2025-01-31", "value": 1881.2 }, { "asOf": "2025-02-28", "value": 1883.4 }, { "asOf": "2025-03-31", "value": 1880.8 }, { "asOf": "2025-04-30", "value": 1881.3 }, { "asOf": "2025-05-31", "value": 1875.3 }, { "asOf": "2025-06-30", "value": 1871.9 }, { "asOf": "2025-07-31", "value": 1869.3 }, { "asOf": "2025-08-31", "value": 1869.4 }, { "asOf": "2025-09-30", "value": 1853.4 }, { "asOf": "2025-10-31", "value": 1839.9 }, { "asOf": "2025-11-30", "value": 1842 }, { "asOf": "2025-12-31", "value": 1836.2 }, { "asOf": "2026-01-31", "value": 1830.4 }, { "asOf": "2026-02-28", "value": 1831.1 }, { "asOf": "2026-03-31", "value": 1832.5 }, { "asOf": "2026-04-30", "value": 1837.4 }, { "asOf": "2026-05-31", "value": 1841.7 }, { "asOf": "2026-06-30", "value": 1845.8 }, { "asOf": "2026-07-31", "value": 1840 }, { "asOf": "2026-08-31", "value": 1837.4 } ], "url": "https://data.bls.gov/timeseries/CES4349300001", "note": "BLS v1, last 10 years" } ], "indicatorStatus": "mixed", "chainEvents": [ { "seq": 256, "ts": "2026-09-07", "kind": "repriced", "hash": "70eecda4e98599b0fa4c829da77ba05bc54b3f5d2765c8d00c942dcdbcfde8c0" }, { "seq": 257, "ts": "2026-09-07", "kind": "repriced", "hash": "b883e67c9eed3e46c1bb9c2286ee0afa00cc4b02ad3511880b038560014bd842" }, { "seq": 287, "ts": "2026-09-08", "kind": "registered", "hash": "c6428fe1b8958e521ce22e192c4ae8050d6c43d62f3cf1d6e4582e48336c5c75" } ], "url": "/predictions/agi-capabilities/RT-33" } ] }