brianletort.ai
Qwen3.8 vs GPT-5.2

Playable game: The Last Lighthouse

A side-by-side artifact view of a cinematic browser game prompt. This is a showcase comparison, not a controlled benchmark: Qwen is preserved from the public demo regeneration flow, while GPT-5.2 is a new unrepaired counterpart call.

Showcase, asymmetric evidenceeditorial showcase with known asymmetry

Publication-safe copies generated from preserved showcase artifacts. Source research paths, local absolute paths, endpoints, and secrets are intentionally omitted.

Prompt and source disclosure
You are generating a single, self-contained HTML file. Output ONLY the HTML.

Create "The Last Lighthouse": a dark, cinematic, three-act choice-driven narrative with at least 3 distinct endings.
Requirements:
- Must be fully self-contained: no network, no external assets, no storage (no localStorage/sessionStorage/indexedDB/cookies)
- No eval, no fetch/XMLHttpRequest/WebSocket, no script src, and NO links to http(s)
- CRITICAL: do not include the substring "http" or "www" anywhere in the HTML (this includes SVG namespaces like w3.org); avoid SVG/data-URI patterns that embed "http"
- Deterministic: do not use Math.random(); outcomes depend only on choices
- UI: Start button, choices that advance acts, restart option after ending
- Acts: I, II, III. Each act has a few beats of text; choices advance to next act and then to an ending
- At least 3 endings that differ materially (not just reworded)
- Optional audio is allowed ONLY if it fails gracefully (no crashes if AudioContext blocked)

Testing hooks (REQUIRED):
- A Start button with data-testid="story-start" (visible on first load; clicking it begins the story)
- Choice buttons with data-testid="story-choice" (every visible choice button must include this exact attribute value)
- An act indicator element with data-testid="story-act" whose text is exactly one of:
  - "Act I"
  - "Act II"
  - "Act III"
  - and after the final choice it MUST include the word "Ending" or "Epilogue"
- A log container with data-testid="story-log" that appends entries; each entry must have class "entry"
- A restart button with data-testid="story-restart" that returns the story to Act I (fresh state)

Mandatory UI skeleton (you may style it, but keep these exact data-testid values):
- <button data-testid="story-start" ...>Start</button>
- <div data-testid="story-act">Act I</div> (update text exactly as acts change)
- <div data-testid="story-log"></div> (append <div class="entry">...</div>)
- For each choice: <button data-testid="story-choice" ...>...</button>
- After an ending is shown: <button data-testid="story-restart" ...>Restart</button>

Output only a complete HTML document starting with <!DOCTYPE html> and including closing </body></html> tags.

Prompt SHA-256

b26085287f4bb9c7c8f6396f6473f7d419d6177241e3a33ce55a5a1ddcd67804

Claims

  • Qwen artifact is preserved from the first-blog public demo regeneration flow.
  • Qwen artifact was regenerated/iterated and acceptance-tested before publishing; it is not a literal one-shot controlled result.
  • GPT-5.2 receives the recovered user prompt, temperature, and completion budget once; output is retained without repair.