TEST 01 · GAME DEV · GAUNTLET V1
Create a game
Measures creative coding, game logic, atmosphere.
Build a playable haunted-house horror game as a single self-contained HTML file (no external assets, libraries, or network requests). The player explores a dark mansion, something is hunting them, and there is a way to win and a way to die. Prioritize atmosphere: darkness, lighting, sound (Web Audio API), and tension. Include brief on-screen instructions. It must work by simply opening the file in a browser.
Live artifact. Exactly what the model produced, sandboxed, no network.
| runs first try | 2 |
| polish | 2 |
| prompt adherence | 2 |
| wow factor | 2 |
| total | 8 |
AI panel: 10/10 — median of 5 blind judges.
First-person raycaster mansion crawl built from scratch: flashlight falloff, film grain, Web Audio ambience, pathfinding hunter, three seals, win/death paths. Very dark by design. 3m 32s at high effort (Ruling #3). Arena panel gave this match to Fable 2-1; the human judge saw it the other way.
4m. 9.9k output tokens.
Live artifact. Exactly what the model produced, sandboxed, no network.
| runs first try | 1 |
| polish | 1 |
| prompt adherence | 1 |
| wow factor | 1 |
| total | 4 |
AI panel: 9/10 — median of 5 blind judges.
Judge's call: the game didn't work in his play-through — scored 1 on every dimension. Top-down torch-and-sigils stealth design (torch/battery meters, multi-state hunter, six sigils, layered Web Audio) that the arena panel preferred 2-1 from the code and artifacts, making this the season's starkest human-vs-panel split. Run counted under Ruling #4 (buildos MCP enabled; model read the producer rules note; stats include its post-game producer-work detour).
20m. 183k output tokens.