Measured harness ledgerPublic result
DeepSeek V4 ProExplorable space-flight game — DeepSeek V4 Pro Max
Build a responsive browser space-flight game with flight controls, a coherent star-system environment, lighting, assets, and a playable game loop.
Max reasoningHeadline result
- Workflow cost
- $0.44
- Wall-clock
- 1h 26m 14.4s wall-clock
- Processed tokens
- 47.73M processed
- Record state
- partial_token_timing_artifact_and_runtime_ledger
Public summary
DeepSeek V4 Pro Max partial_token_timing_artifact_and_runtime_ledger ledger: 1h 26m 14.4s wall-clock, 47.73M processed, and $0.44 API-equivalent usage accounting, not an itemized subscription cash charge.
Run identity and stack
- Result ID: space-flight-game-deepseek-v4-pro-max
- Technical model: DeepSeek V4 Pro
- Provider: OpenCode Go
- Client: OpenCode
- Stack: OpenCode Go
- Stack: OpenCode
- Stack: Technical model/configuration: DeepSeek V4 Pro
- Stack: Three.js / Vite game
- Stack: Generated spaceship assets
- Stack: Harness v1 space-flight prompt
Primary artifact integrity
- Kind: manifest-verified-artifact
- Path: artifacts/space-flight-game-deepseek-v4-pro-max/source/src/main.js
- SHA-256: 468e26b3455a5fbb5b68b0a62f0ef325f5f67d6f559d2629b70e49a07241cfeb
Validation evidence
- Result: FAIL_VISUAL_VALIDITY
Recorded caveats
- Wall-clock is end-to-end workflow latency including tool execution, tests, browser checks, and idle gaps; it is not model-only compute.
- OpenCode splits one agentic benchmark turn into many assistant/API messages.
- Four extractor-counted tool errors are retained in the ledger; the completed build and independent runtime checks passed.
- The 120.1279232461955 FPS figure is a performance-only observation from a visually invalid native rendering path and is not a passing runtime score.
- The exact final production build has a blocking native Metal visual defect: its default UnrealBloomPass path leaves the WebGL scene black while DOM overlays remain visible.
- A no-MSAA isolation remains black, while a no-bloom isolation restores the scene; this identifies bloom rather than MSAA as the trigger in the tested environment.
- Forced-SwiftShader 7-10 FPS diagnostics are software-renderer observations and are not labelled as hardware performance.
- The public record excludes private OpenCode session/export data; source hashes and the stable benchmark-window hash are retained.
- No blind visual-quality evaluation or standardized target-hardware capture is included.
Public result only
This result keeps its public summary and evidence, but it does not currently have a matching Builder test with prompts, projects, Harness workflows or skills.