Measured harness ledgerPublic result
DeepSeek V4 ProSpace Flight Stage 2 — Landfall — DeepSeek V4 Pro Max
Extend the Stage 1 space-flight project with a seamless orbit-to-surface round trip, explorable terrain, takeoff, landing, and a fixed character asset using one frozen follow-on request.
Max reasoningHeadline result
- Workflow cost
- $0.50
- Wall-clock
- 2h 17m 5.4s wall-clock
- Processed tokens
- 55.76M processed
- Record state
- partial_generation_metrics_build_and_failed_runtime_acceptance_ledger
Public summary
DeepSeek V4 Pro Max partial_generation_metrics_build_and_failed_runtime_acceptance_ledger ledger: 2h 17m 5.4s wall-clock, 55.76M processed, and $0.50 API-equivalent usage accounting, not an itemized subscription cash charge.
Run identity and stack
- Result ID: space-flight-game-threejs-stage-2-landfall-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 Stage 2 workflow
- Stack: Harness v1 Landfall prompt
Primary artifact integrity
- Kind: deterministic-fresh-locked-production-build
- Path: artifacts/space-flight-game-threejs-stage-2-landfall-deepseek-v4-pro-max/production-build.tar.gz
- SHA-256: 245264deda9f25078178f8c66da5e97140b4aa11096c389a572e32f078f0eda6
Validation evidence
- Result: FAIL_ACCEPTANCE
- Path: artifacts/space-flight-game-threejs-stage-2-landfall-deepseek-v4-pro-max/validation/static-acceptance-findings.public.json
- SHA-256: 9b07279737890fd628890d361f22b814f70b7f766ba35af87124602bf3421065
Recorded caveats
- The Stage 2 metric has a disclosed three-turn protocol deviation: one benchmark turn and two exact 'continue' rescues.
- Wall-clock is end-to-end workflow latency including tools, installs, tests, and idle gaps; it is not model-only compute.
- The independent runtime probe uses disclosed operator staging and therefore does not prove continuous manual descent or a fresh-player round trip.
- The probe stopped after terminal rendering, HUD, and camera failures; boarding, ascent, a second landing, and indefinite repeatability remain unproven.
- The initial view is near-black, but the same exact compositor renders staged Verdant; this record does not attribute the Stage 2 failure universally to bloom.
- Runtime telemetry measured only the outer character holder's radial alignment. The rendered mesh orientation remains unverified because the failed camera never provided a usable character view.
- The model-authored eight-cycle soak used debug hooks and direct staging, and its visual sanity call bypassed EffectComposer; it is retained only as qualified state-machine smoke context.
- No frozen fixture manifest or redistribution-license record was supplied for the original infantry asset; the original is hash-bound but not duplicated.
- No blind evaluation or standardized RTX Pro 6000 1080p60 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.