Evidence & signatures

# Evidence
- Problem class: js-ggx-importance-sampling-split-sum-mis
- Model: openrouter/deepseek/deepseek-v4.1-flash
- Solved: 2026-10-02T16:30:26.445Z
- Verification: solution produced by pi in sandbox; see signatures.json

{"description": "Implement physically-based image-based lighting in Node: prefilter an equirectangular HDRI into a mip chain using GGX importance sampling (Hammersley sequence, alpha = roughness^2), generate the split-sum BRDF integration LUT over (NdotV, roughness), and evaluate the split-sum approximation against a brute-force Monte Carlo reference. Add multiple-importance sampling with balance/power heuristics so the direct+IBL estimator is unbiased with low variance, and demonstrate white-furnace energy conservation (a constant environment must integrate to the input radiance for every roughness and view angle). Handle non-power-of-two map dimensions with correct solid-angle weighting per texel.", "environment": "node20", "language": "js", "model": "openrouter/deepseek/deepseek-v4.1-flash", "problem_class": "js-ggx-importance-sampling-split-sum-mis", "provider": "openrouter", "solved_at": "2026-10-02T16:30:26.445Z", "version": "20"}