| # Current v3 scope |
|
|
| This is a reproducible finite-light CPU research reference for Certified Innovation Rendering. |
| It adds conservative validity domains to stored visibility facts and eliminates their residual |
| sampling support. It contains six new self-contained mathematical arguments and two new |
| held-out protocols. The broader standalone architecture and v1/v2 evidence remain available. |
|
|
| **Proved under stated assumptions:** matrix covariance contraction for exact partial evidence; |
| causal sequential estimation; real-arithmetic sphere/segment validity; finite-domain query |
| count; deterministic direct-light enclosure; a counterexample to universal variance dominance |
| for arbitrary predictions. None is labeled a historically first theorem. |
|
|
| **Executed:** twenty new procedural scenes; 58 new and retained tests; motion, spatial and |
| known-time query studies; raw results; a portable CPU API; one retained trained visibility prior. |
|
|
| **Estimated:** 68% broader maturity, per-area status and all GPU engineering budgets. These |
| are subjective or derived quantities, not measured production capability. |
|
|
| **Not established:** full SR/RR/FG, game integration, end-to-end speedup, full transport, |
| formal floating-point certificates, hidden-event recovery without evidence, optimal future |
| sampling, cross-game generalization, novel fundamental probability theory or independent replication. |
|
|
| Visibility caching and kinetic certificates are prior art. The contribution candidate is their |
| explicit common-state, validity and output-estimation interface. Current root metadata and |
| results_v3 govern the release; history/v2 and legacy documents retain historical claims. |
| |
