Clearance index: integration validation

PR #92 now contains only the clearance-index/native-cache optimization and library-provenance changes on main c57077df. Its already merged #88 stack was removed by replaying those two commits; the engine changes applied without manual conflict resolution. No validation rule or engine default was relaxed.

Verified campaign receipts

The existing GCP summaries were fetched again on 2026-10-07:

  • Base: 20261007-ladder-c15db5, bundle commit 2215cc8f2091e7ea705fdc2cb98b0bb30fdc6abb, whose parent is #88 head 540784c7d0a16b2aa80b0ddb9790a1b84832132d.

  • New: 20261007-ladder-d3b3dd, bundle commit b2c92f52ce4a1c5cea715612fab655701f0aa427, whose parent is optimization commit c6c1765d51e56dc0e8719a51b58d8d6110ea7b48.

  • Each bundle-only commit adds the same prebuilt Linux library; neither commit was published. Both campaigns completed, with all 96 cell receipts present in each arm.

  • All 96 pairs agree on verdict, opens, DRC violations, unconnected items, vias and copper hash. There are 94 nonempty copper hashes, and 85/96 cells pass in each arm. The other eleven cells remain failures in both arms.

  • Summed task wall time is 2369 → 2041 seconds (-14%); summed place-route time is 1834 → 1467 seconds (-20%).

  • Every native provenance receipt names libpnr_maze.so, source prebuilt, SHA-256 d3f7b82a5fcd0d2139a848b14e59844fd1df99625597f5b8e41aeb06d8abd48a.

These are the original pre-integration campaigns, not a new A/B of the rebased head. They do not include the later dovetail rungs or main’s newer changes. No new paid run was started.

Animation applicability

This optimization has no intended visual behavior: the index skips distant obstacles before applying the same exact tests, and the native wrapper reuses buffers. The verified copper hashes are identical for every cell with copper. An animation would repeat the same board geometry and cannot demonstrate the measured runtime improvement; the receipt comparison is the appropriate evidence for this PR.

Integration checks

Nine focused Bazel targets passed: obstacle_index_test, dense_maze_native_test, exact_route_native_test, hard_rungs_test, tests/unit/exp:test_kinds, and the four release tests. Full CI still needs to pass on the pushed head.