Erik dbddad7a5e test #137: the seam shake reproduced offline — third mechanism characterized ...

The corridor gate FAILED with a changed symptom: shaking at cell seams
(+ purple floor flashing there) instead of the dead stop. Deep-probe
session (step-walk/push-back/indoor-bsp + full resolve capture) traced
the complete chain; docs/ISSUES.md #137 carries it. Short form:

- Corridor floors are double-faced portal slabs over under-rooms; the
  resting foot sphere lives within half a millimeter of three hit/straddle
  thresholds there.
- Crossing a boundary, the foot penetrates the neighbor ramp slab by
  ~0.4mm, steps up onto it successfully (+0.6mm lift, stepped=True) —
  and the lifted check position is then LOST: the following pass runs at
  the unlifted height (the P2 stale-snapshot class; retail step_up
  0x0050b6cc restores only on FAILURE).
- The unlifted re-test grazes the under-room's ceiling (the slab
  underside) within the near-miss window, dispatches a neg-poly step-up
  with a DOWNWARD normal, whose nested step-down finds no walkable at
  exact tangency -> StepUpSlide -> slide_sphere opposing branch ->
  reversed-movement collision -> Collided -> revert. Every frame = shake.

Apparatus committed:
- Issue137CorridorSeamReplayTests: 3 deterministic offline repros
  (snapshot-exact west-boundary from the capture, east deep-straddle,
  the clean-run pin), currently Skip='#137 seam shake' pending the fix.
  Key: THREE portal-ring hydration (the under-room 0x8A020166 is ring-3;
  with fewer rings the flood can't add it and the bug vanishes) + live
  Setup step heights (0.6/1.5) + probe-buffer capture for line-diffing
  offline vs live traces.
- Issue137CorridorSeamInspectionTests: portal-poly world spans (exposed
  the visual-vs-physics polygon-id conflation and the floor-portal
  topology), physics-BSP leaf membership walk, hit-normal candidate
  sweep (|align| both windings), downward-poly sweep.

NEXT: read TransitionalInsert's attempt loop against retail 0x0050b6f0,
find the restore that clobbers the successful step-up position, fix,
un-skip. The purple seam flashing is expected to be the render exposing
the same per-frame oscillation - re-check after the physics fix.

Suites: Core 2553 / App 713 / UI 425 / Net 385, 0 failures (5 skips =
2 pre-existing + the 3 parked repros).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

2026-07-05 19:05:19 +02:00

49 MiB

Raw Blame History

The file is too large to be shown.