[Jpatching]

Summary

Retain an idle Skate Session after normal dismount, install fresh collision before remount, and discard cached state on faults/configuration changes/reset. Also add bounded ride observations, overlapping-floor handling and WBD dictionary inspection; unsupported box geometry refuses complete polygon supply.

normal dismount → suspended Session → fresh collision → remount
fault/reset → discard Session → cold preparation

Candidate: e0bc36cb4e3457dc6e65ffa0967ccc85409615bf. #61 owns current runtime evidence and next action; #20 owns ride acceptance.

Evidence

• Before: normal stop reconstructed the skater; cross-thread reset exposed stale pose. After, retained standalone native evidence: cold preparation 27,119.066 ms; remount 17.875 ms; reset reconstruction 27,485.899 ms. New-slope/fault/recovery regressions passed. These results do not establish staged GTA acceleration.
• Current source checks: clean e0bc36c passes python3 scripts/verify.py (118 tests, 75 documents, with loopback access) and python3 tools/gtaiv-skate/check.py. Earlier authored Windows checks and three candidate CI jobs passed; they were not rerun during cleanup.
• Retained worker observations: two mount/dismount upper bounds of 439.9/475.1 ms with verified recovery. Complete movement/camera/walking acceptance remains pending.
• Current runtime boundary: retained inventory passes but names board candidate 96aba30; saved-session preflight refuses process-identity before input. The owner selected leave runtime unchanged and finish cleanup only. No new clip, runtime installation, restart or gameplay acceptance.
• Supporting-floor selection, moving coverage, board/rider presentation and tricks remain unaccepted. Full pinned upstream host tests remain unverified because src/tests/map_startup.rs is missing.

Merge Danger

Door: two-way. Source is reversible; retained installations, private evidence and dirty worktrees remain preserved.

Blast Radius: gameplay. Keep draft pending relevant live evidence and explicit owner acceptance; no merge or release.

Prepared documentation cleanup

#63 awaits owner merge approval. Its integration is prepared separately at da0764b, with gameplay source unchanged: 114 repository tests, 30 tracked Markdown documents, authored adapter checks and independent Standards/Spec reviews pass. This preparation has not changed this PR's head or the installed runtime. After cleanup approval/merge, publish that integration here while preserving the dirty candidate checkout. #61 owns the current runtime mismatch and cleanup-only authority.

[Jpatching]

Feedback source update: 24ff699 on integration/gtaiv-skate-spec64 adds visible mount refusal and riding-exit reasons, separates coverage escape from the retained floor-relative height guard, and records F6 receipt. The authored coverage-exit assertion failed before the change and passes afterward. Focused adapter checks passed; the working candidate passed the 120-test repository gate with loopback access (including pre-existing unrelated local workflow changes). Independent Standards and Spec reviews found no remaining issues in the two-file change. Unrelated work remains uncommitted. No gameplay acceptance or merge is claimed.

Current live diagnosis: the existing game was attached/focused with owner authority and installed inventory verified against retained board source 96aba30. Mount diagnostic refused fresh-gameplay before input; cleanup not-needed. A temporary read-only script executed three samples: GTA player queries reported playing/on-foot/control enabled, while COMBINE_SKATE_READY returned zero. Installed board source adds board-renderer readiness to that command but omits it from the native status publisher, explaining contradictory readiness reporting. The exact renderer preparation/device/camera/depth/target condition remains unresolved. This is not evidence of a native return-format defect.

The game exited unexpectedly during a subsequent probe attempt; no execution marker for that attempt was captured, and no matching recent Windows application-crash event was returned. Cause unknown. Temporary probe files were removed. The owner restarted; fresh attachment/inventory succeeded and the same status mismatch recurred in windowed foreground gameplay. The owner now requires no further restart. No native plugin replacement or readiness bypass was attempted; the feedback source fix is not installed.

Installed data contains two prepared regions with no same-layer overlap. Existing authored JS/native replacement tests do not prove crossing real GTA collision. Next bounded collision slice remains two overlapping regions from one independently placed real collision resource, using existing refreshSurface/native replacement and unchanged guards; live readiness must first be resolved. No automatic acquisition, full-height coverage or continuous real-region ride is claimed.

[Jpatching]

Coverage source update: f0f3e15 adds rolling local collision windows and explicit qualified mesh coverage without a fixed outer radius. Final checks/reviews and remaining live-data blockers are recorded in #66. This remains a partial source delivery; no coverage bundle was installed and the current live 3 m patches were not expanded. Keep draft and unmerged.

[Jpatching]

Owner reports falling through while stationary immediately after mount. Isolated source fix published at 9070f92 on fix/gtaiv-mounted-ground-clearance, based on installed 96c6322; combined trick/coverage work remains separate and uncommitted.

Cause reproduced at the existing script presentation seam: the host pivot offset was measured against the transient airborne Session root. As that root settled, the frozen GTA character was lowered by the same displacement. Both synchronous and asynchronous mounts now use verified host ground clearance; the async path uses freshly validated ground. Original solved board/root motion is unchanged.

Regression red before fix, green after. Independent Standards/Spec review found no actionable defects. Adapter checks and repository gate passed (111 tests). Exact script with unchanged private surface data staged with hash checks/backups after normal close; restarted windowed. Live stationary contact/owner acceptance remains pending; source tests are not gameplay acceptance.

The isolated original-board/actual-Session fixture measured one transient 18.044mm penetration sample, then stable contact within approximately 1.9mm below its authored plane. It did not reproduce continued falling. That contact observation remains separate from the fixed host-pivot defect and actual GTA ground agreement. No solved GTA character animation is claimed.

[Jpatching]

Archived locally during repository reset to a clean main; not merged.