Fr0zkaandClaude Opus 5.5
4290672d1c
Root cause of liminal seed 5 (measured): passage carve makes the bore air (+3.52), then the liminal route fill stamp re-solidifies it (-0.16) where the deck crosses the tube. Three fixes cleared the bore but left 28-29 road stations unsupported; none kept. New gate: zero solid samples inside any passage interior, with no-passage negative controls. Owner v227 hashes and fast tier pass. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
47 lines
3.5 KiB
Markdown
47 lines
3.5 KiB
Markdown
# L146 report
|
|
|
|
**Result:** No production fix is retained. The cause is measured: the liminal route fill pass runs after the passage carve and makes part of the bore solid again. The new passage-interior capsule audit catches the seed-5 defect. After three fixes cleared the bore but failed the liminal road support gate, work stopped as required by `AGENTS.md`.
|
|
|
|
## Root cause
|
|
|
|
Measured at `(-16.087, 2.915, -722.743) m`, seed 5, Clang W1, composed L145 fixture:
|
|
|
|
- Before passage composition, density was `-3.62890625`.
|
|
- The passage SDF reported `union_distance_m=-0.879889786`; carving changed density to `+3.51955914` (air).
|
|
- The subsequent liminal route stamp changed it to `-0.16169475` (solid).
|
|
|
|
The diagnostic directly identifies the post-carve route-fill pass as the cause. The resolved route/deck geometry indicates the fill overlaps the bore there; that specific deck overlap is inferred from the layout. The half-pipe/open-air path, carve reach, and carve blend are not the cause at this sample: the passage carver ran and produced positive air density before the route stamp.
|
|
|
|
Trace: `Saved/Linux/l146/logs/diagnostic-seed05-clang-w1.log`.
|
|
|
|
## Fix attempts
|
|
|
|
All were temporary and have been reverted; production composition and evaluator version remain at L145 / v227.
|
|
|
|
1. Move fill stamps before passage carving: passage probes cleared, but the liminal road had 29 unsupported stations.
|
|
2. Suppress route fills inside the passage interior: same result, 29 unsupported stations.
|
|
3. Reroute the southern deck around the crossing: passage probes cleared, but 28 road stations remained unsupported.
|
|
|
|
The three seed-5 runs are `probe-seed05-clang-w1.log`, `probe-masked-seed05-clang-w1.log`, and `route-seed05-clang-w1.log` under `Saved/Linux/l146/logs/`.
|
|
|
|
The shared premise that changing stamp order or locally moving the deck could preserve both bore air and road support failed twice. The route adjustment also failed the support gate, so no fourth fix was attempted.
|
|
|
|
## Audit and checks
|
|
|
|
`StreamBenchmarkPipeline.h` now classifies each tested player-capsule sample against its passage interior and requires zero solid interior samples. A no-passage negative control uses the same interior classification and must detect solid samples for every passage.
|
|
|
|
On the final source, the seed-5 audit measured 50,904 interior capsule samples and 289 solid samples for passage 1-2. Both negative controls detected solid interior samples (165,769 and 49,612). The shell summary passed with zero floor or wall violations, and the liminal road audit passed. The composed passage audit exited 10 as expected because the new interior-air gate detected the defect.
|
|
|
|
The final-source evidence is `Saved/Linux/l146/logs/audit-seed05-clang-w1.log`.
|
|
|
|
- Clang W1 final-source build: passed; log `Saved/Linux/l146/fast/logs/build-fast-clang-final.log`.
|
|
- Fast verification tier: passed in 362.5 s; summary `Saved/Linux/l146/fast/fast-summary.json`.
|
|
- Owner canonical v227 hash rows: passed for both pinned owner worlds; W1 and W12 signatures matched.
|
|
- `git diff --check`: passed.
|
|
- The 32-run pad matrix and the liminal Clang/GCC W1/W8/W12 matrix were not run because the three-failed-fix stop condition was reached. The 16-seed cave-planner sweep is not on this branch.
|
|
- No Unreal `Source/` files changed.
|
|
|
|
## Remaining risk
|
|
|
|
The seed-5 passage interior remains partly solid in production. A new design must define how route floor support and passage air share their overlap before changing composition again. No passage-composition change or owner hash change is retained.
|