Fr0zkaandClaude Opus 5.5
bd7bba994c
The FI->TN wall beat (4.77 cm at 7.06 m) was the lower-mouth flare: the full mouth radius was assigned to the penultimate control point, a one-segment radius jump. The shared route builder now blends to the mouth radius over 12 m with a smoothstep. Wall ripple 1-2: 7.27 -> 1.62 cm. All three passages pass support/blocked/grade/ripple/naturalness; controls detected. Fast tier 269/269 incl. the new-plan stage; W1/W8/W12 equal, Clang and GCC. Hash pins: FI style 2 re-pinned at v213. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
49 lines
4.0 KiB
Markdown
49 lines
4.0 KiB
Markdown
# Linux68 — FI→TN 7 m wall beat
|
||
|
||
## Cause and fix
|
||
|
||
The 7.06 m beat came from the authored-radius profile. In the failing 32 m window, a radius-only diagnostic reproduces 4.77 cm at 7.0625 m. The route’s interior controls are generally 5.6–6.3 m apart; the last 3.0→4.5 m radius change spans 5.803 m, followed by an 8.8 m constant-radius span. The measured beat does not repeat at either spacing. Removing relief left 7.46 cm, and restoring the earlier carve cutoff left 7.55 cm. The shorter 114.23 m route moved the lower-mouth flare into the failing window; the L67 split rose from 6.78 to 14.52 cm.
|
||
|
||
The shared route builder now blends the lower-mouth radius over 12 m with a smooth distance-based taper, replacing the single-segment radius jump. Evaluator version is **v213**. L67 support changes remain in place.
|
||
|
||
## Passage audits
|
||
|
||
Clang and GCC produced identical results under the authored-radius gauge with cubic detrending per 32 m window. Limits are unchanged: wall ≤3.5 cm, floor ≤2.5 cm.
|
||
|
||
| Interface | Wall ripple | Period | Floor ripple | Blocked / unsupported / grade | Result |
|
||
|---|---:|---:|---:|---:|---|
|
||
| FI 0–1 | 1.45 cm | 4.562 m | 0.06 cm | 0 / 0 / 0 | PASS |
|
||
| FI 1–2 | 1.62 cm | 7.250 m | 1.33 cm | 0 / 0 / 0 | PASS |
|
||
| TN 2–3 | 0.92 cm | 7.375 m | 0.32 cm | 0 / 0 / 0 | PASS |
|
||
|
||
Naturalness, mouth profile, and ripple audits passed 3/3. Synthetic sphere, floor bead, floor step, L23 smooth-wall, and L23 terrace controls were detected. Blocked controls were detected 3/3. The new-plan generator/style maps, 3/3 interfaces, and three transition audits passed; width and thickness negative controls were rejected.
|
||
|
||
Clang and GCC builds passed with source stamp `33a8215f36abc00fadcdde5d6f3ff2d7054454fc2158e32470fb059e68bbd12e`. W1/W8/W12 hashes matched for all four strata on both compilers (8 compiler/stratum rows, 12 hash rows per worker count). The final fast tier passed **269/269** checks in 265.1 s. Its new-plan stage passed. The first fast attempt exposed the expected v213 floating-islands hash change against the old L52 pin; the current W1/W12 hashes match, and the v213 pin is [here](hash-baselines/owner-hash-floating-islands-style2-v213.log).
|
||
|
||
## Captures
|
||
|
||
All files are below 300 KB. The 1–2 camera is at the old ripple window. The 2–3 cameras are inside the passage and aimed through the mouths. The upper 2–3 exterior is sparse in the capture mesh, so that view reads as a dark opening.
|
||
|
||
| View | File | Bytes | SHA-256 |
|
||
|---|---|---:|---|
|
||
| 1–2 mid-tunnel | [JPEG](captures/passage-1-2-0-mid-tunnel.jpg) | 29,282 | `39381b01ae4d1674800391c6d8c65383e0d6866abf3cd70a8330062c677d06fa` |
|
||
| 2–3 arrival mouth | [JPEG](captures/passage-2-3-0-arrival-mouth.jpg) | 20,123 | `f73ce987932a5e2ba165a3272b18d8786fcce3f815d0c2a1edcf13c99cc3f7b4` |
|
||
| 2–3 departure mouth | [JPEG](captures/passage-2-3-0-departure-mouth.jpg) | 26,024 | `7665166f8f2e78b86774931e53af32243a5f40d2fc95e1b7ad4bf8cd6d310bd5` |
|
||
|
||
## Whole-view cost
|
||
|
||
Three interleaved L67/L68 pairs per stratum; all 24 runs exited 0 with 12 hash rows.
|
||
|
||
| Stratum | L67 median | L68 median | Change |
|
||
|---:|---:|---:|---:|
|
||
| 0 | 1.011 s | 1.074 s | +6.23% |
|
||
| 1 | 1.494 s | 1.503 s | +0.60% |
|
||
| 2 | 0.581 s | 0.577 s | −0.69% |
|
||
| 3 | 0.210 s | 0.210 s | 0.00% |
|
||
|
||
## Failed runs and discarded diagnostics
|
||
|
||
The first fast attempt failed only on the stale L52 floating-islands hash expectation; the updated v213 pin and final fast run passed. Relief-off and old-cutoff experiments still failed the wall limit. A generic control-point refinement reduced ripple but introduced one unsupported sample; a post-route refinement retained 7.23 cm ripple. Those experiments were reverted. One early audit used the landmark-demo flag on an incompatible passage fixture and exited during route setup; the corrected audit passed.
|
||
|
||
Evidence is under `Saved/Linux/l68`: `build-clang.log`, `build-gcc.log`, `logs/final-{clang,gcc}-owner.log`, `logs/transition-clang.log`, `verify/owner-w1-w8-w12-determinism.json`, `cost/cost-summary.json`, and `verify/fast-tier-final.log`.
|