# L159 — pad stamp only

The stamp is implemented and built as evaluator v240, with no planner edits. All 12 captures passed image quality, but the seed-0 fast tier and both 16-seed composed sweeps failed; no baselines were added.

## Seed-0 tunnel-end captures

Each row matches the L151 eye-height pose and adds a pulled-back L159 wide view.

| Tunnel end | Before (L151) | After, eye height (L159) | After, wide (L159) |
|---|---|---|---|
| `passage/0-1/0` upper | ![L151 upper](../Linux151/captures/after-l151-0-1-0-upper-landing.jpg) | ![L159 upper eye](captures/after-l159-0-1-0-upper-landing.jpg) | ![L159 upper wide](captures/after-l159-0-1-0-upper-wide.jpg) |
| `passage/0-1/0` lower | ![L151 lower](../Linux151/captures/after-l151-0-1-0-lower-landing.jpg) | ![L159 lower eye](captures/after-l159-0-1-0-lower-landing.jpg) | ![L159 lower wide](captures/after-l159-0-1-0-lower-wide.jpg) |
| `passage/1-2/0` upper | ![L151 upper](../Linux151/captures/after-l151-1-2-0-upper-landing.jpg) | ![L159 upper eye](captures/after-l159-1-2-0-upper-landing.jpg) | ![L159 upper wide](captures/after-l159-1-2-0-upper-wide.jpg) |
| `passage/1-2/0` lower | ![L151 lower](../Linux151/captures/after-l151-1-2-0-lower-landing.jpg) | ![L159 lower eye](captures/after-l159-1-2-0-lower-landing.jpg) | ![L159 lower wide](captures/after-l159-1-2-0-lower-wide.jpg) |
| `passage/2-3/0` upper | ![L151 upper](../Linux151/captures/after-l151-2-3-0-upper-landing.jpg) | ![L159 upper eye](captures/after-l159-2-3-0-upper-landing.jpg) | ![L159 upper wide](captures/after-l159-2-3-0-upper-wide.jpg) |
| `passage/2-3/0` lower | ![L151 lower](../Linux151/captures/after-l151-2-3-0-lower-landing.jpg) | ![L159 lower eye](captures/after-l159-2-3-0-lower-landing.jpg) | ![L159 lower wide](captures/after-l159-2-3-0-lower-wide.jpg) |

Capture quality: **12/12 PASS**. Poses use the seed-0 layout resolved for L151; the six after-eye views reuse those camera poses. The image manifest and capture commands are in [captures/INDEX.md](captures/INDEX.md) and [RunL159Captures.py](RunL159Captures.py).

## Change

- Added a last-step stamp to both `ApplyResolvedStrateComposition` branches. It uses the existing landing-chamber floor center as P, radius `max(3 m, mouth radius + 1 m)`, fill depth 1.25 m, headroom 3 m, 12° cones, and an 8 m blend reach. The tube-only distance mask leaves the bore open; the smooth-min width rises to 0.5 m over the first metre beyond the flat radius.
- Removed the landing chamber box/shell composition cases. Tunnel endpoints and planner output were left unchanged.
- Extended tile feature bounds, proof bounds, fallback overlap bounds, and the proof-miss sampler for the pad influence.
- Added `LandingPadAudit` for floor flatness, support, headroom, blend grade, and four negative controls.
- Bumped `WorldAssemblyEvaluatorVersion` from 239 to 240.

No Unreal `Source/` files changed. Changed Standalone files: [ShapeWorldInternal.h](../../../Core/src/ShapeWorldInternal.h), [VoxelForgeCore.h](../../../Core/include/VoxelForgeCore/VoxelForgeCore.h), and [StreamBenchmarkPipeline.h](../../../Source/StreamBenchmarkPipeline.h). Planner files are unchanged.

## Gates

| Gate | Tunnel Network | Vertical Shafts | Result detail |
|---|---:|---:|---|
| Seed-0 fast tier | FAIL | — | Stopped before composition: `owner-floating-islands-style2-clang` rejected because `passage/0-1/0 lower landing chamber does not fit inside its host content band` (exit 2, 0.465 s). |
| Combined composed run | 0/16 PASS | 0/16 PASS | Each seed returned failure from the resolved passage audit. |
| Pad audit | 0/16 seeds PASS | 0/16 seeds PASS | 8/96 endpoints passed per mode. Across the 96 pads: 49 had flatness misses (369 samples total; worst reported variation 0.74946 m), 75 had blend-grade misses (2,072 samples; max 80.754°), support was 9,384/10,368 probes, and headroom passed on 93/96 pads. There were 1,682 missing blend crossings. All four negative controls were detected on every pad: 384/384 per mode. |
| Route plan grade | 16/16 PASS | 16/16 PASS | Planned grades stayed within the 15° limit. |
| Selected-stratum route audit | 0/16 PASS | 1/16 PASS | Vertical Shafts passed only seed 2; its selected route had 0 blocked or unsupported samples. Tunnel Network failed all 16: blocked samples were present on 15 seeds and unsupported samples on 13. The full composed passage walkability summary still failed 16/16 in each mode; `route_walkable` was 0 on 15 seeds and 2 on seed 2. |
| Interior air | 0/16 PASS | 0/16 PASS | Every seed had solid capsule samples (839–1,588 per seed). |
| Shell | 11/16 PASS | 11/16 PASS | Failed seeds 2, 3, 4, 5, and 14: 2,571 floor violations and 570 wall violations in total. Other shell samples passed. |
| Ripple | 5/16 PASS | 5/16 PASS | Passing seeds: 3, 7, 11, 12, and 14. |
| Layout envelope | 8/16 PASS | 8/16 PASS | Conflicts on seeds 1, 4, 6, 7, 8, 10, 13, and 15, on existing route-edge, room, or shaft primitives near passage endpoints. |
| Tile seams | NOT REACHED, 0/16 rows | NOT REACHED, 0/16 rows | The command included `--audit-passage-tile-seams`, but every run returned from the failed passage summary before reaching the seam-audit block. Seam behavior is therefore unverified. |
| Pad proof/feature bounds (L130 sampler) | 16/16 PASS | 16/16 PASS | Per mode: 413,952 sampled points, 305,571 sign-relevant; zero proof misses and zero tile-feature misses. |
| Mode blocked-route control | 16/16 PASS | 16/16 PASS | Deliberately blocked controls were detected as expected. |

Per-seed detail: pad `F/G` is the number of endpoints with any flatness/blend-grade failure; route `B/U` is the selected-stratum blocked/unsupported sample count. Shell `F (x/y)` gives floor/wall violations.

| Seed | Pads passed | Pad F/G | Tunnel Network B/U | Vertical Shafts B/U | Interior solids | Shell | Ripple | Envelope |
|---:|---:|---:|---:|---:|---:|---:|:---:|:---:|
| 0 | 1/6 | 3/5 | 324/12 | 507/0 | 1,233 | PASS | FAIL | PASS |
| 1 | 1/6 | 3/4 | 190/0 | 157/0 | 839 | PASS | FAIL | FAIL |
| 2 | 1/6 | 3/4 | 159/2 | 0/0 | 915 | FAIL (513/110) | FAIL | PASS |
| 3 | 1/6 | 3/4 | 0/14 | 521/0 | 1,588 | FAIL (522/117) | PASS | PASS |
| 4 | 0/6 | 3/5 | 140/0 | 222/0 | 855 | FAIL (512/114) | FAIL | FAIL |
| 5 | 0/6 | 3/4 | 152/15 | 379/0 | 892 | FAIL (516/116) | FAIL | PASS |
| 6 | 0/6 | 3/6 | 436/24 | 495/0 | 1,350 | PASS | FAIL | FAIL |
| 7 | 1/6 | 3/5 | 299/5 | 374/0 | 1,011 | PASS | PASS | FAIL |
| 8 | 1/6 | 3/4 | 177/15 | 541/0 | 1,079 | PASS | FAIL | FAIL |
| 9 | 0/6 | 3/4 | 325/16 | 540/0 | 1,205 | PASS | FAIL | PASS |
| 10 | 1/6 | 3/5 | 171/16 | 385/0 | 921 | PASS | FAIL | FAIL |
| 11 | 0/6 | 3/5 | 150/30 | 537/0 | 1,025 | PASS | PASS | PASS |
| 12 | 1/6 | 3/4 | 266/4 | 353/0 | 1,161 | PASS | PASS | PASS |
| 13 | 0/6 | 4/6 | 153/15 | 376/0 | 1,146 | PASS | FAIL | FAIL |
| 14 | 0/6 | 3/4 | 277/18 | 386/0 | 1,075 | FAIL (508/113) | PASS | PASS |
| 15 | 0/6 | 3/6 | 173/0 | 265/0 | 970 | PASS | FAIL | FAIL |

Full per-seed records: [Tunnel Network summary](../../../../Saved/Linux/l159/seeds/tunnel-network/summary.json) and [Vertical Shafts summary](../../../../Saved/Linux/l159/seeds/vertical-shafts/summary.json).

The pad, route, shell, interior-air, ripple, and envelope results are composition audits, one run per seed 0–15 with `--workers=8 --clip-radius=3 --max-clip-level=8 --max-level=0 --force-generate --recipe=surface-world --audit-resolved-layout-passages --audit-passage-tile-seams` and the same source stamp (`5b8ab5e84364e26c1a555428fba8bedb23faed14ef0f6b7d397edf654883d640`). Both modes produced the same shared-field results. The selected-stratum route row and blocked-route control are mode-specific.

The seed runner recorded 6,711.2 s summed child wall time for Tunnel Network and 6,762.5 s for Vertical Shafts (monotonic windows from each seed process launch to exit). These are sweep execution windows, not performance benchmark results; no L80 timing was taken.

## Checks and remaining risks

- Build: `bash Standalone/BuildLinux.sh --compiler=clang++-19 --variant=l159-clang --fp-contract=off --jobs=4` — passed.
- Captures: `RunL159Captures.py` — 12 views, 0 quality failures.
- Fast tier: `VOXELFORGE_L130_PASSAGE_REACH_PROOF_AUDIT=1 python3 Standalone/Tools/verify/verify.py fast --out Saved/Linux/l159 --overall-timeout 1740` — failed at the owner floating-islands fixture before composition.
- Composed sweeps: `RunL159SeedSweep.py --kind tunnel-network` and `--kind vertical-shafts`, seeds 0–15, with `VOXELFORGE_L130_PASSAGE_REACH_PROOF_AUDIT=1` — completed, all aggregate gates failed.
- `git diff --check` passed before the final report was written. No Unreal build or timing benchmark was run.

The pad flatness, support, headroom, and blend gates remain open, and all composed passage summaries fail. Shell failures are seed-dependent. The proof-tree and tile-feature bound sampler passed; fallback overlap bounds were extended but have no separate standalone audit. Seam behavior remains unverified because the process exits before that gate. No baselines were added because the requested gates did not all pass.