passages: continuous floor h(s) along the dense route; shell-probe oracle fix (L115, v223)

Floor height from the planner profile h(s) by horizontal projection + arc-length interpolation;
adjacent fine segments share endpoint values. Floor zero-crossing offsets max 7.8 mm (L114: up
to 73 mm); rock-floor close-ups show no ribs. The 2 L114 shell misses were oracle false positives
(points inside an overlapping bore); the oracle now checks every overlapping span; no gate relaxed.
Fast tier PASS; Clang/GCC W1/W8/W12 deterministic. Speed regression grows: near-ready +17..+31% vs
L109 (fine-segment/sample pairs 49.0M vs 19.8M). Not merged.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1

Standalone/Research/WorldAssemblerM3/Linux115/REPORT.md

@@ -0,0 +1,46 @@
+# Linux115 report
+
+Source version 223. Floor height now comes from the planner profile `h(s)` on the dense route, using horizontal projection and arc-length interpolation. In-range projections interpolate on their fine segment; out-of-range projections use the global profile lookup. Adjacent fine segments share the same endpoint value. The plan-time grade audit remains enabled: maximum sampled grade was 10.687° (limit 15°).
+
+## Floor zero crossings
+
+Each station has 51 density samples. Offsets below are in millimetres.
+
+| Passage | Station | Max absolute | Median absolute | P95 absolute |
+|---|---:|---:|---:|---:|
+| 0–1 | 25% | 0.088 | 0.009 | 0.066 |
+| 0–1 | 50% | 0.079 | 0.016 | 0.064 |
+| 0–1 | 75% | 0.225 | 0.016 | 0.161 |
+| 1–2 | 25% | 7.475 | 7.259 | 7.459 |
+| 1–2 | 50% | 7.272 | 7.215 | 7.265 |
+| 1–2 | 75% | 7.272 | 7.121 | 7.245 |
+| 2–3 | 25% | 7.849 | 7.712 | 7.844 |
+| 2–3 | 50% | 7.712 | 7.351 | 7.694 |
+| 2–3 | 75% | 7.573 | 7.506 | 7.566 |
+
+Maximum offset is 7.849 mm, below the 10 mm target. Full data: [CSV](../../../../Saved/Linux/l115/density-probe/zero-crossing-offsets.csv) and [summary](../../../../Saved/Linux/l115/density-probe/zero-crossing-summary.json).
+
+## Shell probes and captures
+
+The two L114 southwest wall misses were probe-oracle false positives. Both positions were inside an intentionally carved overlapping passage bore, where host air is correct; the audit had only considered the nearest route span when classifying bore interior. The oracle now recognizes interior air on any overlapping route span. No wall gate was relaxed. The final two-strata audit reports 0 wall violations in 31,361 probes (8,848 overlap-interior skips), 0 floor violations, and 0 interior violations.
+
+The final capture set has 23/23 programmatically verified images: 9 rock-floor close-ups and 14 open-air outside/interior close-ups across 7 sections. Quality labels are 18 `pass` and 5 `low_contrast_geometry_visible`; all five still passed visibility verification. [Capture index](captures/capture-index.json).
+
+## Speed
+
+Coarse candidate totals are L109 3,132, L114 14,384, and L115 3,720 (74.1% fewer than L114). The inner sweep still visits 48.95M fine-segment/sample pairs for both L114 and L115, versus 19.81M for L109, so fewer coarse candidates did not reduce fine traversal enough to improve readiness.
+
+Five interleaved measured runs per version followed warm-ups at W8/W12 for all four strata. The host was quiet before timing (1-minute load 1.30; `top` 99% idle). Values below are L115 near-ready median deltas.
+
+| Stratum | W8 vs L109 / L114 | W12 vs L109 / L114 |
+|---|---:|---:|
+| 0 | +16.8% / +8.7% | +16.8% / +9.2% |
+| 1 | +30.4% / +13.0% | +31.1% / +14.5% |
+| 2 | +25.2% / +12.4% | +25.9% / +12.0% |
+| 3 | +23.8% / +8.6% | +18.1% / +2.9% |
+
+Across the eight strata/worker combinations, median whole-view deltas were +10.7% versus L109 and +4.9% versus L114. Full run data: [timing summary](../../../../Saved/Linux/l115/timing/summary.json).
+
+## Verification
+
+Fast tier passed in 341.1 s, including walkability, passage shell, camera sweeps, native field, and motif capsule reachability. Clang/GCC hashes matched at W1/W8/W12 for all four owner strata groups. The final source stamp is `7eb1fa49f91954c0690173aba4ed832b3594e50531bd4b20690d122970943274`; no Unreal-compiled source changed, so no Unreal build was needed.