Fr0zkaandClaude Opus 5.5
7d654ff861
Layers may publish certified arrival/departure pads; the passage planner binds endpoints and tangents to them; no-pad layouts serialize unchanged (owner canonical hashes match v227). Pads removed the endpoint grade failures; seeds 0-4 passed at a 45 m margin, but seed 5 still has 289 blocked passage-capsule samples ~29 m before the departure pad, where the composed field stays solid (-0.162) inside the tube over native ground (-3.6). Ripple audit now keeps true route distance across masked gaps (with a synthetic control). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
52 lines
6.5 KiB
Markdown
52 lines
6.5 KiB
Markdown
# Linux145 — Core.Liminal.Field v2 landing pads
|
||
|
||
**Result: endpoint landing pads alone did not make the liminal join solvable.** The pads bind the passage mouth to certified ground and remove grade failures in the final seed 5 probe, but the passage capsule still intersects native Surface World terrain about 29 m from the departure pad. This matches R48's warning that a pad must cover the approach and tube exclusion envelope; stop here and use a bounded joint plan.
|
||
|
||
## What changed
|
||
|
||
- Added an opt-in landing-pad record to resolved strata: stable ID, arrival/departure role, position, floor height, Q16 approach direction, clearance, support depth, and a certification bit. The passage planner validates the record and binds the endpoint and tangent to it. Empty pad lists are omitted from canonical serialization, preserving legacy owner layouts and hashes.
|
||
- Added `Core.Liminal.Field` v2 pads on the field profile: flat ground, hill/valley profile, road endpoints, and pillar-grid clearings. Its layer route connects the pads; the road uses that route as its road centerline.
|
||
- Added an uncertified-pad negative control. Corrected the ripple audit to retain each valid sample's actual route distance across masked gaps; a 5 m synthetic gap control now confirms that the measurement detects a 5.8 cm ripple and rejects the same smooth cubic control.
|
||
- Fixed the composed audit's footing placement so it counts actual structural supports. No Unreal `Source/` files changed.
|
||
|
||
## Count first: failure cause
|
||
|
||
L144's fixed-mouth route already showed both end-connector grade and capsule blockage. For the flat-field seed 0 probe, the liminal route had 222 blocked capsule samples and 7 grade failures, with no unsupported samples or structural-support failures. Pillar-grid seed 3 had 289 blocked samples and 9 grade failures. In the separate seed 5 audit, interface 1–2 had 10 blocked passage-capsule samples and the liminal route had 9 blocked samples, both with zero grade failures. These results say support alone was not the cause; the fixed endpoint geometry sometimes made the approach too steep, while other cases had occupied space at the join. [L144 report](../Linux144/REPORT.md); logs: `Saved/Linux/l144/logs/probe8-seed00.log`, `probe8-seed03.log`, `audit-seed05.log`.
|
||
|
||
The pad experiment removed the endpoint-grade part in seed 5, but not the occupied-space part:
|
||
|
||
| Probe | Passage 1–2 capsule | Grade / support | Shell | Liminal route |
|
||
|---|---:|---|---|---|
|
||
| L144 seed 5, fixed mouths | 10 blocked | 0 grade failures; 0 unsupported | not the failing gate | 9 blocked; 0 grade failures |
|
||
| L145 seed 5, 45 m ground margin | 140 blocked; 20 interior shell violations | 0 grade failures; 0 unsupported | FAIL | PASS |
|
||
| L145 seed 5, final 20 m ground margin | 289 blocked | 0 grade failures; 0 unsupported | PASS | PASS |
|
||
|
||
The final L145 Clang W1 probe (source stamp `82e3b58c…`) certified both pads. The departure pad was `(12, 0, -716.451) m`, with floor `-717.581 m`, 6 m clearance, and 2 m support depth. Yet the interface-1–2 passage reported 289/50,904 capsule samples blocked, with 0/4,040 unsupported and 0 grade failures. A blocked sample at `(-16.087, 2.915, -722.743) m` had native density `-3.629` and composed density `-0.162`; it is about 29 m from the departure pad. The sample is inside the passage tube, so the layer's field remains solid across the approach even though its pad itself is certified. The lower interface, both shell audits, both ripple audits, and the liminal walk audit passed; the final composed passage summary failed and seam audit did not run for this seed. See `Saved/Linux/l145/logs/audit-seed05-clang-w1.log`.
|
||
|
||
For the 45 m attempt, the original seed 0–4 Clang W1 probes passed their composed gates, then seed 5 failed. The seed 5 log was overwritten by the final 20 m rerun; its counts above are from that earlier run. The corrected ripple fitter also removed a false 5.99 cm reading caused by compressing masked samples: at 45 m, seed 0 measured 2.87 cm after retaining route distances. These are probe results, not timing measurements.
|
||
|
||
The third repeated seed 5 failure reaches the requested stop condition. This does not show that a full candidate-and-approach contract is impossible; it shows the tested endpoint-only pad contract is too weak. The passage planner needs to choose a candidate while checking the full capsule approach and tube against the layer field, as R48 recommends.
|
||
|
||
## Captures
|
||
|
||
One 1.6 m eye-height capture was rendered at each arrival pad. All four passed `capture_quality.py`; mesh export used `--capture-skip-walkability`, so these images show appearance and are not walkability evidence.
|
||
|
||
| Variant | Capture |
|
||
|---|---|
|
||
| Flat field + monolith |  |
|
||
| Repeating hills |  |
|
||
| Road over cloud floor |  |
|
||
| Pillar grid |  |
|
||
|
||
## Checks
|
||
|
||
- Clang 19 headless and offscreen-capture binaries reported the final source stamp. Four arrival captures passed quality checks.
|
||
- Fast tier passed on the final source: `Saved/Linux/l145/fast/fast-summary.json` (280.8 s tier duration). The initial default-output attempt refused a stale L42 binary before tests started; rerunning into the isolated L145 output directory with the current L145 binary passed.
|
||
- Owner canonical hash checks passed: Surface World style 1 and Floating Islands style 2 each matched all 12 pinned L130-v227 rows at W1 and W12, and each recipe's W1/W12 rows matched. These exercise the no-pad owner path.
|
||
- `git diff --check` passed.
|
||
- The landing-pad composed matrix stopped at seed 5 after the repeated failure. It did not complete 8 seeds per variant (32 runs). Clang/GCC W1/W8/W12 identity for the liminal fixture was not run. Final-source seeds 0–4 were not rerun after the margin returned to 20 m. No performance timing was collected.
|
||
|
||
Changed standalone files: `Standalone/Core/include/VoxelForgeCore/VoxelForgeCore.h`, `Standalone/Core/src/ResolvedLayoutInternal.h`, `Standalone/Core/src/WorldAssemblerInternal.h`, `Standalone/Source/StreamBenchmarkHeadless.cpp`, `Standalone/Source/StreamBenchmarkPipeline.h`, and `Standalone/Source/WorldAssemblyPrototype.h`.
|
||
|
||
Remaining risk: the current pad record certifies a local landing and direction, not an approach corridor. A future joint planner must reserve/check the full capsule and tube against the layer before finalizing pad choice and passage geometry.
|