Fr0zka/VoxelForge
1
0 Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

research: diagnosis of the 7 unsupported landing probes (L154)

Browse Source 
(1) L152 rotated the upper chamber floor plane down 6-11 cm, so the chamber carve sits 4-14 mm inside the Surface World
route's 0.25 m support search; (2) two chamber probes lie inside the open passage bore (chamber audit lacks the
passage-interior exemption); (3) FI lower chamber extends 3.69 m above its host content band (fast tier red at HEAD).
Bisect: introduced by L152; L150/L151 pass.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
This commit is contained in:
Fr0zkaandClaude Opus 5.5 committed 2026-11-01 20:25:13 +01:00
1 parent bea40502af
commit 989a3ee646
1 file changed
+72
Download Patch File Download Diff File Expand all files Collapse all files
Standalone/Research/WorldAssemblerM3/Linux154/REPORT.md
+72
View File
@@ -0,0 +1,72 @@
# Linux154 — unsupported-probe diagnosis
**Result:** L152 introduced the seven failures by rotating the upper chamber floor plane and extending its carve over the Surface World route’s support samples. Five samples end 4–14 mm above that floor at the route audit’s 0.25 m limit. The other two belong to the upper chamber audit but sit inside the open passage bore. The Surface World endpoint and chamber center did not move. The Floating Islands content-band rejection also reproduces at HEAD. No production source was changed.
## Seed 2: the seven probes
The positions below are the failed support samples reported at the audit’s deepest sample. Distances to the chamber floor center and resolved passage spline are calculated from the L152 layout and spline. The sample category is measured from the audit that emitted it. Coordinates and densities are in metres and Core density units; SDF is in metres. Negative native density is solid, negative passage/chamber SDF is inside the carve, and positive composed density is air.
| # | Audit path | Position (m) | To chamber center (m) | To spline (m) |
|---|---|---|---:|---:|
| 1 | Surface World layer route | `(30.000, -8.170, -374.928)` | 6.503 | 6.691 |
| 2 | Surface World layer route | `(29.750, -8.170, -374.920)` | 6.253 | 6.446 |
| 3 | Surface World layer route | `(29.500, -8.170, -374.911)` | 6.003 | 6.201 |
| 4 | Surface World layer route | `(27.250, -8.170, -374.875)` | 3.754 | 4.046 |
| 5 | Surface World layer route | `(27.000, -8.170, -374.880)` | 3.504 | 3.817 |
| 6 | `passage/0-1/0` upper chamber, west spoke | `(18.229, -8.450, -375.624)` | 5.349 | 1.122 |
| 7 | `passage/0-1/0` upper chamber, west spoke | `(17.750, -8.450, -375.617)` | 5.821 | 1.087 |
At each deepest sample, the fields were:
| # | Max search depth | Native density | Passage-union SDF | Upper chamber SDF | Composed density |
|---|---:|---:|---:|---:|---:|
| 1 | 0.250 m | -0.371094 | -0.003868 | -0.003868 | +0.015472 |
| 2 | 0.250 m | -0.550781 | -0.008297 | -0.008297 | +0.033188 |
| 3 | 0.250 m | -0.527344 | -0.012726 | -0.012726 | +0.050903 |
| 4 | 0.250 m | -0.386719 | -0.014011 | -0.014011 | +0.056046 |
| 5 | 0.250 m | -0.535156 | -0.004768 | -0.004768 | +0.019073 |
| 6 | 1.125 m | -11.773438 | -0.032436 | +0.874998 | +0.129742 |
| 7 | 1.125 m | -9.085938 | -0.110422 | +0.875004 | +0.441689 |
The Surface World search samples at 0.0625, 0.125, 0.1875, and 0.25 m. At all five locations the chamber SDF stays negative through that search; at its limit it is only 4–14 mm inside the floor plane. Native density there is already solid, but the upper chamber carve keeps the composed field air. **The five route failures are caused by the chamber floor lying just beyond the route audit’s support depth.** At 2 m below the route feet, native density is -7.371 to -7.551 voxels and both passage-union and chamber SDFs are +1.736 to +1.746 m, so the carve has ended and solid support is present below the audited range.
The chamber audit samples from 0.125 through 1.125 m. The native field is solid throughout, but the passage-union SDF stays negative through 1.125 m while chamber SDF is +0.875 m there. At the next 0.125 m step the passage SDF is positive. At 2 m below the chamber feet, passage SDFs are +0.817 and +0.732 m, chamber SDFs are +1.750 m, and native density is -15.273 and -12.586 voxels. **The two chamber failures are inside the passage bore; the passage carve leaves air there.** They are about 1.1 m from the spline, unlike the five Surface World samples, which are 3.8–6.7 m away.
No failed sample came from the passage-approach audit; `passage/0-1/0` reported zero unsupported samples. All seven points are within the upper chamber’s 7.5 m radius. Thus the five Surface World samples are under the chamber footprint, and the two chamber-audit samples are physically at its west passage opening.
Raw profiles and audit rows: [extended seed-2 trace](../../../../Saved/Linux/l154/l154-seed2-extended.log), [L152 seed-2 audit](../../../../Saved/Linux/l152/seed2-plane-fit.log).
## Bisect: L150 → L152
Same owner world, seed 2, W8. L150 and L151 binaries were replayed for this audit; L152 is the saved focused run.
| Revision | Surface World route unsupported | Upper `passage/0-1/0` chamber unsupported | Result |
|---|---:|---:|---|
| L150 `9ac60d5` | 0 | no landing-chamber audit | route passes |
| L151 `5bd158d` | 0 | 0 | route and chamber pass support |
| L152 `11b59a9` | 5 | 2 | first appearance of these seven |
The Surface World route endpoint is `(23.500, -8.000, -374.831)` in all three layouts. L151 and L152 also keep the upper chamber floor center at that same position. The full `passage/0-1/0` control-point positions are identical between L151 and L152.
**Measured change:** L151’s upper floor gradient was `(94, 0)` Q16 (`+0.001434` in X); L152’s was `(-1015, 0)` Q16 (`-0.015488` in X). L152 changed the chamber fit to the farthest receiving-route point in its radius and made the chamber floor profile the passage field’s floor profile. Across the failed route samples, the plane therefore moved down by 0.059–0.110 m. That moves the chamber carve across the deepest 0.25 m support probes without moving the route endpoint, chamber center, or passage spline control points. L152 is the introducing change; adding the landing chamber in L151 did not produce these support failures.
Evidence: [L150 replay](../../../../Saved/Linux/l154/l150-seed2-audit.log), [L151 replay](../../../../Saved/Linux/l154/l151-seed2-audit.log), [L152 report](../Linux152/REPORT.md).
## Probe-placement fixture
**Measured:** the same placement logic was run against a hand-built flat floor and a known open mouth. It returned `flat_floor=PASS`, `flat_unsupported=0`, `open_mouth=EXPECTED_FAIL`, `mouth_unsupported=3`, `status=PASS` ([fixture output](../../../../Saved/Linux/l154/l154-seed2-extended.log)). The helper places support samples correctly on the flat control and detects columns under an intentionally open mouth.
**Inferred:** probes 6–7 are correctly sampling the open bore, not a misplaced point. The chamber audit applies its support test there without the passage-interior support exemption used by the passage-route audit. These two should be interpreted as open-mouth samples. The fixture does not show a placement error in the five Surface World route probes: their SDF profile independently places them inside the chamber floor carve until just beyond the route audit’s 0.25 m search limit.
## Floating Islands content-band check
**Measured at HEAD:** the fast tier still fails its first case, `owner-floating-islands-style2-clang`, in 0.5 s. It stops with `passage/0-1/0 lower landing chamber does not fit inside its host content band`; later fast checks do not run. The restored-source run uses stamp `b299076d…` ([fast summary](../../../../Saved/Linux/l154/fast-summary.json), [log](../../../../Saved/Linux/l154/owner-floating-islands-style2-clang.log)). This is the same failure L153 recorded.
A diagnostic print at the validation site measured the lower chamber’s required Z extent as `[-130.202148, -122.557617] m`. Its Floating Islands host content band is `[-625.750000, -126.250000] m`. The chamber’s upper extent crosses the host top by **3.692383 m**. Its floor center is `-128.629882 m`, only 2.380 m below that boundary; the full 4.5 m clearance, floor-plane reach, and 0.75 m shell extend above it. The upper chamber fits its host band. The planner rejects the lower chamber before writing a layout snapshot. The temporary diagnostic print was removed after this measurement; its output is saved in [the content-band diagnostic log](../../../../Saved/Linux/l154/floating-islands-contentband-diagnostic.log).
## Checks and limits
- Passed: temporary Clang diagnostic build; known-answer placement fixture; source restoration (`git diff` is empty).
- Ran: seed-2 audits for L150, L151, and L152; restored-source HEAD fast tier.
- Failed: HEAD fast tier at the Floating Islands content-band validation, before downstream checks. This remains a red required gate.
- Not run: timing, other seeds, or fixes. No Unreal `Source/` files changed.