erik/acdream
1
0
Fork 0
Code Issues Pull requests Projects Releases 6 Packages Wiki Activity Actions

docs(issues): #464 — the S4-c1 transcript gate replayed at the artifact pose PASSES: punch/seal events match retail exactly; the divergence is punch geometry or shell depth

Browse source 
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
This commit is contained in:
Erik 2026-09-03 22:30:31 +02:00
parent de33d49f22
commit d26a2ed035
1 changed files with 2 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

2
docs/ISSUES.md
View file

@ -113,6 +113,8 @@ then `/teleloc F4180113 39.502899 19.589222 181.119247 1 0 0 0`, zoom out, tilt
**RETAIL CAPTURE `cathedral-stair-arch` (2026-09-03 23:0x, owner in the paired 2013 client at the pose, tilted as in the screenshot; `docs/research/2026-09-01-overhaul/oh-capture/cathedral-stair-arch.walk.log` + `.alphadepth.log`, 5 frames each, commit `4f12cd06b`; retail screenshot posted in chat: the far face above the arch is SOLID, a doorway with sky shows only through the arch):** retail eye `(39.58,16.99,182.79)`, root `f4180114`. **Walk: retail admits EXACTLY our set** — `DI f4180114`, `DC ov=7 n=3: f4180114 f4180113 f4180112`, eleven `BLD` turns, and under `BLD f418000a` the same seven hall groups in the same order (`f4180109`; `f4180109 f4180108 f418010e f418010b`; `f4180109`; `f4180108 f4180107`; `f4180108 f4180107 f418010f f4180110 f4180106`; `f4180107`; `f4180107 f418010f f4180110 f4180106`), interior repaint `112 113 114` last. Only differences: acdream also turns the neighbouring blocks' `BLD f3170034`, `BLD f317003d` and draws `f3180105` (retail does not — a block-admission difference, unrelated to the artifact), and retail draws `f4180104` four times to our three. **Admission is eliminated as the cause.** **Alpha-depth: retail's frame is `PC ov=7 counter=0008 fc=0`, then for each building-portal group a `PM mode=1` FAR-DEPTH PUNCH of the portal polygon immediately BEFORE that group's `PC ov=0` cell draw (sixteen punches: `12980fd4`, `12981004`×3, `1cc24bd4`, `1cc24c34`×4, `4dbb5e2c`, `4dbb5dcc`, `4dbb5dfc`, `4dbb5d6c`, `4dbb5e5c`, `4dbb5d3c`, `4dbb5d9c`), then eight `PM mode=0` true-depth SEALS with `counterBefore` 0..7 (the exit views), `FL` by site 1029/11/1/1; 249 `AM` lines, all `clip=0`.** So in retail the hall can appear only inside the punched doorway outlines; the building shell's depth hides it everywhere else, including above the arch. Acdream lets it show above the arch, so the divergence is one of: the punch fans wider than the portal outline there (AD-17's unclipped fallback for a >8-edge or multi-polygon outside view), the punch/seal ORDER against the cells, or the shell's depth missing at that face. **This capture is the fifth pose of S4-c1's PM/PC transcript gate; running that gate names the divergence line by line. The fix contract follows from it (S4).** **RETAIL CAPTURE `cathedral-stair-arch` (2026-09-03 23:0x, owner in the paired 2013 client at the pose, tilted as in the screenshot; `docs/research/2026-09-01-overhaul/oh-capture/cathedral-stair-arch.walk.log` + `.alphadepth.log`, 5 frames each, commit `4f12cd06b`; retail screenshot posted in chat: the far face above the arch is SOLID, a doorway with sky shows only through the arch):** retail eye `(39.58,16.99,182.79)`, root `f4180114`. **Walk: retail admits EXACTLY our set** — `DI f4180114`, `DC ov=7 n=3: f4180114 f4180113 f4180112`, eleven `BLD` turns, and under `BLD f418000a` the same seven hall groups in the same order (`f4180109`; `f4180109 f4180108 f418010e f418010b`; `f4180109`; `f4180108 f4180107`; `f4180108 f4180107 f418010f f4180110 f4180106`; `f4180107`; `f4180107 f418010f f4180110 f4180106`), interior repaint `112 113 114` last. Only differences: acdream also turns the neighbouring blocks' `BLD f3170034`, `BLD f317003d` and draws `f3180105` (retail does not — a block-admission difference, unrelated to the artifact), and retail draws `f4180104` four times to our three. **Admission is eliminated as the cause.** **Alpha-depth: retail's frame is `PC ov=7 counter=0008 fc=0`, then for each building-portal group a `PM mode=1` FAR-DEPTH PUNCH of the portal polygon immediately BEFORE that group's `PC ov=0` cell draw (sixteen punches: `12980fd4`, `12981004`×3, `1cc24bd4`, `1cc24c34`×4, `4dbb5e2c`, `4dbb5dcc`, `4dbb5dfc`, `4dbb5d6c`, `4dbb5e5c`, `4dbb5d3c`, `4dbb5d9c`), then eight `PM mode=0` true-depth SEALS with `counterBefore` 0..7 (the exit views), `FL` by site 1029/11/1/1; 249 `AM` lines, all `clip=0`.** So in retail the hall can appear only inside the punched doorway outlines; the building shell's depth hides it everywhere else, including above the arch. Acdream lets it show above the arch, so the divergence is one of: the punch fans wider than the portal outline there (AD-17's unclipped fallback for a >8-edge or multi-polygon outside view), the punch/seal ORDER against the cells, or the shell's depth missing at that face. **This capture is the fifth pose of S4-c1's PM/PC transcript gate; running that gate names the divergence line by line. The fix contract follows from it (S4).**
**Gate replay (lead, 2026-09-03 22:29, scratch worktree `s4-c1-gate` at the S4-c1 fix-round commit `2bd353234` with the capture copied in and a fifth `[Fact]` added):** `AlphaDepthTranscript_CathedralStairArch_MatchesRetailFrame2` **PASSES** — acdream's PM (mode, counterBefore) and PC (ov, counter, forceClear) sequence at the artifact pose is EXACTLY retail's: the sixteen building-portal punches, the eight exit seals, the counter. So the divergence is NOT which punches/seals happen or their order; it is their GEOMETRY on screen (what each punch fan covers, what each seal writes) or the building shell's depth at that face — the transcript gate cannot see either. AD-17 records the one known geometric deviation exactly there: a >8-edge or multi-polygon OUTSIDE view (the stairwell's open face IS multi-polygon: three exit polys of `0x114` + two of `0x113`) yields a zero-plane slice and that view's punch fans draw UNCLIPPED, while retail CPU-clips every portal polygon exactly. Next: compare the punch-fan geometry per portal against retail's clipped polygons (retail's `DrawPortalPolyInternal` clips the polygon against the current portal view's planes before drawing; a vertex-dump capture at the pose, or the replay harness's own clipped polygons, gives the retail-side outline) and check that the cathedral's shell is drawn with depth before the punches. (Side finding of the same replay: `AlphaDepthTranscript_TerraceEdge` FAILS at `2bd353234` — see the S4-c1 fix-round record.)
## #463 — Soak `turn` samples can report a 0.1 ms GPU p50 (the turn window catches non-world frames) ## #463 — Soak `turn` samples can report a 0.1 ms GPU p50 (the turn window catches non-world frames)
**Status:** OPEN — found 2026-09-03 by the Campaign OVERHAUL v2 S3 review's perf re-measure. **Status:** OPEN — found 2026-09-03 by the Campaign OVERHAUL v2 S3 review's perf re-measure.