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

fix(p1): membership already matches retail — the 0/11 was a cdb capture artifact

Browse source 
The P1 "doorway membership lags retail" premise is FALSIFIED. acdream's swept
ResolveWithTransition already matches retail's true per-frame curr_cell: the
production gate ProductionPath_IndoorCrossings reads 9/9 on the indoor 0170<->0171
crossings with NO code change, once fed an aligned retail golden.

Root cause of the false 0/11: CPhysicsObj::SetPositionInternal calls change_cell
(acclient_2013_pseudo_c.txt:283456) BEFORE set_frame writes m_position (:283458),
so the original golden (find-cell-list-capture.cdb, read at the change_cell BP)
paired each frame's NEW cell with the PREVIOUS frame's position — a one-frame skew.
Verified 3 ways: the decomp ordering; golden_picked[i] == geom(golden_position[i+1])
for all 22 rows; acdream's static pick == golden_picked[i-1] for all rows. Both
retail and acdream pick with center-only point_in_cell on global_sphere[0] (no XY
lead; cache_global_sphere @ pc:274196). curr_cell commits via validate_transition
(@ pc:272608, curr_cell = check_cell) = the find_cell_list pick, structurally
identical to acdream's RunCheckOtherCellsAndAdvance -> FindCellSet -> SetCheckPos.
There was nothing to port; a swept advance would make membership LEAD by a frame.

- tools/cdb/find-cell-list-capture-aligned.cdb: re-capture reads the committed
  position from the set_frame that follows change_cell (cell+position same instant).
- Fixtures/find-cell-list-threshold.log: replaced with the aligned capture.
- ThresholdPortalCrossingReplayTests / FindCellListConformanceTests: rewritten from
  documents-the-bug to assert retail truth (per-segment / per-indoor-pick equality).
- handoff + notes + README + memory: banners correcting the disproven premise.

Still open (NOT indoor membership, which is DONE): outdoor->indoor 0031<->0170 entry
conformance (needs landcell + building stab in the gate cache); master-plan cleanups
(delete CheckBuildingTransit, unify find_env_collisions, demote ResolveCellId) refactor
working retail-faithful code -> need explicit user approval.

Conformance 60 pass / 1 skip / 0 fail; full Core 1309 pass / 5 fail (pre-existing
2 BSPStepUp + 3 door-collision = P2) / 1 skip.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This commit is contained in:
Erik 2026-06-03 18:54:27 +02:00
parent db94cb1c90
commit 9017107960
7 changed files with 197 additions and 64 deletions
Download patch file Download diff file Expand all files Collapse all files

16
docs/research/2026-06-03-p0-conformance-apparatus-notes.md
View file

@ -167,7 +167,21 @@ evidence. **Do NOT design or code a P1 membership "fix" before the production-pa
RED/GREEN is read.** The P0 `..._DivergesFromRetail_PendingP1` test is a UNIT-level pin only, NOT
evidence of a production divergence.
## ✅ RESOLVED — the production path DIVERGES (the "probe artifact" hypothesis is FALSIFIED)
## ⚠ SUPERSEDED 2026-06-03 — the "production path DIVERGES" conclusion below was a CAPTURE ARTIFACT
The section below concluded the production path genuinely diverges (0/11 → "port the swept curr_cell
advance"). **That is wrong.** The 0/11 came from a one-frame skew in the cdb golden:
`CPhysicsObj::SetPositionInternal` calls `change_cell` (`acclient_2013_pseudo_c.txt:283456`) BEFORE
`set_frame` writes `m_position` (`:283458`), so the original capture paired each frame's NEW cell with
the PREVIOUS frame's position (`golden_picked[i] == geom(golden_position[i+1])`, all 22 rows). An
aligned re-capture (`tools/cdb/find-cell-list-capture-aligned.cdb`, position read from the following
`set_frame`) makes the production gate read **9/9 with NO code change** — acdream's center-only
`point_in_cell` pick already IS retail's true per-frame membership. Canonical corrected finding:
`memory/project_retail_membership_criterion.md` + the RESOLVED banner in
`docs/research/2026-06-03-p1-membership-swept-advance-handoff.md`. The text below is retained as the
investigation trail.
## ✅ (HISTORICAL) RESOLVED — the production path DIVERGES (the "probe artifact" hypothesis is FALSIFIED)
Built `ThresholdPortalCrossingReplayTests.ProductionPath_IndoorCrossings_DivergeFromRetail_PendingP1`
(replays the golden indoor `0170↔0171` segments through the REAL `ResolveWithTransition` — engine

28
docs/research/2026-06-03-p1-membership-swept-advance-handoff.md
View file

@ -4,6 +4,34 @@
> (do NOT branch/worktree; do NOT push without asking; NEVER `git stash`/`gc`). PowerShell on
> Windows; launch logs are UTF-16. Read this FIRST, then the linked docs as needed.
## ⚠ RESOLVED 2026-06-03 — premise REVERSED (read this first)
The thesis of this doc ("port retail's swept `curr_cell` advance") is **FALSIFIED**. acdream's doorway
membership already matches retail. The "0/11 production divergence" was a **cdb CAPTURE ARTIFACT**:
`CPhysicsObj::SetPositionInternal` calls `change_cell` (`acclient_2013_pseudo_c.txt:283456`) BEFORE
`set_frame` writes `m_position` (`:283458`), so the original golden paired each frame's NEW cell with
the PREVIOUS frame's position — a one-frame skew (`golden_picked[i] == geom(golden_position[i+1])`,
verified on all 22 rows; and acdream's static pick `== golden_picked[i-1]` on all rows). Re-capturing
the COMMITTED position from the following `set_frame` (`tools/cdb/find-cell-list-capture-aligned.cdb`)
aligns cell+position, and the production gate `ProductionPath_IndoorCrossings_MatchRetail` reads
**9/9 with NO code change**. Both retail and acdream pick with **center-only `point_in_cell` on
`global_sphere[0]`** (the foot sphere does not lead the foot; `cache_global_sphere` @ pc:274196);
`curr_cell` commits via `CTransition::validate_transition` (@ pc:272608, `curr_cell = check_cell`) where
`check_cell` is the `find_cell_list` pick — **structurally identical to acdream's
`RunCheckOtherCellsAndAdvance` → `FindCellSet` → `SetCheckPos`**. There is nothing to port; porting a
swept advance would make membership LEAD the foot by a frame (a bug to satisfy a mis-aligned golden).
Corrected finding + current state: `memory/project_retail_membership_criterion.md` and
`tests/AcDream.Core.Tests/Conformance/Fixtures/README.md`. The two decomp questions below ARE answered
(Q1: no XY lead; Q2: commit is the `find_cell_list` pick via `validate_transition`) — the text below is
retained as the investigation trail. **Still open** (separate from indoor membership, which is DONE):
outdoor→indoor (`0031↔0170`) building-entry conformance (the building-only gate cache can't promote an
outdoor seed — needs the landcell + building stab loaded); the master-plan cleanups (delete
`CheckBuildingTransit`, unify `find_env_collisions`, demote `ResolveCellId`) refactor WORKING
retail-faithful code → need explicit user approval.
---
## State both altitudes
- **Milestone:** M1.5 — Indoor world feels right.