test(p0): retail find_cell_list trace parser + cdb value-capture tooling

P0 Task 5. RetailTrace parses the [fcl] golden format (seed/pos/picked,
RetailCellPick); 4 TDD tests green. find-cell-list-capture.cdb targets
CPhysicsObj::change_cell (commit-on-diff) to capture retail's accepted
membership sequence at the doorway; README is the operator runbook
(dt offset verification + decode_retail_hex float decode). The live run
is P0's one user-gated step (Task 6 mines existing traces first).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

tools/cdb/README-find-cell-list-capture.md

@@ -0,0 +1,75 @@
+# find-cell-list-capture — operator runbook
+
+Captures retail's **accepted membership sequence** at the cottage doorway so P0 can pin
+acdream's `CellTransit.FindCellList` against the real client (P0 Task 6 — the P1 gate).
+This is the one user-gated step of P0. Everything else (parser, fixture loader, goldens) is
+already headless + green.
+
+> **Try the autonomous path first.** Before running this, P0 Task 6 greps the *already-committed*
+> retail traces under `docs/research/2026-05-21-a6-captures/` for a usable membership pick. Only
+> run this capture if those traces don't yield one.
+
+## What you get
+
+A log of lines in the golden format the parser (`RetailTrace.ParseFindCellList`) reads:
+
+```
+[fcl] seed=0xA9B40031 px=0x43200000 py=0x41200000 pz=0x42BC0000 picked=0xA9B40170
+```
+
+`seed` = the cell the player was in; `picked` = the cell retail committed; `px/py/pz` = the player
+world origin as **raw IEEE-754 hex** (decode with `decode_retail_hex.py`). One line per accepted
+cell change, so a doorway walk yields a short clean sequence (e.g. `0031 → 0170 → 0171`).
+
+## Prerequisites
+
+1. **PDB matches the binary** (CLAUDE.md retail-debugger toolchain):
+   ```
+   py tools/pdb-extract/check_exe_pdb.py "C:/Turbine/Asheron's Call/acclient.exe"
+   ```
+   Expect `=== MATCH ===`.
+2. **Retail in-world at the Holtburg cottage doorway** (the building at world ≈ (161.9, 7.5, 94),
+   outdoor landcell `0xA9B40031`, vestibule `0xA9B40170`, room `0xA9B40171` — see
+   `docs/research/2026-06-03-p0-conformance-apparatus-notes.md`). Stand just outside the door.
+3. `cdb.exe` (x86) at `C:\Program Files (x86)\Windows Kits\10\Debuggers\x86\cdb.exe`.
+
+## One-time: verify the struct offsets (the script ships with placeholders)
+
+Launch cdb attached, let it break once, then dump the layouts and edit `$t4/$t5/$t6` in the
+`.cdb`:
+
+```
+dt acclient!CPhysicsObj @ecx        # note m_position (origin x,y,z) offset -> POS_OFF ($t6)
+                                    # note the current-cell pointer field   -> CELLPTR_OFF ($t5)
+dt acclient!CObjCell poi(@esp+4)    # note cell_id offset                   -> CELLID_OFF ($t4)
+```
+
+`change_cell` is `thiscall`: `this` (CPhysicsObj*) is `@ecx`; the new cell arg is `poi(@esp+4)`.
+The defaults in the script (`0x18 / 0x08 / 0x1C`) are best-guesses from `acclient.h` — **confirm
+them** before trusting a capture.
+
+## Run
+
+```powershell
+& "C:\Program Files (x86)\Windows Kits\10\Debuggers\x86\cdb.exe" `
+    -pn acclient.exe -cf tools\cdb\find-cell-list-capture.cdb *>&1 |
+    Tee-Object -FilePath "find-cell-list-capture.console.log"
+```
+
+Then, in retail, **walk slowly in and out of the doorway 5–10 times** (out → vestibule → room →
+back). The script auto-detaches (`qd`) after 400 cell changes; retail keeps running. Do **not**
+`Stop-Process` cdb — that kills retail (CLAUDE.md watchout).
+
+## Fold into the golden
+
+1. Decode the hex floats to decimals:
+   ```
+   py tools/cdb/decode_retail_hex.py find-cell-list-capture.log > find-cell-list-threshold.log
+   ```
+   (or hand-decode; each `0xHHHHHHHH` is a little-endian float). Result lines must read
+   `[fcl] seed=0xA9B40031 px=160.0 py=10.0 pz=94.0 picked=0xA9B40170`.
+2. Place the decoded file at
+   `tests/AcDream.Core.Tests/Conformance/Fixtures/find-cell-list-threshold.log`.
+3. Run `dotnet test … --filter FindCellList_DoorwayThreshold_MatchesRetailTrace`. GREEN = acdream
+   already matches retail at the threshold. RED = a real divergence → that is the **P1** work,
+   left as a documents-the-bug conformance test (no weakening the assertion — master plan §4).