phase8: rules check after P8-T4

phase-ends/CURRENT_PHASE.md

@@ -16,7 +16,7 @@ matched bodies** (from 34), every one verified by the coordinator's own gate.
 - [x] **P8-T2 — Claim-merge tooling and the protocol record** (complete)
 - [x] **P8-T3 — Dispatch both charters and verify cycle 1** (complete: 18 claims verified, 57 regions)
 - [x] **P8-T4 — Verification cycle 2 with re-partitioning** (complete: 58 claims verified, 115 regions / 109 bodies — milestone met)
-- [ ] **Rules check**
+- [x] **Rules check** — re-read `AGENTS.md` mandatory behavior after P8-T4 and stated the required continuation notice.
 - [ ] **P8-T5 — The SDK-shape investigation, bounded** (worker C, in progress)
 - [ ] **P8-T6 — Verification cycle 3 and consolidation**
 - [ ] **P8-T7 — Cookbook, conventions, protocol record, and phase gate**
@@ -178,6 +178,22 @@ site.
 | Ghidra's PSX loader **collapses COP2 sequences** into one pseudo-op | Worker A found `0x8001AE3C` rendered as a single `ldbkdir a0,a1,a2`, so Ghidra's instruction count can be *lower* than reality. Never use it as a size; the extent table is authoritative. |
 | Two `$gp`-switch thunk halves are not C regions | `0x80108034`, `0x8010804C` (worker B). Added to the worklist's named exclusions. |
 
+## Rules check
+
+Re-read `AGENTS.md` mandatory behavior after P8-T4. One task at a time; explain before changing code or
+architecture; never overwrite blind; evidence before assumption with Ghidra and PCSX-Redux as the
+oracles; a match is byte-for-byte with the full-binary hash green; unmatched code stays fallback; verify
+from clean state and read exit codes; check duplicates before matching and share a verified body through
+the documented mechanism; document uncertainty; stop after two distinct unexplained failures.
+
+The ROM/repository firewall and the `git clean -x` prohibition stand. Two of the rules earned their
+keep this cycle in an unusual way: *verify from clean state* is what caught the stale worklist before it
+wasted two workers, and *a match is byte-for-byte* is what stopped a bad candidate cycle from reaching
+the tracked registry — the worker claims were plausible, locally verified, and still wrong for one
+region.
+
+`Rules check — re-read complete. Continuing with P8-T5 coordination.`
+
 ## Notes and limits
 
 - Three sessions share one worktree. The safety property is that a worker's claim is never trusted: the