phase12: merge worker D's final row 0x8002D014 (685 bodies / 694 regions, +83) + record my attribution error

phase-ends/logs/Phase12.md

@@ -1776,3 +1776,33 @@ from a worker reading its own input file before trusting it, and none of them fr
 here **only so the working tree is clean at the stop**, and it is marked as unverified: it is not a claim,
 it is not in the registry, and nothing depends on it. Recording the distinction because "committed" and
 "verified" are different states and this file is the former.
+
+### Worker D's final row — and it is the file I had committed as UNVERIFIED
+
+**`0x8002D014` (76 B) MATCHES**, so the `src/func_8002D014.c` committed two commits ago "only so the tree
+is clean, marked unverified" is now a verified claim. **FINAL: 685 bodies / 694 regions. Phase: 602 →
+685 = +83.** D's lever is worth keeping:
+
+> **THE FRAME SIZE PINS A LOCAL AGGREGATE'S SIZE.** Written with a struct of exactly the twelve bytes the
+> code touches, the row is 76 B at the right instruction count with **four differing bytes that are all
+> frame arithmetic** (`addiu sp,sp,-40` / `sw ra,32(sp)` against the original's `-48` / `40(sp)`).
+> A 16-byte struct still gives 40. A **24-byte** struct — twelve used plus twelve tail — gives the
+> original's 48 and matches byte-exact.
+
+**So size a local aggregate from the FRAME, not from the fields you can see.** And D flagged the mirror of
+the per-row epilogue trap on the same row: `0x8002D014` must NOT use `--fill-epilogue`, unlike the five
+other rows of D's that need it.
+
+### MY ATTRIBUTION ERROR, found by worker A
+
+**In the stop messages I credited worker D's rows to A and to B.** The four rows `0x80100334`,
+`0x800FBDC0`, `0x800F88F0`, `0x80042DD4` are **D's**, and `0x80102F58` is D's; I wrote them into the
+messages addressed to A and B. **Worker A caught it and said so plainly** — and it is exactly the kind of
+error that corrupts a handover, because it changes two workers' counts in opposite directions and the
+ledger cannot detect it (the ledger tracks addresses, not authorship).
+
+**Recorded because the count in this file was already wrong once this session** (a commit message stating
+672/681 when the truth was 679/688, corrected in its own commit). Authorship and counts are the two
+things a later session trusts without re-deriving, and I have now been wrong about both. **The authority
+for bodies is `config/regions.tsv` counted at the merge; the authority for authorship is each worker's
+own `claims.tsv` as the checksum in their own directory.**