phase11: merge 18 + cookbook 87-88 — 530 bodies / 539 regions

Worker A's 0x8009C904: 548 bytes, matched on the FIRST spelling. Cumulative evidence for the
redundancy ranker is now 548B/1st, 248B/3 spellings, 1232B/2, 700B/3 -- against 176B/9
failures for a tie-break-dense row.

Cookbook 87 is the important negative: cc1 does NOT unroll a constant-trip-count loop (the
for-loop form is 212 B against the original's 780 B), so whenever an original is unrolled the
SOURCE is unrolled too. That explains why the >800 B band is full of cheap rows -- those
bodies are unrolled in the source, and an unrolled repetitive body is exactly what the
ranker scores highest.

config/regions.tsv

@@ -358,6 +358,7 @@
 0x80099E14	0x80099E34	src/func_80099E14.c
 0x8009AC08	0x8009AC28	src/func_8009AC08.c
 0x8009B464	0x8009B4A0	src/func_8009B464.c
+0x8009C904	0x8009CB28	src/func_8009C904.c
 0x8009D8A0	0x8009D8E0	src/func_8009D8A0.c
 0x8009E8D0	0x8009E95C	src/func_8009E8D0.c
 0x8009F0E8	0x8009F120	src/func_8009F0E8.c

docs/MATCHING_COOKBOOK.md

@@ -1387,3 +1387,28 @@ return block reached with a `j` (+8 bytes), whereas the original branches each g
 to the shared epilogue with `move v0,zero` in each delay slot. A named-result-local spelling gets
 to 92 bytes but puts the local in v1 instead of v0. Named untried lever: a `goto` to a shared
 return label (finding 43, trigger 3) to defeat the cross-jump.
+
+### 87. cc1 does NOT unroll a constant-trip-count loop (worker A)
+
+Worker A's `0x800BFF20` (780 B): the `for (i = 0; i < 6; i++)` form is **212 B** against the
+original's 780 B. So **whenever an original is unrolled, the SOURCE is unrolled too** — do not
+write a loop and hope cc1 unrolls it. This is the single most useful negative in the phase for
+large bodies, because it explains at a stroke why the >800 B band is full of cheap rows: those
+bodies are unrolled *in the source*, and an unrolled repetitive body is exactly what the
+redundancy ranker scores highest.
+
+Same row, second negative: an explicit two-base spelling (`p = q + 12532` for the last 8 checks)
+is 784 B — one instruction extra, same allocation mismatch.
+
+### 88. Redundancy-rank evidence, cumulative
+
+| row | size | rep-density | outcome |
+|---|---|---|---|
+| `0x8009C904` | **548 B** | 2.26 | **FIRST spelling → MATCH** |
+| `0x8009F798` | 248 B | 2.10 | 3 spellings → MATCH |
+| `0x8010AF50` | **1232 B** | repetitive | 2 spellings → MATCH (worker D) |
+| `0x8006BC74` | 700 B | repetitive | 3 spellings → MATCH (worker D) |
+| `0x80017C6C` | 176 B | tie-break-dense | **9 spellings, 42 bytes, classified** |
+
+Worker A, on adopting the ranker: *"I have not attempted a tie-break-dense row since adopting the
+ranker and I do not intend to."* That is the correct use of the measurement.