diff --git a/docs/matching-cookbook.md b/docs/matching-cookbook.md
index 050ea470a..cea293352 100644
--- a/docs/matching-cookbook.md
+++ b/docs/matching-cookbook.md
@@ -2504,6 +2504,23 @@ crack (the type-using triage cracks AND the Fable5 cores) flows through to reach
 bodies, no reconcile) still template via plain `--hseq` (Task 5: 399 banked). Key: reconcile PER SIBLING from
 the RAW draft, never remap the ov077-reconciled body.
 
+### §31-triage — R17 applies to CODEGEN residuals, never to a compile ERROR (Phase 26 session 7, Drew asked)
+
+Before reaching for `tools/reference/gcc-2.7.2/` or the `docs/gcc-2.7.2-map/`, classify the failure:
+
+| Symptom | Cause | Tool |
+|---|---|---|
+| The build SUCCEEDS but the bytes differ | a codegen decision (regalloc / sched / cross-jump / CSE / loop) | **R17** — read the pass, or gdb-on-cc1 (§45-B) |
+| The build FAILS to compile (`conflicting types`, `undeclared`, `parse error`) | plain C89 semantics — gcc is CORRECT | **our tooling.** Reading gcc source tells you nothing |
+
+Byte-example: the `func_8015AE2C` ×133 sweep blocker is `conflicting types for D_801812A4` — two incompatible
+file-scope decls of one identifier in one TU. That is not a gcc quirk; `reconcile_decls` was picking a
+**fleet-majority** canonical type instead of the type the TU can actually SEE (its §8b carried decl layer).
+A `loop.c`/`global.c` read would have been pure waste. Contrast `func_8017BEBC` (close=2, an `allocno`-priority
+tie in `global.c`): compiles fine, wrong bytes → exactly an R17/§45-B target.
+
+**Rule of thumb: "wrong BYTES" → read the compiler. "won't COMPILE" → read our Python.**
+
 ## §41 — The DEF-SIDE canonical-sig wall: mechanically banking a drafted giant past `conflicting types` (Phase 25 T5b batch-2, 2026-07-09; `tools/canon_sig_reconcile.py`, byte-proven on `func_8013B274`)
 
 **The wall (dominant for GIANTS — ~universal, vs ~35% clean-bank for small fns):** a drafter writes an
diff --git a/phase-ends/CURRENT_PHASE.md b/phase-ends/CURRENT_PHASE.md
index d2e78d9b0..7ff998a7c 100644
--- a/phase-ends/CURRENT_PHASE.md
+++ b/phase-ends/CURRENT_PHASE.md
@@ -63,11 +63,23 @@ banked set + carve-ownership read from a `.s` splat never emits for matched fns
 carves** · `p16_permute` `hide_asm` eating GTE `#define`s (permuter silently no-op'd `0s` on ALL renderer
 drafts) + hardcoded to one overlay.
 
-### NEXT (in priority order)
+### NEXT (in priority order) — with the R17 routing call (Drew asked, 2026-07-13)
+> **Do NOT read the gcc-2.7.2 source for the sweep blocker.** R17 (read the compiler internals) is for
+> **codegen** residuals — regalloc / scheduling / cross-jump / CSE — i.e. things no C change reaches. The
+> sweep blocker is a **C FRONT-END diagnostic** (`conflicting types`: two incompatible file-scope decls of one
+> identifier in one TU). gcc is correctly rejecting plain C89; the bug is in OUR Python (`reconcile_decls`
+> picks a fleet-majority type instead of the type the TU can actually SEE). Reading `cse.c`/`loop.c`/`global.c`
+> would tell you nothing. **Triage rule: "the compiler produced the wrong BYTES" → R17. "the compiler refused
+> to COMPILE" → our tooling.**
+
 1. **Unblock the ×133 sweep** (above) → banks 562×133 ≈ 75K templatable ins. Highest ROI, well-diagnosed.
-2. **`func_8017BEBC` close=2** — permuter failed; needs a §45-B gdb-on-cc1 read of `allocno_live_length`
-   (the agent's own recommendation) or a targeted Fable5/Opus follow-up. ×113 reach.
-3. **`func_80178D40` close=39** — all in case 0x5C; agent named 4 concrete residuals. Cheap-Opus or permuter.
+   **Python fix, NOT a gcc read.**
+2. **`func_8017BEBC` close=2** (×113) — **THIS is where the gcc-source read belongs.** The permuter failed
+   (25 min, no close). The agent localized it to `global.c`'s allocno-priority TIE and named the exact move: a
+   §45-B **gdb-on-cc1 read of `allocno_live_length`** (the original's length quantization plausibly split the
+   tie). Two instructions from a 107K-ins bank — the best R17 target in the queue.
+3. **`func_80178D40` close=39** (×134) — all 39 in case 0x5C; agent named 4 concrete residuals (incl. gcc
+   peeling a loop iteration the original doesn't → likely `loop.c`). Cheap-Opus/permuter first, R17 if it walls.
 4. Then the rest of the 191 heavy jr cores. **R27: prompt Drew before any further Fable5.**
 
 ---