docs: mark the triage-ladder spec BUILT with its three corrections; accelerator #16

next-session-triage-ladder.md was still written as a to-build spec. It now leads
with the shipped status, the acceptance numbers, and the three things the spec got
wrong (the '32 free banks' were 0/28; the autodecl arm is worse in-tree than the
raw draft; PRE and POST cannot be the same pass because residual_rules_b needs a
draft), plus the one found by building it — never classify on a moving tree.

accelerators #16: a 'verified, just bank it' claim must name the compilation it
survived. Day-one kit material for a new decomp: any per-function oracle compiles
in isolation, every real bank compiles in context.

docs/accelerators.md

@@ -357,3 +357,29 @@ So at every session/phase close, review the tooling against the idioms learned t
 *which scanner's denominator just got wider?* — the answer converts new knowledge into free banks.
 S68's own §332 sweep is the worked example: one idiom review, ten functions / 1,027 instructions
 reclassified from "hard" to "not bankable at all", and one in-flight escalation stopped mid-spend.
+
+## #16 — A "VERIFIED, JUST BANK IT" CLAIM MUST NAME THE COMPILATION IT SURVIVED (P31 S69)
+
+**What happened.** A session closed with "32 FREE BANKS ARE WAITING", 10 of them personally verified
+with `match_one` at closeness 0. The next session gated them: **0 of 28**. Not one failure was a
+codegen miss — every one was a declaration conflict inside the real translation unit.
+
+**Why the claim looked true.** `match_one` compiles the draft ALONE, with its own externs. The TU the
+function must actually live in already carries a forward declaration written for a call site, and the
+draft's real signature conflicts with it. **Closeness 0 is a statement about the BODY. Banking is a
+statement about the TU.** They are different oracles and only one of them was consulted.
+
+**Why this belongs in a NEW decomp's day-one kit.** The trap is structural, not local to this project:
+any per-function matching oracle compiles in isolation, and every real bank compiles in context. So:
+
+* every "already matches" claim ships with **which compilation it survived** (isolated / whole-TU /
+  whole-binary), the way every rate ships with its denominator (R41);
+* the isolated oracle's output is a **GATE-FIRST candidate**, never a bank;
+* budget for the integration lever, because it is the one that actually pays: 8 of those 28 banked the
+  moment the missing lever existed (the self-caller cast, cookbook §378), including one in `main` —
+  the project's most expensive binary — for **zero agent tokens**.
+
+**The corollary that cost the most time.** Fixing the first declaration error only REVEALS the second:
+no-proto the conflicting decl and the draft's own definition becomes the prototype in scope, so the
+same call site now fails with `too few arguments`. A class can look dead after one fix and be three
+mechanical steps from banking. Drive the chain to a byte verdict before writing anything off.

docs/next-session-triage-ladder.md

@@ -1,4 +1,32 @@
-# NEXT SESSION — build the TRIAGE LADDER (spec, written S68 with full context)
+# THE TRIAGE LADDER — **BUILT AND SHIPPED (P31 S69)**
+
+> **Status: DONE.** The tool is `tools/triage_ladder.py`. It is wired into `wave_args.py` (draw time)
+> and `tools/workflows/escalate_fable.js` (escalation time), documented in `docs/SETUP.md` and
+> `docs/wave-playbook.md` §4b, and its acceptance harness is green over the whole corpus:
+> **false-skip 0/1367 open stubs · recall 426/426 matched · wall tier fires on exactly the 10
+> enumerated walls (0 extra, 0 missing)**. Run `tools/triage_ladder.py --acceptance` to re-verify.
+>
+> **Three things in the spec below were WRONG, and the corrections are the valuable part:**
+>
+> 1. **"32 FREE BANKS ARE WAITING" was not free.** 4 had already banked; the other 28 gated **0/28**.
+>    Every failure was a declaration conflict inside the real TU. `match_one` compiles the draft
+>    ALONE — a standalone closeness of 0 proves the BODY and says nothing about the TU accepting the
+>    SIGNATURE (cookbook **§376**). 8 of the 28 banked once the missing lever existed (**§378**, the
+>    self-caller cast, `tools/cast_self_callers.py`); the remaining 20 are on a named ledger at
+>    `.run/S69_class376_ledger.json`.
+> 2. **The autodecl arm is WORSE in-tree than the raw draft** — the `extern` it adds to satisfy the
+>    standalone probe is a second conflicting declaration. Gate the raw draft.
+> 3. **The spec conflated PRE and POST.** `residual_rules_b` needs a draft and runs `match_one`, so
+>    it cannot run at DRAW time. Only the target-side tiers (BANKED · WALL-332 · PARKED) can — and
+>    those are the ones that save a whole agent rather than one iteration. The tool splits them.
+>
+> One more, added by building it: **the ladder must not classify on a moving tree.** A merging gate
+> makes the stub oracle wrong in BOTH directions (§377), which manufactures the exact false skip the
+> ladder exists to prevent. It refuses unless the tree is quiescent.
+
+---
+
+## The original spec, kept as written (S68)
 
 > Written deliberately so a session with NONE of S68's context can build this. The companion tool,
 > `tools/neighbor_ref.py` (retrieval), is already BUILT and committed — this is the other half.