phase12: credit-outage recovery (10 bodies recovered, 635/644) and the negatives re-dispatch

2026-09-24 18:16:47 -04:00
parent 6da840f3ab
commit e074d947fb
1 changed files with 73 additions and 0 deletions
@@ -551,3 +551,76 @@ invariant is a *diagnostic*, not a gate. **To be reconciled at the phase close**
 census is available and the negatives-index reconciliation is done anyway — the two are probably the
 same question, since the divergence appeared on the first merge whose additions were all
 negatives-index rows that no earlier merge had registered.
+
+---
+
+## THE CREDIT OUTAGE AND THE RECOVERY (2026-09-24 ~21:45)
+
+**All four workers stopped mid-flight when the account ran out of credit.** The developer re-upped
+and instructed me to continue. The workers did **not** die — all four came back **idle** — so this
+was a resumption, not a roster rebuild, and §4.0's "spawn fresh" rule did not apply.
+
+### The recovery: 10 verified bodies out of staging that no report had yet announced
+
+| | count |
+|---|---|
+| `src/` files produced this phase and still untracked | 11 |
+| distinct addresses claimed in ANY staging file | **29** |
+| `src/` files claimed by **nothing** (true orphans) | **0** |
+| claims whose source file is missing | **0** |
+
+**All 29 claimed rows were re-verified by me from fresh `--work` directories with `--symbols` (and
+`--cc1 gcc-2.8.1-psx/cc1` for the four `cc1bin` rows) BEFORE merging: 29 of 29
+`differing_bytes=0 result=MATCH`, exit 0.** One combined `apply --skip-registered`: 19 skipped as
+already merged, **10 added**, 0 rejected.
+
+```
+sf3_match gate   c_regions=644  differing_bytes=0  result=MATCH
+                 sha1 e173426c157384ebf1b6caf8c6fea18a85a14af9  (unchanged)
+make check       exit 0        extents-verify  regions=644 disagreements=0 AGREE
+registry audit   644 rows, 635 distinct bodies, 0 missing, 4 carry cc1bin
+```
+
+**A defect in my own audit, recorded because it manufactured a false alarm.** My first orphan check
+globbed `.run/p12/w-*/claims.tsv` and reported `src/func_801008DC.c` as unclaimed. It is claimed —
+in `claims-altcc1.tsv`, a **deliberately separate pool** because those rows need the `cc1bin`
+override. **The audit was wrong, not the worker.** A `src/`-against-claims audit must glob *every*
+claims file the charter defines, not the default one, or it invents orphans and the next worker
+investigates a phantom.
+
+### Position: **602 → 635 bodies = +33, past the cycle-1 checkpoint (+30)**
+
+| worker | bodies | the signal |
+|---|---|---|
+| D | **13** | 5 fresh-band + 4 Goal-B + 2 twins + 2 GTE |
+| C | **9** | 4 default + 3 `cc1bin` + 2 GTE |
+| B | **6** | 5 in ONE translation unit + 1 adjacency |
+| A | **5** | **adjacency, 3-for-3 across three bands** + 2 |
+
+### The re-dispatch: all four workers onto the negatives (developer's decision)
+
+The developer chose **"keep 750, re-dispatch to the negatives"**. The rationale is entirely measured:
+
+| source | cost per body |
+|---|---|
+| worker B's fresh band | **4.2 attempts per match** (and it hit the leading indicator twice) |
+| worker A's fresh band | 4-13 spellings, bimodal; **all three failures were allocation/block-order** |
+| worker D's **classified negatives** | **4 bodies from 4 attempts at ONE SPELLING each** |
+| worker C's negatives | **15 rows → 9 matched, 0 abandoned without a mechanism** |
+
+**The mechanism is already named in those rows, so the per-row reading cost is pre-paid** — which is
+exactly what worker A asked for when it requested a cheap key away from the allocation/block-order
+class and noted that finding 116 records the naive proxy failing. The negatives index *is* that key.
+
+New pools, split by `crc32(address) % 4` so membership cannot move:
+
+| worker | pool | rows | prio 1 (named mechanism) | prio 2 (`cc1bin`) | prio 3 (no mechanism) |
+|---|---|---|---|---|---|
+| A | `negatives-a-pool.tsv` | 47 | 19 | 0 | 28 |
+| B | `negatives-b-pool.tsv` | 45 | 23 | 2 | 20 |
+| C | `negatives-c-pool.tsv` | 43 | 31 | 1 | 11 |
+| D | `negatives-d-pool.tsv` | 41 | 19 | 1 | 21 |
+
+**176 unregistered negatives rows; 92 carry a named mechanism; 8 are in the multi-exit `cc1bin`
+class.** D was told to take its **three in-flight `wip` twins first** (`0x800161E0`, `0x80016224`,
+`0x8008F478`) before the pool, because twins have closed at one spelling all phase.