/reports/attempts-coverage.md · raw
Attempts ledger: coverage check (2026-10-14)
This is an independent check of coord/attempts.tsv. I did not rely on the backfill reports. I found each attempt in the original records and then looked it up in the ledger.
Counts
- Ledger: 327 rows after this check (239 Claude, 88 Astra).
lintpasses with 0 errors and 0 warnings, andtools/test_attempts.pypasses 26/26. - Spot check: 25 attempts taken from the records (17 Claude, 8 Astra; 2026-10-10 to 10-14; kills, negatives, parks, cryptids, refutations).
- 18 were in the ledger with the right outcome.
- 7 were missing. I added 8 rows (ledger lines 321-328).
- Whole-folder checks:
- Every
machines/*/folder (94) has at least one row. - Every KILLED machine (31) has its own kill row, and every kill row is a KILLED machine.
- Rows touch all 740 lines.
The 25 checks
| # | Attempt (source) | Agent | In ledger? |
|---|---|---|---|
| 1 | 156 W4-4 observe stage: rules, Eqv_v3 TM23, kill #11 template (notes) | C | missing, added at line 321 |
| 2 | 11: FractalType7 TM20 S1+Inc template transfer, 0/20 (notes) | C | missing, added at 322 |
| 3 | 644 W4-8 seams stage, stop rule fired (notes, decisions) | C | yes |
| 4 | 693 stage 2: CTL62 sweeps, about 190 runs, 0 true (notes) | C | yes |
| 5 | 10 W6-1: sobc1_round_invariant fails because of the fossil (notes) |
C | missing, added at 323 |
| 6 | 22-proof audit: 9 PASS, 13 CONCERN (coord/audits) | A | yes (outcome kill; see gap 1) |
| 7 | Bouncer preflight 102/328/463/638/641 (astra-log) | A | yes |
| 8 | K42 stage 2: 5^k port, cryptid family, parked (reports/k42) | C | yes |
| 9 | CB3 closure explorer, stage 1, running now (reports/cb3/explorer.md) | C | missing, added at 324 as in-progress |
| 10 | coqchk uncapped on W4-3b (512) and W4-3c (588) (reports/coqchk-uncapped) | C | missing: line 133 filed all three under W4-3's string, so query 512 and query 588 missed it. Added at 325 and 326 |
| 11 | BEC sweep of 21 lines (reports/bec-sweep) | C | yes |
| 12 | 593 follow-up: 6 local modules, potentials, obstruction (coord/audits) | A | yes |
| 13 | 443 (#13): edge lemma (E'), SFT/CPS closures (overnight ledger, notes) | C | yes |
| 14 | CB1 executor on 7 lines (reports/halting-search/cb1) | C | yes |
| 15 | qsim deep runs on 19 closed-rule Collatz lines (reports/halting-search/qsim) | C | yes |
| 16 | 318/483 primitive grafts, 210 + 61 trials (coord/audits) | A | yes |
| 17 | 646 ghost-history entry and loss reuse (coord/audits) | A | yes |
| 18 | RWLAccHalt reuse, plain and cached, on 32/262/616/110/319/340/553 | A | yes |
| 19 | Short-opening CTL: 48 calls, 18 false, 30 timeouts (astra-log) | A | yes (rows 63-65 add up to 48) |
| 20 | 64: Sk1like_TC TM1 twin port refuted (notes) | C | yes |
| 21 | W5-2 (409): TM10 port, cryptid, halting hunt (notes) | C | yes |
| 22 | 530: rejected in wave 6, 1 transition from CounterClass2 (wave6 ranking) | C | missing, added at 327 |
| 23 | 147/389 synthesis pilot: 22 calls, 18 UNSAT, 4 timeouts (astra/independent) | A | yes (rows 78-81 add up) |
| 24 | R9 (625): niced simulation to 1e12 (overnight ledger) | C | yes |
| 25 | R5 (269) observe stage: rules, TM204 template "does not apply" (notes) | C | missing, added at 328 |
Query tests
I ran tools/attempts.py query on lines 693, 409 (W5-2), 10 (W6-1), 64 (K42) and 593, plus 147 and 493. Each line was queried four ways:
- by line number;
- by the exact string;
- by a renamed and mirrored variant;
- by the string of the same table started from another state.
All four returned the same hit set for every machine (11, 6, 7, 9, 9 and 15 rows). The hits were the expected ones: - 693 finds its 4 machine rows and the Bell eats counter class row. - 64 finds the K42 class rows and the refuted ports of 124 and 305 as "same table, other start". - 493 finds 147's SAT rows and Astra's 147/493 table rows.
Known gaps
- No outcome for a useful result that is not a kill.
- Astra is holding 3 rows from
astra/three-directions-20261014: the width-12 block cache (2.32-2.95x faster), 8 synthetic certificates, and 15 local theorems. The cache row also carries a negative: adding traversal costs 6-34% more time. - Audits that only confirm an existing kill are filed askill(lines 21, 22, 95, 133, 325, 326). - Both need a decision on the outcome set. - Early stages are often folded into later rows. Observe stages and template checks often have no row of their own (items 1, 2, 5 and 25). A template check that says "does not apply" is the negative most worth keeping, and I checked only these four machines. The other parked machines may have more of the same.
- One row can cover several machines. Line 133 filed three machines under one string (item 10). Rows like that need one row per machine, or a
listrow with explicit lines. - A string that starts with
---breaks the query. This can only happen with an other-start string.query '---1LB_...'fails in argparse;query -- '---1LB_...'works. - Open rows.
- Line 29's uncapped coqchk is still running.
- Line 10 is waiting on Dennis's decision.
- CB3 explorer, line 324: its stage must close the row with the same target,
closure-explorerandclaude. - Wave 7 rows are being added live (lines 317-320). - Astra's review is not finished. Astra has not yet reviewed the content of its 69 backfilled rows (message of 18:55).