# Paper 1 recovery

Status: C1 recovery execution and F1 resource probe completed; final science acceptance pending.

C1 resumes only after the old job is absent globally. F1 is a resource and numerical-closure probe using old fitted geometry, not a publication product. Background and removal decisions remain with the user.

Code/configuration: plan.json. Activate syndiff; submit generated job.sub files to an explicitly recorded scheduler after global inventory. No automatic submission.

Conclusion: operational recovery completed; this does not certify the final paper products.

## Execution update 2026-10-02T18:51:45-04:00

Old C1 job 1586.0 on plscience4 was verified and removed; all OS4 products retained. New jobs submitted on plscience3: C1 317, F1_probe 318. Code pin ef21b95c5d1f5e4cee9ee2931cf0c95cc4c8f656. Global queue checks include plscience4. Old contribution arrays remain held.

Validation: 84 passed, 4 deselected, 1 expected cross-projection failure. OS4 inventory: 944/944; OS1 initial inventory: 936/942. The failed-final-task cell 2130.004 also passes ZIP integrity.

Run data: C1/condor/job.*, C1/runs/, F1_probe/condor/job.*, F1_probe/probe_results.json. No scientific result or recovery-complete claim yet.

### Recovery is running 2026-10-02T18:55:00-04:00

C1 attempt 317 exited 1 before any recovery writes because compute-node credentials cannot query every scheduler. The replacement 320 (plscience3 scheduler, plscience13 worker) uses a fresh authenticated global-query receipt produced on the submit host, bound to the exact code SHA/output root and expiring after 15 minutes. C1 code is isolated at 97498e5 on codex/paper1-c1-recovery-20261002; no running F1 source was edited. F1 job 318 remains on ef21b95 and is running on plscience13. Both branches are pushed. Logs for the C1 replacement: C1/condor_v2/.

All original held f03 arrays and their waiting launchers remain untouched; no final-fitting job was submitted.

## Old arrays retired 2026-10-02T19:00:21-04:00

All 100 verified held tasks in plscience4 clusters 1572/1611/1612/1627/1650 were removed after archiving classads, configs, submit files, logs, launcher scripts and product inventories. A post-removal query returns no matching jobs. 7483 original per-band product files were checked: all retain their original size and mtime. No matching old launcher existed on plscience3 or plscience4; no unrelated processes were killed.

F1 probe 318 exited 0 and all three sequential contributions passed the existing closure gate. Peak process RSS was 23,729,208 KiB (about 22.6 GiB); the three chosen cells had no cross-projection correction. This supports single-worker follow-up, not a claim that all cells or multi-worker jobs fit the same limit. Broad replacement submission remains on hold until preparation choices and boundary coverage are settled. C1 recovery 320 is separate and continues.

## Latest status 2026-10-02T21:52:48-04:00

C1 job 320 exited 0. All 942 OS1 entries resolved; native and F4 output files are present, and recovery_result.json records the produced template. F1 job 318 also exited 0. Final image inspection/science acceptance remains a separate gate. Current agent inventory: agents.html.

## Decision review 2026-10-02T23:51:06-04:00

Reviewed completed corrected neighbour A/B pilot (inconclusive), ongoing C4 ledger DAG, installed extraction commands and corrected-cohort tables, and failed temporal phase-null calibration. Recommendation: one better-powered intact-core A/B comparison before final five-field fitting; complete extraction acceptance in parallel. No science jobs changed. Illustrated decision review: decision.html; evidence snapshots and hashes: decision_evidence.json.