The first extracted-archive audit found that the old Job136 training split manifest was not bundled. Offline inference had already run, but the split-separation audit correctly failed with FileNotFoundError. Its full report is retained as packaging-first-failure.json. Added the unchanged136/data/manifest.json dependency; no model, score, input or assertion was changed. The archive is then rebuilt and all extracted checks repeated. A preceding publication preflight also refused the stale failed audit summary left during the editorial hash correction; that refusal is retained at preflight-stale-audit-summary.json.
