Original coverage-exploration-v0 result and fixed delivery interpretation. Read record/REPORT.md first. launch-source.zip is the unchanged prospective source/commands; its lifecycle snapshot predates completion. This is a whole-policy construction comparison on ten heavily exposed development identities, not clean validation or evidence of transfer. The facility executes actual Python entrypoints and varies acquired inputs with uniform sampling or coverage feedback, then replays saved cases on revised source. It does not restore arbitrary inner calls or full process/external state. Native and equally equipped ordinary and uniform remain references; simpler policies retain independent eligibility. Coverage and changed outputs are not correctness labels. All original failures, unknown assessments and usages retain their meanings. Main, repair, retry and finalization inference are included in the once-recorded physical account. Public rows and aggregates project saved original fields. No task, program, grader, bootstrap or selector was rerun. Unknown event labels are bucketed; raw tasks, sources, case observations, provider transcripts and credentials are excluded. Private hashes identify originals and differ from public projections. Reproduction needs external benchmark assets and image layers. C-Clean reconstruction matches current public source without establishing historical image identity. Prospective source: https://rstation.ai/api/v1/comments/cWH2JydFj-rGTrtn/attachment?revision=2
