0.8.1: final review/operation regression suite: 84 tests passed in 89.871s. Additional SAME/recurring-payment rejection regression: 1 test passed in 1.533s. Related dedupe/candidate/coding/queue/research/report suite ran 141 cases in 93.131s: 140 passed; the new native-receipt-persistence integration case exposed lost diagnostic metadata, was corrected to retain the check on the execution/test receipt, and passed in the final 84-case suite. Tests use fake model providers/executors; they do not establish scientific validity.
Two actual networkless Docker executions passed with the retained real EPUB: native RSS creation and independent pointer readback, missing title/guid/link rejection and malformed XML rejection. Old actual JSON helper/source-only probe rejected, disjoint immutable source ranges resolved, stale component flag cleared in a copy. Original checkpoint unchanged; paid API calls 0; external actions 0. RSS reference is human-authored, not evidence of autonomous future model success.
