0.8.0: 144 related regressions passed (109.106s). 17 new mechanism-quality tests passed (17.421s), including actual SDK tool routing with fake provider/executor, helper classification, source reuse, independent failed assertion probes, input/implementation immutability, print-only rejection, byte/attempt/runtime/cleanup guards and bounded continuation after repeated reuse. A subsequent focused run also passed coding/submission checks after the reuse-retention change.
Actual final captured Open Library extractor was run in isolated networkless Docker with a reviewer-authored assertion challenge: exit=1 because the actual decoder accepted the tampered checksum, implementation unchanged, cleanup confirmed. Old SUBCOMPONENT approval refused. Dedupe/reuse matched the two real Open Library records.
Source checkpoint unchanged; paid API calls 0. This validates the framework behavior for these cases, not arbitrary future model judgment or the original candidate mechanism.
