HEIMEL_Governed_Effects_Submission_Draft_v0_5_canonical

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

HEIMEL Governed Effects — Canonical Evidence v4 This record contains the frozen canonical evidence package for the HEIMEL governed-effects evaluation conducted on 30 September 2026. The evaluation tests structural authorization for consequence-bearing AI-agent actions, including fresh authority at consequence time, exact-effect binding, deterministic non-promotion, exclusive governed effect paths, atomic single-use permit consumption, uncertainty-preserving recovery, and evidence binding. A concurrency falsification identified a TOCTOU defect in an earlier permit-consumption implementation: two concurrent attempts using one single-use permit could both cross the consequence boundary. The defect was traced to non-atomic permit consumption, corrected by moving an atomic durable claim before consequence reachability, elevated to the HEIMEL atomic_single_use_consequence invariant, and re-tested against canonical upstream commits. Canonical post-remediation results: REHT 339 passed / 1 skipped; focused concurrent replay 9/9 passed; Gateway 245/245 passed; HEIMEL 30/30 passed; RACS 231/231 passed; governed-effects adversarial gate 78/78 passed. Observed duplicate effects: 0. Observed unauthorized effects: 0. The package includes machine-readable evidence, the validation report, reproduction manifest, raw test logs, and SHA-256 checksums. Claims are intentionally bounded. The evidence does not establish arbitrary distributed exactly-once execution, multi-host consensus, or automatic resolution of indeterminate external outcomes.