Skip to content

Add stale-plan and authority-provenance regressions with bounded research comparison - #20

Draft
altrudev wants to merge 1 commit into
mainfrom
research/assurance-regressions-20260924
Draft

altrudev wants to merge 1 commit into
mainfrom
research/assurance-regressions-20260924

Conversation

@altrudev

Copy link
Copy Markdown
Owner

Agent Replay needs regression coverage for fresh memory paired with an obsolete plan and for untrusted instructions repeated through summaries, retries and handoffs.

This PR adds six synthetic parametrized cases, including replanning and explicit-grant controls, and verifies that absent normative assertions remain unassessed. It also records a pinned comparison of ATC with EBL-Core, with source attribution and explicit chronology limits.

Validation: 19 Replay reconstruction tests passed (six new, 13 existing); 14 existing ATC verifier tests passed separately. The published Git tree matches the locally tested tree exactly.

No production code changes. These tests cover reconstruction of supplied evidence, not runtime enforcement or authentication. EBL-Core source execution was not reproduced; the comparison is based on the paper and inspected ATC code. Existing evidence-model and provenance PRs remain untouched. No GitHub Actions were used.

Created for Valentyn Rukhaylo / Altru.dev.

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant