v0.2.65

Summary

• add a public validate_maintainer_rerun_attestation API
• add anvil leaderboard validate-rerun for explicit maintainer-side attestation checks
• document validate-rerun in the leaderboard and CLI guides

Verification

• CI passed on PR #209 and PR #210
• local gate passed: ruff format --check, ruff check, ty check, pytest -q

v0.2.64

Summary

• expose maintainer rerun evidence fields in generated leaderboard rows, JSON, and CSV
• align the Hugging Face leaderboard renderer with maintainer rerun evidence aliases
• update leaderboard documentation and checked-in index schema

Verification

• CI passed on PR #206 and PR #207
• local gate passed: ruff format --check, ruff check, ty check, pytest -q

Agent Anvil v0.2.63

Highlights

• Verify maintainer rerun GitHub Actions runs when leaderboard build uses --github-run.
• Reject failed rerun runs, repository mismatches, and SHA mismatches.
• Document that --github-run verifies maintainer rerun attestations too.

Verification

• CI passed on PR #203 and PR #204.

Agent Anvil v0.2.62

Highlights

• Add anvil leaderboard attest-rerun for generating maintainer rerun attestations.
• Validate original and rerun leaderboard submissions before writing maintainer_rerun overlays.
• Document the generated maintainer rerun attestation workflow.

Verification

• CI passed on PR #200 and PR #201.

Agent Anvil v0.2.61

Highlights

• Add typed maintainer rerun attestations for leaderboard rows.
• Add leaderboard build overlay support via --maintainer-reruns.
• Validate attestation evidence hash, headline metrics, agent/benchmark identity, and GitHub Actions rerun URL.

Verification

• CI passed on PR #197 and PR #198.

Agent Anvil v0.2.60

Highlights

• Reject direct leaderboard submissions that self-claim maintainer_rerun trust.
• Reserve maintainer_rerun labels for maintainer-side rerun attestations.
• Update leaderboard workflow examples to v0.2.60.

Verification

• CI passed on PR #194 and PR #195.

Agent Anvil v0.2.59

Summary

• Reject versioned results.json files when summary trial counts or pass rate do not match grades.
• Document summary-vs-grades validation for persisted results artifacts.
• Add parametrized storage tests for tampered aggregate summaries.

Verification

• PR #191 CI passed: Demo Eval, Test Python 3.12, Test Python 3.14.
• Release PR #192 CI passed: Demo Eval, Test Python 3.12, Test Python 3.14.

Agent Anvil v0.2.58

Summary

• Validate that results.grades trace_path values point back to matching run traces.
• Reject missing, mismatched, or path-traversal grade trace paths during run validation.
• Surface grade trace path verification in artifact trust summaries.

Verification

• PR #188 CI passed: Demo Eval, Test Python 3.12, Test Python 3.14.
• Release PR #189 CI passed: Demo Eval, Test Python 3.12, Test Python 3.14.

Agent Anvil v0.2.57

Summary

• Add opt-in strict trace step validation for trace and run artifacts.
• Keep default trace loading permissive for legacy/custom observer events.
• Document strict validation for CI artifact consumers.

Verification

• PR #185 CI passed: Demo Eval, Test Python 3.12, Test Python 3.14.
• Release PR #186 CI passed: Demo Eval, Test Python 3.12, Test Python 3.14.

Agent Anvil v0.2.56

Summary

• Share the assertion JSONPath grammar and resolver in anvil.jsonpath.
• Make scenario validation and deterministic grading use one parser contract.
• Add contract tests for supported, unsupported, missing, and array-indexed paths.

Verification

• PR #182 CI passed: Demo Eval, Test Python 3.12, Test Python 3.14.
• Release PR #183 CI passed: Demo Eval, Test Python 3.12, Test Python 3.14.