[rmcdaniel]

Closes #228.

Change

durable_workflow.history_bundle_verify now fails a retention-pruned bundle (workflow.details_state == "pruned") with the history.details_pruned error, matching PHP Core's BundleIntegrityVerifier.checkRetainedDetails from workflow#980 (merged 05ffc5794f8d554c4fb10c29a62da60e1a9e946c):

• Finding: the same rule ID, severity, message text (including the optional at <details_pruned_at>), path workflow.details_state, and context {details_pruned_at, history_complete}. A missing, empty or non-string timestamp becomes null, as in PHP's stringValue.
• Unchanged: bundles with details_state: "retained", and older exports without the field. Only the exact string "pruned" triggers the error.
• Dedupe key: the pruned bundle's extended dedupe_key (…:details-pruned:<timestamp>) is accepted as exported. Python performs no dedupe-key check, so there's no new rule.
• Simulation and CLIs: because the verdict is failed, simulate_bundles and durable-workflow-replay-verify --simulate-bundles block the bundle (block_and_investigate), and durable-workflow-history-bundle-verify exits nonzero. Bundle simulation doesn't execute workflow code; this is the integrity layer.

Regression fixture

tests/fixtures/history_bundles/retention_pruned/pruned-bundle.json holds the exact bytes from #228: a real TestGreetingWorkflow run, completed, pruned and exported by the workflow#980 head 2e562997. The merge 05ffc579 kept that head's files unchanged (the compare shows no file diff). File sha256 is 57f12ffb…bbf; the canonical checksum 8fa57d07…01ac still matches.

PHP's corpus entry tests/Fixtures/V2/ReplayRegression/retention-pruned-run-export.json (sha256 7016af05…) is a scenario descriptor that generates pruned exports at test time and carries no bundle bytes. provenance.json records that, along with PHP's expected verification result. The test asserts that Python's findings equal PHP's expected findings exactly.

Evidence (local, macOS, Python 3.12.4)

• New tests: tests/test_history_bundle_pruned.py has 14 tests: fixture bytes untouched, exact PHP parity, dedupe key, a pruned marker without a timestamp, retained and absent unchanged, near-miss markers, and simulation plus both CLIs.
• Counterfactual: with main's verifier, the 7 detection tests fail and the 7 unchanged-behavior controls pass.
• Related suites: history-bundle and replay tests, 355 passed.
• Lint and types: ruff and mypy pass.
• Full suite: running; result posted below.

Expected to conflict with #229 only in CHANGELOG.md (both add under Unreleased).

🤖 Generated with Claude Code

[rmcdaniel]

Exact head fb97191c20bb8382b611573826dc65230d21f2ff: all 21 hosted checks pass. Clean local full unit suite on this branch: 3,730 passed, 4 skipped, with the 2 bash-3.2 scanner tests deselected. An earlier local run was discarded because the branch changed during it. Ready for review. Only CHANGELOG.md should conflict with #229 (both add under Unreleased); I'll resolve it on whichever branch lands second, if you want me to.

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud

[rmcdaniel]

Merged main (8b95f28, which includes #229) normally into this branch. The only conflict was CHANGELOG Unreleased, now listing Added (#168) then Fixed (#228). Exact head 7d840977fa3d32b976dc5b44a7dca14b78266e46: all 21 hosted checks pass; local full unit suite 3,756 passed / 4 skipped (2 bash-3.2 scanner tests deselected).