[anth-volk]

Fixes #1035

Same-Repository Draft PR Check

• This PR is a draft.
• This PR is from a branch in PolicyEngine/policyengine-sim-api, not a personal fork.

Changelog

• This PR adds a changelog fragment, changelog.d/1035.fixed.md, whose type is fixed.

Summary

• Stage12CachedSimulation.ensure() recalculated an incomplete Stage 12 cache file inside the except (TypeError, ValueError): block of its output-plan check, so any exception from run() or save(), including a Modal preemption's KeyboardInterrupt, carried the output-plan ValueError as __context__. The Public API's telemetry error report fingerprints an error by the first traceback in its stack, so these errors were grouped under the check instead of where they were raised.
• The except block now only logs the same WARNING and notes that the file is incomplete; the recalculation (incomplete outcome, run(), save(), clearing .py's in-memory cache) runs after the try statement.
• Everything else is unchanged: cache miss, missing output plan, cache hit, the WARNING message, the recorded outcomes, and exceptions other than TypeError or ValueError from the check, which still propagate.
• This is the same pattern Run the simulation outside the saved-output lookup handler in Simulation.ensure() policyengine.py#746 fixes in .py's Simulation.ensure(), which the cache-miss path here still goes through until the service moves to a .py release with that fix.

Merge order: merge after #1033 (policyengine-observability 3.0.7), so the service records preemptions at WARNING before any error's fingerprint changes. Deploys run one at a time in merge order, so merging after #1033 is enough.

Testing

• New tests in projects/policyengine-simulation-executor/tests/test_stage12_cache_runtime.py, run by node ID:
  • test_recalculation_error_carries_no_output_plan_error[run] and [save]: an error from run() or save() while recalculating an incomplete file is raised unchanged, with __context__ and __cause__ of None, and the outcome is incomplete.
  • test_preemption_during_recalculation_carries_no_output_plan_error: a KeyboardInterrupt from run() carries no __context__.
• Red (commit 26782a5, tests only): all 3 failed, each with the output-plan ValueError as __context__. Green (with the change): all 3 pass.
• Ruff 0.15.22 (the version CI pins): format --check and check pass on both changed files. Pyright was not run locally; CI runs it.
• No other tests were run locally.

Review

Self-review by the authoring session; findings only, no review disposition.

• Round 1 (a fresh reviewer given the diff and Recalculate an incomplete Stage 12 cache file outside the output-plan check's except block #1035): no bugs found. It checked every path through ensure() against main, confirmed the flag can't be unbound, and ran the 3 tests against main's module, where all 3 fail. Review stopped after round 1.
• Noted, not changed here: src/modal/segmented_national.py:381-383 respawns an overdue child inside except TimeoutError:, so a spawn failure carries the poll's TimeoutError as context. Child run errors are not affected.

🤖 Generated with Claude Code