ERD_INDEXERS.md requires retention and reputation rollup jobs to honor time-window semantics (1h, 24h, 7d) and retain deterministic aggregation behavior.
Existing tests covered retention purge and one-hour rollup behavior, but explicit coverage for 24h/7d boundary inclusion-exclusion was missing.
Phase 11 still listed Add job runner tests for retention and rollups as incomplete.
Decision:
Add a revaer-data job-runner test that validates multi-window rollup boundaries:
includes events within 24h and 7d windows;
excludes events older than each target window;
verifies derived success and acquisition metrics for both 24h and 7d windows.
Mark the Phase 11 checklist item complete.
Alternatives considered:
rely only on SQL inspection: rejected because boundary mistakes are subtle and regress easily.
Consequences:
Positive outcomes:
rollup window boundaries now have executable regression coverage for non-1h windows;
retention/rollup job-runner coverage aligns with checklist intent.
Risks or trade-offs:
fixture timestamps are relative to test clock; if ERD windows change, test expectations must be updated.