Issue #1882 reports repeated PBS history observations. The existing regression test does not actually rebuild the registry or verify persistence after closing the store, leaving the repaired observation-time behaviour exposed to regressions.
Exercise repeated registry rebuilds across three source observations and SQLite close/reopen, checking canonical identity and all four storage metrics. Retain later equal-valued observations rather than deduplicating by value. Production behaviour is unchanged.
Validation: focused normal and race tests passed in the original execution, including ten race repetitions; a source-time mutation made the new test fail. This amendment changes only commit metadata.
Change-source: pulse-maintainer
pulse-triage[bot]
c590a01f33