Review follow-up: durable receipt provenance is passed explicitly into the merge. An authoritative same-run partial/error result replaces stale success; workflow-only history still cannot overwrite the final Pulse result. Errors from a different outcome are no longer attached to the retained status.

Local follow-up for G35 / PUL-3565D07C (PLAT-080/219 related). No live workflow schedules, run records, or finding tracking rows were changed.

• Preserve in-memory running status only while that exact active run still has a scheduler run-context registration. Session IDs are not ownership: resumed sessions can span multiple runs.
• Without ownership, adopt the latest terminal evidence for the same run instead of indefinitely retaining stale running status. Clear stale errors and active-run IDs when adopting the terminal record.
• Prefer the durable scheduler terminal receipt over workflow-only history, preserving partial/stopped Pulse outcomes.
• Sort history explicitly by start time before selecting the newest record.
• Calculate next_run from loaded enabled jobs at projection time and again after Pulse finishes, instead of reusing the timestamp from before Pulse. Calendar jobs select the earliest remaining future item; disabled, deleted, or exhausted schedules have no next occurrence. Workflow scopes stay isolated.

Regression coverage exercises the actual GetRuntimeStateForWorkflow reader against a mock workspace API, active-run registration/release, same-time terminal history, reused sessions, unsorted history, durable partial Pulse results, long Pulse delays, calendar exhaustion, disabled jobs, and workspace scope isolation. The individual-step retry recovery issue G67 is separate and is not changed by this patch. This is not a claim of deployed verification or a historical SQLite repair.