Parent: #163
Add tests that append interleaved frontier, memory, task, critic, and worker events, reopen the database, replay projections from zero, and compare them against stored projection cursors and BootContext. Include duplicate idempotency keys and out-of-order projection advancement attempts.
Acceptance evidence: wrong replay ordering or stale cursor advancement fails the suite.
Customer story
A maintainer deciding whether an autonomous patch is semantically trustworthy benefits because restart and replay behavior is tested against failure modes, not only the happy path.
Source
Generated by Forge brainstorm dry-run from repo-local product vision and axes.
Parent: #163
Add tests that append interleaved frontier, memory, task, critic, and worker events, reopen the database, replay projections from zero, and compare them against stored projection cursors and BootContext. Include duplicate idempotency keys and out-of-order projection advancement attempts.
Acceptance evidence: wrong replay ordering or stale cursor advancement fails the suite.
Customer story
A maintainer deciding whether an autonomous patch is semantically trustworthy benefits because restart and replay behavior is tested against failure modes, not only the happy path.
Source
Generated by Forge brainstorm dry-run from repo-local product vision and axes.