Distillation

Date: Sep 16, 2026

• 🔴 [requested-tests] (10:09) User supplied an Effect test asserting recovery behavior "records a recovery continuation without forging an assistant identity", using execution ID "continue-only" and expecting exactly 1 row in SessionRecoveryTable.
• 🔴 [requested-tests] (10:09) User supplied an Effect test asserting the invariant "never recovers a live execution lease"; the test expects SessionInputTable to remain empty.
• 🔴 [requested-tests] (10:09) User supplied an Effect test asserting the invariant "never recovers an execution superseded by a newer assistant".
• 🔴 [requested-tests] (10:09) User supplied recovery scanner tests involving poisoned entries, wake tracking via wake: (id) => Effect.sync(() => wakes.push(id)).pipe(Effect.asVoid), and batching based on SessionRecovery.BATCH_SIZE; one expectation is wakes having length SessionRecovery.BATCH_SIZE * 2.
• 🔴 [requested-tests] (10:09) User supplied an Effect test asserting recovery behavior "redrives one existing recovery after its replacement claim crashes before promotion".
• 🔴 [requested-tests] (10:09) User supplied an Effect test asserting the invariant "never recovers a pre-promotion claim when input commits at the recovery boundary"; the scenario includes input with delivery: "steer" and expects SessionRecoveryTable to be empty afterward.
• 🔴 [requested-tests] (10:09) User supplied recovery tests that invoke SessionRecovery.recover().pipe(Effect.provideService(SessionExecution.Service, execution)).
• 🔴 [requested-tests] (10:09) User supplied an Effect test asserting the invariant "never promotes a recovery superseded before scanner validation".
• 🔴 [requested-tests] (10:09) User supplied an Effect test asserting the invariant "never promotes a quarantined recovery continuation as an ordinary steer".