Dashboard › opencode › Distillation
a8f6c40b-a922-4387-9689-f201e31ed9e8["lore_tm_v1_BVsyfZ4Vjf0NHfpaEQa0fSiFHq-eMNwJueMByA4gdtQ","lore_tm_v1_K_tY51ksOihS9NuX8PALqqmLEql-HyaH7PEUHafcv0Q","lore_tm_v1_B3jSFub0i3-THu-bwMeQd5hv7U79mMd7q3jgdnq05tc","lore_tm_v1_M0836WlSLou65gyho_tYtIeMRrtzwFkun-eY82u9KY8"]
Date: Sep 15, 2026
packages/core/test/session-projector.test.ts:1192 tests that distinct creator events cannot reuse one projected message ID: a SessionEvent.Synthetic creates msg_creator_collision with text "keep me", a subsequent SessionEvent.Step.Started using the same ID fails, and the existing row remains { type: "synthetic" }.packages/core/test/session-projector.test.ts:1233 tests that SessionMessageUpdater.memory({ messages: [stale, completed] }).getCurrentAssistant() returns undefined rather than reviving an earlier incomplete assistant when a newer completed assistant exists.packages/core/test/session-projector.test.ts:1258 verifies SessionEvent.Step.Ended updates only the newest incomplete assistant projection: msg_assistant_1 remains without time.completed, while msg_assistant_2 receives finish: "stop" and time.completed: DateTime.makeUnsafe(1).packages/core/test/session-projector.test.ts:1316 verifies an event targeting the newer completed assistant does not revive the stale incomplete assistant: SessionEvent.Text.Started targets msg_assistant_completed with textID: "text-stale"; the completed message keeps time: { created: DateTime.makeUnsafe(1), completed: DateTime.makeUnsafe(2) }, while msg_assistant_stale remains incomplete and unchanged.packages/core/test/session-execution-local.test.ts uses SessionExecutionLocal, an in-memory database via Database.layerFromPath(":memory:"), and session ID ses_execution_local.packages/core/test/session-execution-local.test.ts:43 verifies a successful execution.resume(sessionID) drain removes the SessionExecutionTable claim.packages/core/test/session-execution-local.test.ts:52 verifies closing the child Scope of a running execution preserves its claim; claim count remains 1 before and after Scope.close(child, Exit.void).packages/core/test/session-execution-local.test.ts:68 verifies execution.interrupt(sessionID) explicitly clears the execution claim.packages/core/test/session-execution-local.test.ts:82 verifies a second SessionExecutionLocal owner cannot run while the first owner’s lease is live: secondRuns remains 0 and exactly 1 claim remains.packages/core/test/session-execution-local.test.ts:105 verifies a second owner does not overwrite an expired lease: after setting the first claim’s expires_at to 0, second.resume(sessionID) leaves secondRuns at 0 and preserves the stale claim’s exact id and owner_id.packages/core/test/session-execution-local.test.ts:136 verifies an execution never re-enters its own uncertain live lease: the first run changes its claim phase to "unknown" and dies with "simulated failure"; a second resume leaves total runs at 1 and retains exactly 1 claim.packages/core/test/session-execution-local.test.ts:163 verifies heartbeat expiry is renewed from the current clock: initial expires_at is 30_000, then becomes 40_000 after TestClock.adjust("10 seconds"), and 50_000 after another 10 seconds.packages/core/test/session-execution-local.test.ts:186 verifies heartbeat never revives an expired claim: after manually setting expires_at: 0 and advancing TestClock by 10 seconds, expiry remains 0.packages/core/test/session-execution-local.test.ts:204 verifies cancellation settles before a pending successor starts: interruption waits on the first run’s finalizer, a wake queues the successor, releasing the finalizer allows the second run, total runs becomes 2, and all claims are cleared.packages/core/test/session-runner.test.ts defines toolExecutionsReady = 5, model fake-model, replacement model replacement, recovery model recovery, all using provider "fake", and fragment kinds ordered as ["text", "reasoning", "tool input"].packages/core/test/session-runner.test.ts:598 tests redriving only the synthetic recovery continuation for execution ID "lost-recovery-wake" in phase "ready"; the recovery provider receives user texts ["Session execution interrupted by server restart", "continue"].packages/core/test/session-runner.test.ts:672 tests redriving an ordinary prompt after its advisory wake is lost, using message ID msg_lost_prompt_wake; recovery ultimately clears wake_pending.packages/core/test/session-runner.test.ts:707 tests that no provider is dispatched after ownership changes; the claim is replaced with { id: "provider-successor", owner_id: "successor", expires_at: 60_000 } while model resolution is blocked.packages/core/test/session-runner.test.ts:733 tests that streamed output is never published after ownership changes; the stale output fixture uses text fragment ID "text-fenced" and content "stale output", and ownership changes to { id: "publication-successor", owner_id: "successor", expires_at: 60_000 }.packages/core/test/session-runner.test.ts:760 tests that a local tool never starts after ownership changes: an echo call with ID "call-fenced" and input { text: "must not run" } is fenced by ownership changing to { id: "tool-successor", owner_id: "successor", expires_at: 60_000 }.packages/core/test/session-runner.test.ts:798 tests that interrupted tools are never failed after ownership changes; it uses assistant message ID msg_interrupted_tool_fence, call ID "call-interrupted-fence", tool "echo", provider: { executed: false }, and successor ownership { id: "interrupted-tool-successor", owner_id: "successor" }; the stale runner exits with failure.