All courses · Course overview · Lesson sequence
Skip to lesson textLocal synthetic runbook
Owner: the learner running this fictional exercise. Scope: in-memory museum-label comparisons, fixture decisions and zero-write simulation receipts. No real source, user, account, message or record may be introduced as part of this course run.
Start with the core test command from the course folder. Run advanced_demo.py, then the baseline and repeat evaluations. Keep the small case report or terminal output needed for your explanation; there is no service to leave running. Closing the process ends the RAM stores. An explicit exported JSON string can be restored only as the documented transfer simulation; it is not automatic backup.
| Observation | Meaning and permitted local response |
|---|---|
| Input validation fails | Repair fictional input and use a clear request identity; no tool was called. |
| RequestConflict | Existing ID belongs to different input/source/policy. Inspect the difference; do not overwrite the old identity. Use a new ID for intentional new work. |
| StaleRevision | Reload current state and reconsider the decision. Do not merely substitute a new revision into an old approval. |
| Unsupported proposal or wrong target | Stop and inspect the planner/contract. Do not add a broad execution helper. |
| Invalid tool output or unknown exhibit | No supported draft is available. Inspect the fixture or intended scope; do not guess the missing fact. |
| Known transient read failure | The controller may reserve another read within both budgets. Failed reservations remain charged. |
| IN_FLIGHT after injected crash | Outcome is uncertain. Default recovery stops. Explicit abandonment is allowed only for this synthetic read and consumes the old reservation. |
| Budget exhausted | Explain which counter ended the run. Change policy only as a deliberate new execution contract, with tests and a fresh identity. |
| REVIEW | Inspect the exact draft/source/digest/revision. A note cannot approve it. Use only fixture decision data. |
| Invalid, backdated or expired review | Preserve the outcome. A stale draft needs reconsideration; do not bypass the check. |
| APPROVED_FOR_SIMULATION | New finalization still checks content/source/time/budget. It can only create a simulation receipt. |
| REJECTED | No finalization path. Explain the rejection; do not relabel it approved. |
| SIMULATED | Saved receipt says zero real writes. Replaying it adds no new work. |
For optional graph execution, use the pinned isolated runtime, disable both tracing flags, and run the retained framework test suite. An optional skip means not executed. The graph and Engine both use in-memory checkpoints; thread IDs are routing keys rather than access credentials.
Record version, command, expected/actual result and untested limits. The reference recommendation is local synthetic demonstration only. Any real integration, identity/time infrastructure, persistent recovery, live-model evaluation or publication requires separate scope and authorization; this runbook does not grant it.