SYNTHETIC RECOVERY DESIGN · NOT LIVE PROVIDER EVIDENCE
WAIT_RESOURCE: Recovery Outside the Model Turn
A long-scroll visual companion to the article. It summarizes the article’s evidence boundary and method; it does not extend the underlying result.
At a glance
Evidence in scope
- V20: 1,000 simulated WAIT_RESOURCE lanes
- V20: 1,000 simulated WAIT_AUTH lanes
- V20: 2,000 synthetic resume events
Evidence boundary: V20 counts describe simulated lanes/events. V21’s six test cases are a separate report and are not added to V20.
How to use the diagram
Persist the job/checkpoint, classify the wait, apply bounded backoff, reconcile uncertain effects, resume, then verify.
What this visual explains: A durable job record lets an orchestrator wait and resume without asking a transient model turn to remember the job.
What it does not establish
Synthetic fake-adapter harness only: zero provider calls/tokens and zero production jobs. Live recovery effectiveness is unknown.