Skip to content
← Evidence archive

Public evidence summary

Historical interrupted-call replay test

Record date
28 July 2026; re-run 13 October 2026
Scope
Laboratory or bench evidence; production behavior is not inferred from it.
Current deployment
Not independently verified by this historical evidence. Current availability

Recorded result

On 28 July 2026, a laboratory Claude Code session survived a forced stop and resume. Its records reconciled, and a completed interrupted response was replayed without being billed again. The test also detected deliberately degraded replay and record behavior. The result depends on the client returning for the held response; it is not a promise about every agent’s retry behavior. The specimen’s task outcome was reported separately and did not determine whether the infrastructure test passed.

The 13 October 2026 re-run

Paid re-run with production gateway source

Recorded result: green

On 13 October 2026 the test re-ran on the bench with gateway and engine source identical to production. In the passing run the retry was served from the store, citing the interrupted call, and 24,678 tokens were excluded from spend; the deliberately broken control run failed exactly the ledger checks, as it must. An earlier run that day did not replay: Claude Code rebuilt its request after a stall of 5 minutes 44 seconds, which the gateway treated as new work and billed once.

Date: 13 October 2026

Scope: Laboratory or bench evidence with production gateway source; one client, one model, one day.

Limit: Whether a client re-sends the same request after a resume is its own behavior; a rebuilt request is new work.