Ship a provider canary and stronger physical iMessage QA flow

Goal

Make it practical to catch transport and Apple-client regressions before users by extending the deployed Eve evals through the provider path and preserving a small physical-device acceptance flow.

Scope

• Isolated test number/account and deterministic fixtures.
• Signed inbound, read/typing, delivery status, thread continuity, one reversible write, undo, feedback, and exact-link canary.
• Correlated safe traces across the full path.
• Scheduled and pre-release execution with volume caps.
• Physical-client checklist evidence format.

Acceptance criteria

• No canary can contact a customer or mutate a customer resource.
• Critical cases pass twice consecutively before a release is considered healthy.
• Failures distinguish model variance, provider transport, product service, and rendering regressions.
• Maintainers can adopt the flow without committing provider credentials.