This release includes a transparent update on infrastructure reliability issues and upcoming improvements:
Outages: Infrastructure changes made during scaling caused reliability issues. Failover didn't perform as expected. Redundancy is now in place so a single provider outage won't take Flow down, along with health monitoring and an on-call rotation to catch issues faster. This is now the top priority across engineering and infrastructure.
Accuracy: The auto-cleanup setting may have affected other settings. A rollback is actively being tested. An upgraded speech recognition model is also in progress, targeting both accuracy and latency improvements for all users.
Longer term, better internal tooling is being built to catch issues before they reach users.