The Replit team experienced a 2.5 hour outage on September 29, where an incomplete build was pushed to production, causing affected Repls to become read-only or stop working, with 98% of the affected Repls recovered and the remaining 2% still in progress. The root cause was a lack of guardrails around deployments, which led to the deployment of a faulty build to production clusters. To prevent similar issues in the future, the team is implementing changes such as automated canary analysis, adjusting deployment speeds, and updating the backup system. Users who are affected by the outage can report their Repls for further updates on recovery progress.