Describe a time you made a mistake at work. How did you handle it?

Interviewers ask this to test psychological safety and ownership. A perfect-record answer is itself a red flag.

STAR walkthrough — sample answer:

Situation: I was leading testing on a major UI release on a previous team. The release was scheduled for Friday afternoon. On Wednesday I'd updated the staging deploy script to include a new seed data step, and I assumed the script's behaviour in production would be identical.

Task: Verify the release in staging, sign off, then re-verify post-deploy in production.

Action: Staging passed. Production deployment went out Friday at 2pm. By 2:15pm, several customer support tickets came in saying the dashboard wouldn't load for some accounts.

I'd missed a critical detail: my staging seed data step had silently masked a real schema mismatch in production. The new code expected a JSON column to be present in a specific table; staging had it because I'd seeded it; production didn't because no migration had run yet.

I noticed the trend in the support channel within minutes. Three things, in order:

1. Acknowledged immediately, before I knew the full picture. Posted in the eng channel: "Issue affecting some accounts post-deploy. I'm investigating, suspect missing migration. ETA on update: 15 min." Pinged the engineer who'd built the feature.

2. Diagnosed and rolled back fast. Within 15 minutes we'd confirmed the migration had been written but not deployed. Decision: roll back the application, run the migration, redeploy. Total customer-facing impact: ~25 minutes of dashboard breakage.

3. Owned the post-mortem honestly. I wrote it the same day. Didn't blame the missing migration on anyone — described what I had missed: I should have caught the schema dependency in pre-deploy review, and my staging seed had given me false confidence by masking the gap.

Result: Customer impact was small (25 min, partial population). The post-mortem led to a checklist change: any release with schema-touching code now requires a "migration run on production-like env without seed-mask" step in pre-deploy. The seed data utility itself got updated to error out if it overlapped a recent migration's responsibility.

What I learned: convenient tooling (seed scripts that "just work") can mask real failure modes. From then on I distrust environments that work too smoothly without me understanding why.

Why this works: real mistake (not a humblebrag), fast acknowledgement, owned the gap matter-of-factly, structural fix that would catch the same class of issue next time. Doesn't blame anyone else.