loading…
For each drill, apply the same operating method: define impact, identify the boundary, collect evidence, mitigate safely, recover, and prevent recurrence. Kintsu is used only where AI behavior matters.
Events arrive as e1, e1, e2, e1. Decide where deduplication lives, how multiple instances coordinate, how long keys persist, and what acknowledgement means.
The CRM is unavailable for 20 minutes. Decide which calls retry, how to prevent synchronized retry storms, when to degrade, and when to alert a human.
Refund succeeds; ticket update fails. Name the durable state, resume point, idempotent operation, and reconciliation job.
Failure rate moves from 2% to 11% after deploy. Identify the fastest safe mitigation and the evidence needed before re-release.
HTTP metrics are stable while escalation and customer complaints rise. Define the missing quality signal, affected slice, rollback rule, and new regression case.
One enterprise customer's integration worked last week and now fails with TLS errors. Your clean environment succeeds. Build the timeline, comparison matrix, safe workaround, and escalation packet. State what evidence would distinguish an SDK bug from a proxy or certificate issue.
Stripe records a successful subscription, but the database write times out. Decide the source of truth, webhook acknowledgement rule, idempotency key, retry behavior, event storage, customer experience, and reconciliation process.