Incident explanation unverified claims
After a partial outage, leadership wants a one-paragraph explanation for a customer-facing status update. You draft it with an AI from the incident timeline. The AI writes: 'A brief networking issue caused some requests to slow down; no data was lost or exposed.' The actual cause was a deploy that exhausted a connection pool, and you haven't yet confirmed the 'no data exposed' claim. What's your process for verifying an AI-simplified incident explanation before it becomes a public statement?
review_public_claims(claims: list[str], confirmed_fact_ids: list[str]) → list[str][["A brief networking issue caused some requests to slow down|f1","No data was lost or exposed|f2"],["f1"]]out["publish","remove"][["Requests were slower than usual|"],[]]out["hedge"][["All customer data is safe|f3"],["f3"]]out["publish"]Treat the AI’s output as a draft to verify, not an answer to trust. Name the specific flaw and the input that triggers it, say how you’d catch it (tests, edge cases, reading critically), and how you’d re-prompt or decompose to get it right.
Vibe & agentic: describe the solution in plain language (or narrate it) and the coach grades your approach.