← Benchmark deep diveAll leaderboards

DevOpsBench-100 · agent evaluation

The work behind the score

Inspect a task.

Original instructions, observed actions, and the verdict for each attempt.

Execution error

gpt-6-astra · 222.3s · 43 recorded actions

Original task

How many customer-facing incidents in the last 7 days? Leadership wants the number of customer-facing incidents over the last 7 days (today is day 420, so days 414-420 inclusive). Reconcile the governing definition, time window, identities, and conflicting systems; give the supported answer and explain every material exclusion or unresolved ambiguity. For operational reporting portfolio, a credible answer has to reconcile the governing definition, source precedence, reporting window, duplicate treatment, and the decision made from the reconciled result. Turn the fragmented operating evidence into one defensible result, carry through any justified write, and reopen it to verify persistence. The reconciled answer is due in the 2026-03-09 operating review for operational reporting job. Record whether the normal cycle meets the window, the constraint that governs it, and the consequences of waiting, expediting, or seeking broader authority.

Grading criteria & verdicts

Executable checksUngraded

Recorded benchmark evaluator · devopsbench-100:3.2.7:original-verifier

Model execution did not complete; see retained trajectory