What CI/CD practice best validates a new agent build before promotion using real municipal traffic patterns?
Select an answer to reveal the explanation.
Short Explanation
Replaying last week’s anonymized 311 traces is a dress rehearsal with the real choreography. String-length unit tests and one hello world don’t stress tools or peak shapes. Staging replay catches surprises before residents do.
Full Explanation
Pre-production validation for civic agents should replay anonymized production-like 311 traces in staging before promotion. Replaying last week’s anonymized 311 traces exercises real tool paths, utterance mix, peak shapes, and failure modes that unit tests of string length or a single hello-world dialog will never stress, reducing promotion risk for municipal traffic patterns. Staging replay is dress rehearsal with the city’s actual choreography, privacy-preserving.
Promoting immediately if unit tests of string length pass confuses trivial syntactic checks with behavioral readiness under real load. Skipping staging because anonymized traces are “never useful” discards the best available rehearsal of real demand. Validating only with a single hand-written hello world dialog misses tool chains, multi-turn flows, and peak-hour shapes.
Exam caveat: staging replay of anonymized municipal traces is the CI/CD practice that best validates agent builds against real traffic patterns before promotion. Operational check: export last week’s anonymized 311 traces, replay them in staging against the candidate build, compare error rate, latency, and tool-success to baseline, and block promotion on regressions beyond agreed thresholds with an automatic report attached to the release.