A public-health outbreak model must warn leadership when quality falls. Which monitoring control fits?

Triggering an alert when precision drops below a negotiated threshold is correct because a public-health outbreak model must warn leadership when quality falls, not rely on logged trends reviewed only weekly. Explicit alert rules tied to civic quality floors (precision, recall, or related metrics) create accountable, timely notification so epidemiologists and MLOps can respond before degraded scores mislead response planning. Negotiated thresholds also create an auditable contract between model owners and leadership.

Logging precision trends against a negotiated threshold for weekly review fails the warning requirement because an outbreak can accelerate between reviews, and logged trends consulted only weekly still hide clinically relevant quality loss until the next review. Triggering retraining when precision drops below a negotiated threshold fails because starting a retraining job does not itself notify leadership when quality falls, and it does not create the timely warning the question requires. Triggering an alert when training loss rises above a negotiated threshold fails because training loss is a training-process signal, not production outbreak-model quality, and it will not warn leadership when precision drops.

Exam caveat: stems that ask which monitoring control warns when quality falls want threshold-based alerts on the relevant ML metric—not logged trends for weekly review, retraining triggered by a precision drop, or training-loss alerting.

Operational check: set a precision (or agreed metric) alert at the leadership-negotiated threshold on the outbreak model’s production monitor, verify a synthetic precision drop pages the right on-call or leadership channel, and document the threshold in the municipal runbook.