AWS Certified AI Practitioner (AIF-C01) · Objective 1.3
Defining rollback triggers
1.3Standard
Release a new model gradually.
Return to the current version if quality or safety declines.
Choose two triggers or procedures.
Hint 1: Focus on the requirement
Define measurable triggers for reversal and rehearse the data, cache, notification, and monitoring steps that follow it.
Hint 2: Key terms
Review rollback conditions and recovery procedures.
Rollback condition
A measured threshold or event that triggers return to a previous approved version.
Recovery procedure
The ordered steps for restoring service, data, notifications, and monitoring after a failure.
Release threshold
A limit for quality, latency, errors, safety, or cost used to judge a release.
Rollback rehearsal
A test that verifies a rollback can be completed safely and consistently.
× Incorrect
Set versioned thresholds for latency, errors, quality, refusal, safety, and cost, and document data, cache, notice, and reevaluation steps after rollback.
Detailed explanation
A
IncorrectStop monitoring the new model after rollback
Investigation evidence is lost.
B
IncorrectWait for complaints before defining triggers
Impact can expand.
C
CorrectDefine versioned thresholds for p95 latency, errors, quality, safety, refusal, and cost
Abnormal behavior can be detected.
D
CorrectDocument data, cache, notice, and reevaluation steps after rollback
Recovery inconsistencies are managed.
Check official documentation
Compare the following points with the official documentation.
Path & Pass is an independent learning site and is not affiliated with any certification body. Questions are original practice content based on published objectives.