Collect a fresh: Which action provides credible release evidence?
Repeated tuning contaminated the validation file; credible release evidence requires a fresh representative set protected from tuning.
The question
A software-procurement model was tuned repeatedly on the same validation file until it met an accuracy target. The team proposes calling that file the test set, but has no untouched records. Which action provides credible release evidence?
Preparing for AIGP? Take the free 5-min readiness quiz →
- Relabel the tuned file as the test set.Repeated tuning exposed the file to development decisions, so relabeling it does not restore independent evaluation.
- Release using training accuracy and a vendor benchmark as corroboration.Training accuracy is not an unseen-data measure, and a vendor benchmark may not represent this procurement use case or restore local independence.
- Collect a fresh, representative held-out test set. ✓A fresh set representative of deployment conditions can provide independent release evidence if it remains protected from further tuning.
- Use cross-validation on the tuned file for final release evidence.Cross-validation can be useful during development, but repeated adaptive tuning on the same records still compromises the credibility of final evidence.
The trap
If records influenced tuning, they cannot serve as independent final-test evidence. How to remember it
Repeated tuning contaminated the validation file; credible release evidence requires a fresh representative set protected from tuning.
How many of these would you get right?
One of 1581 AIGP questions on Certsqill. Take a free five-minute check and see your score per domain — not one number, but which section to open tonight.
Test your AIGP readiness — freeMore Understanding How to Govern AI Development questions
- Test subgroup outcomes and inspect correlated inputs: Which mitigation best addresses the remaining risk? →
- Report pediatric and adult error rates: What evidence controls release? →
- Compare evaluation scope with deployment conditions: Which model-card distinction matters most? →
- All 426 Understanding How to Govern AI Development questions →
Part of the Certsqill AIGP question bank · Understanding How to Govern AI Development ·
Every answer, right and wrong, comes with its own explanation.