Guard Set
Preserve important regression cases for repeated evaluation.
Before you start
An open project with routed evidence.
Quick start
- Select a guard case and review its images and context.
- Use available evaluation controls to compare model behavior on the same evidence.
Functions and choices
Guard images remain test evidence until explicitly promoted. They are not automatically included in training.
Result
Evidence with a deliberate role and preserved context.
Next step
Compare regressions in evaluation or review them in Failure Bank.