Reading findings and confidence
Every finding carries a confidence band and a brittleness flag. Read both before treating it as actionable.
01
Confidence band
- High: grader was deterministic; no edge case in the rubric.
- Medium: grader matched a documented edge case.
- Low: grader bailed out; needs human review.
02
Brittleness flag
- keyword: defense depends on string-level matching.
- control: defense is structural (tool registry, rate limit).
- unknown: defense classification not yet wired.