What This Looks Like
A confidence score, certainty label, risk score, probability-like value, or threshold result from the AI triggers a workflow action. The system may escalate, approve, reject, notify, route, block, or execute based on the score before enough calibration, approval, or authority control is applied.
Why It Matters
Scores can look more authoritative than they are. If a score triggers action directly, the AI may become a decision authority through a numeric threshold. This is risky when the score is not calibrated, the threshold is not governed, or the action requires human or policy approval.
Structural Signal
A model-generated score crosses an action threshold without a sufficient authority boundary. The issue is not simply that the score is wrong; it is that the score has been connected to action in a way that may exceed its declared authority.
Common Triggers
- Workflow automation treats confidence as approval
- Thresholds are configured without review or calibration
- Risk labels trigger routing or escalation directly
- The model emits a score without explaining its authority limits
- A downstream system consumes scores as executable state
- No distinction is made between advisory score and approved decision
When to Use This Issue
Use this Issue when a confidence, certainty, risk, or probability-like score triggers or controls an action without adequate approval, calibration, or authority structure.
When Not to Use This Issue
Do not use this Issue when a score is displayed only as advisory information and no action is triggered. Do not use it when the action is governed by a separate valid approval path.