Overview

Interpret confidence as a review signal derived from evidence quality—not as probability that a legal conclusion is correct.

Audience

• Lawyer
• Compliance
• Ops

Product Area

AI Governance

Prerequisites

• AI-assisted result with confidence or evidence-strength indicators

Steps

1. Inspect the contributing signals

Review source coverage, citation alignment, recency, jurisdiction match, contradiction, extraction quality, and model uncertainty.

1. Apply the configured threshold

Use thresholds to route work for review or block automation; never convert a score directly into a legal conclusion.

1. Document reviewer disposition

Record whether the result was accepted, corrected, rejected, or escalated and why.

Expected Result

The reviewer can reproduce why the output received its current confidence label and identify which missing evidence could change it.

Security and Audit Notes

• Use the least-privileged role that can complete the task.
• Confirm the resulting change or decision appears in the workspace audit trail when the workflow changes customer data or access.

Limits and Preconditions

• Scores may change when retrieval, evidence, evaluation, or scoring policy changes.
• Scores are comparable only within the documented version and workflow.

Troubleshooting

• A high score is treated as certainty: Re-check the underlying evidence and decision context; confidence is not a warranty or legal opinion.

Related Articles

• Handle Low-Confidence Output
• Review AI Citations
• What Lawie AI Does and Does Not Do

Escalation

If the documented result cannot be reached after the checks above, capture the workspace identifier, affected product area, timestamp, and a redacted error message, then use Support. Never include secrets, privileged legal content, or customer documents in the initial report.