BEBEL — Decision Evidence Checklist
Public research artifact · August 2026

Use this checklist before a system produces a comparison, recommendation, or
decision-support answer.

1. State the question
   - What decision is the system helping a person make?
   - Who is accountable for the final call?

2. List the evidence
   - Identify the source for every material input.
   - Mark each input as observed, estimated, missing, or contradictory.
   - Do not let a default value silently become an observed fact.

3. Define the answer boundary
   - What minimum evidence does this question require?
   - Which missing or conflicting inputs should stop the answer?
   - Which comparisons are invalid because their units differ?

4. Exercise refusal
   - Test the accepted path.
   - Test missing, estimated, stale, contradictory, and incomparable inputs.
   - Make a refusal explain what evidence would allow the work to continue.

5. Preserve human control
   - Show the evidence status with the answer.
   - Keep the consequential judgment with the accountable owner.
   - Do not imply that an automated output is the decision itself.

6. Record corrections
   - Let new evidence supersede the prior answer.
   - Keep the earlier state and reason for correction legible.
   - Re-run the boundary after a correction.

7. State the limits
   - Name what was tested.
   - Name what was not tested.
   - Do not generalize a bounded result into a universal claim.

This checklist is a public explanation of a BEBEL evaluation method. It is not
product access, certification, or legal, financial, or medical advice.
https://bebel.io/work/incomplete-data/