# AI Source Verification Checklist

Version 1.0.0 · CC BY 4.0

## Purpose

A claim-by-claim checklist for opening, matching, and recording sources.

## Before you begin

Record the decision, audience, permitted evidence, confidentiality limits, tool/version where relevant, date, and accountable reviewer. Do not paste confidential material without authority.

## Working fields

1. **Decision:** What must this work help someone decide?
2. **Evidence:** List source IDs and observations separately from interpretation.
3. **Constraints:** Name exclusions, format, time, privacy, and legal limits.
4. **Alternatives:** Record at least two plausible choices and disconfirming evidence.
5. **Evaluation:** Score traceability, specificity, contradiction handling, usability, and repair effort.
6. **Judgment:** State the choice, rationale, uncertainty, and owner.

## Review checklist

- [ ] Every consequential claim maps to an opened source or is labeled interpretation.
- [ ] Product behavior includes a test/access date and visible version where relevant.
- [ ] Contradictory evidence remains visible.
- [ ] Generic claims that could describe any project have been removed.
- [ ] Private data, credentials, client names, and personal details are absent.
- [ ] A human reviewer has tested the resulting artifact.
- [ ] Limitations and manual interventions are recorded.

## Output contract

Return an evidence table, alternatives considered, the recommended artifact, open questions, and a claim-review list. Never invent a missing source, quotation, result, or product behavior.

## Usage example

For a presentation, write the one-sentence audience decision, attach source IDs to each proposed claim, order claims as an argument, and only then describe slides. A reviewer should be able to remove any unsupported claim without rebuilding the entire artifact.

## Limitations

This resource improves process visibility; it does not verify facts automatically or guarantee model behavior. Adapt it to your domain and data policy.