# Evidiam Methodology and Evidence Standards

Evidiam uses a conservative evidence process so a customer-facing finding can withstand review by the manufacturer's product experts.

## Six phases

1. Scope the products, markets, buyer journeys, competitors, AI platforms, and material claim classes.
2. Establish current authoritative records for the exact products and configurations.
3. Test natural buyer questions without revealing the expected answer to the tested system.
4. Classify correct answers, cautious answers, unsupported claims, source conflicts, verified AI errors, and visibility gaps.
5. Review identity, authority, materiality, and repeatability before a finding becomes customer-facing.
6. Retest after corrections and monitor future changes.

## Finding classes

- Verified AI error: A captured answer contradicts authoritative exact-product evidence.
- Verified public-source conflict: Current public sources disagree, but a captured AI error has not been established.
- Visibility gap: A qualifying product is materially absent or displaced in a defined buyer journey.
- Source-coverage gap: AI relies on weak, stale, or third-party sources when stronger manufacturer evidence is available.
- Unresolved: Evidence is incomplete, ambiguous, conflicting, or insufficient for a customer-facing conclusion.
- Correct answer: The tested response agrees with current controlling evidence.

## Limits

AI responses can vary by interface, prompt, retrieval mode, location, session, and model update. Evidiam records observable conditions and does not imply access to private ranking systems or training data.

Human-readable page: https://evidiam.com/methodology