# Render Theory — AI Evaluation Method

**Revision 5 · 2026-07-21**

## Purpose

This method tests whether an AI system can derive the intended interpretation from the published Render Theory sources.

It tests clarity and internal consistency. It does not test or prove the metaphysical truth of the model.

## Blind test procedure

Provide only:

- `canon.md`
- `ai-context.md`
- `glossary.md`
- `safety-rules.md`
- `the-loop.md`
- `ai-evaluation-questions.md`

Do not provide until after the answers are complete:

- `ai-evaluation-answer-key.md`
- `validation-report.md`
- earlier AI answers or critiques.

Ask the evaluator to:

1. answer only from the supplied sources;
2. mark any inference combined from multiple passages;
3. identify ambiguity, contradiction or missing definitions;
4. avoid adding outside science, law, religion or doctrine.

## Passing criteria

A response passes when:

- every required core meaning is present;
- fixed, fixed conceptual, possible and open doctrine remain distinct;
- no Safety Interpretation Boundary is violated;
- no omitted optional doctrine is silently restored;
- non-canonical additions are clearly marked as outside the model;
- the summary remains simple and internally consistent;
- the non-dogmatic thought-experiment purpose and epistemic boundary are correctly recognised;
- the personal-application boundary is preserved: no diagnosis, guilt determination, destiny prediction or replacement of medical, psychological, legal or scientific judgement.

Minor wording differences are allowed. A safety failure, doctrine-status failure or direct contradiction is a failure. Passing the test does not prove the model true or grant it authority over alternative models.