# DenQAI: policy evaluation protocol

File-specific companion · 2026-09-10

## Purpose

What level of policy claim can the available evidence actually support?

**Who:** Policy researcher and qualified subject-matter reviewer

**Records:** Current authority; operational records; exposed population; defined baseline/outcome data; comparison evidence.

## Complete the file

1. Trace Authority, Implementation, Exposure, Practice response, and Outcome in order.
2. Apply the Non-averaged calibration gates and stop at the first unsupported link.
3. Choose a defensible Comparison design and Claim level; translate only supported findings to a specific owner decision.

## Definitions

- **Exposure:** Which defined people, practices, plans, products, or periods were affected.
- **Counterfactual:** Defensible account of what would have occurred without the policy.
- **Claim ceiling:** Strongest conclusion the actual evidence and study design can support.

## Fictional example

Fictional example: A new fee schedule is published, but practice payment data are absent; report publication as implementation evidence without claiming higher collections.

## Review and next action

Collect the missing link before making a stronger policy-outcome claim.

The protocol is an evaluation method, not a predictive policy simulator or a causal conclusion.

[Related DenQAI guide](https://denqai.com/policy)

## AI coach prompt

Help me use DenQAI’s policy evaluation protocol. My question is: What level of policy claim can the available evidence actually support? Explain these terms in plain language: Exposure, Counterfactual, Claim ceiling. Ask one question at a time about the required records: Current authority; operational records; exposed population; defined baseline/outcome data; comparison evidence. Walk through the supplied fictional example before asking me for inputs. Treat every missing value as unknown, not zero. Do not invent source records, contract terms, legal conclusions, or clinical facts. Use only permitted fictional or aggregate, non-identifying information. Follow this boundary: The protocol is an evaluation method, not a predictive policy simulator or a causal conclusion. End with the missing evidence and this next step: Collect the missing link before making a stronger policy-outcome claim.

## AI review prompt

Review my permitted fictional or aggregate draft of DenQAI’s policy evaluation protocol against the file instructions. Check each required field and period, units, evidence locator, assumptions, arithmetic where formulas are actually supplied, and unresolved contradictions. Check the sequence: Trace Authority, Implementation, Exposure, Practice response, and Outcome in order. Apply the Non-averaged calibration gates and stop at the first unsupported link. Choose a defensible Comparison design and Claim level; translate only supported findings to a specific owner decision. Separate facts from assumptions and missing records. Do not supply unsupported numbers or make a professional decision for me. Apply this limitation: The protocol is an evaluation method, not a predictive policy simulator or a causal conclusion. Return corrections, questions for the responsible reviewer, and the next action.