# DenQAI policy-evidence and calibration protocol

Version: 0.1.0  
Reviewed: July 23, 2026  
Status: Public method; no predictive policy simulator

## Purpose

This protocol tests whether a dental-practice policy changed an observable mechanism and whether available evidence supports a descriptive, associational, or causal claim. It does not advocate for a policy, estimate a guaranteed owner result, or replace legal, actuarial, economic, payer, workforce, or research review.

## Required evidence chain

1. **Authority:** Verify the enacted text, regulation, official guidance, order, program term, version, jurisdiction, scope, exclusions, adoption date, and effective date.
2. **Implementation:** Verify the operational event required to make the policy real: rule, form, system, report, appropriation, contract, privilege, payment, enforcement process, or other implementation record.
3. **Exposure:** Define which plans, practices, clinicians, patients, codes, products, geographies, and periods were actually affected.
4. **Practice response:** Measure the pre-specified payment, administration, staffing, participation, capacity, contract, cash, or owner behavior that the mechanism predicts.
5. **Outcome:** Measure the pre-specified owner or patient outcome using a stable numerator, denominator, unit, geography, and period.

Stop at the first unsupported link. Do not treat adoption as implementation, implementation as exposure, exposure as response, or response as outcome.

## Non-averaged calibration gates

- The controlling authority and relevant version are verified.
- Adoption, effective, and operational dates are separated.
- The exposed population and unaffected comparison are defined.
- Numerator, denominator, unit, geography, payer product, and data-generation process are stable.
- The baseline shows trend and seasonality where they matter.
- The post-period begins after a defensible implementation lag.
- Concurrent benefit, rate, workforce, economy, ownership, coding, technology, enforcement, and market changes are recorded.
- Missing data, revisions, exclusions, and measurement breaks remain visible.
- The claim does not exceed the design.

A failed gate limits the claim. Gates are not averaged into a score.

## Claim levels

- **Verified authority:** the controlling text or official program term is current for the stated scope.
- **Verified implementation:** the required operational event occurred.
- **Descriptive observation:** a defined measure changed in a defined population and period.
- **Association:** exposure and outcome covary after specified adjustments, without establishing causation.
- **Causal estimate:** the design supports a defensible counterfactual and qualified independent review accepts the assumptions and uncertainty.

DenQAI’s current public policy layer stops at verified authority, verified implementation, and bounded descriptive observations unless a separate reviewed evaluation supports more.

## Comparison design

A simple before/after difference is usually weak because other events can change at the same time. Depending on the question and data, qualified reviewers may consider:

- an interrupted time series with enough pre- and post-observations;
- a comparable unexposed jurisdiction or population;
- difference-in-differences with assessed pre-trends;
- regression discontinuity around a real eligibility threshold;
- a randomized design where ethical and operationally possible; or
- a carefully bounded case study when population inference is not claimed.

The comparison must be similar on the factors most likely to affect the outcome. Statistical adjustment cannot repair an indefensible comparison or unstable measurement.

## Owner translation

Keep public-policy observations separate from the target practice. A statewide rate, carrier ratio, license pathway, award, or access measure becomes relevant to an owner decision only after the exact address, entity, contract, payer product, fee schedule, staffing model, clinical days, costs, debt, household runway, and implementation facts are connected.

## Data and privacy boundary

The public Policy Evidence File is a local drafting tool. Do not enter protected health information, patient-level records, credentials, secrets, unredacted contracts, identifiable employee information, or confidential payer data. Use coded locators pointing to records held in an authorized system.

## Core sources

- Medicaid fee-schedule transparency: https://www.ecfr.gov/current/title-42/chapter-IV/subchapter-C/part-447/subpart-B/section-447.203
- Massachusetts dental insurance and loss ratio: https://www.mass.gov/info-details/dental-insurance
- Dentist and Dental Hygienist Compact status: https://ddhcompact.org/
- NHSC site eligibility: https://nhsc.hrsa.gov/sites/eligibility-requirements
- Minnesota noncompete statute: https://www.revisor.mn.gov/statutes/cite/181.988
- CMS 2026 dental and oral health measures: https://www.medicaid.gov/medicaid/quality-of-care/downloads/dentaloralhealth-ta-resource.pdf
- Medicaid clean claims: https://www.ecfr.gov/current/title-42/chapter-IV/subchapter-C/part-447/subpart-A/section-447.45
- GAO evaluation-design reference: https://www.gao.gov/assets/DESIGNING_EVALUATIONS.pdf