Statute or regulation, court, government agency, professional association, peer-reviewed research, trade press, firsthand account, forum, vendor, target record, or DenQAI method.
Methodology & evidence standards
Traceable assumptions are more useful than confident averages.
DenQAI is designed to make the work inspectable: what was observed, what was calculated, what was assumed, what remains uncertain, and what would reverse the decision.
Claim-level evidence
One grade cannot explain authority, support, recency, and commercial interest.
DenQAI records these as independent fields. A current government page can still only partially support one claim; a professional association can provide strong guidance without becoming binding law; a firsthand account can reveal a mechanism without proving prevalence.
Directly supports, partially supports, context only, contradicts, or does not support.
Verified, partially verified, unverified, or disputed.
Binding, official nonbinding guidance, professional guidance, research evidence, target record, experience, opinion, or method.
Federal, state, locality, payer, contract, target, or not applicable.
Source date, reviewed-on date, effective period, and next-review date are separate fields.
Disclosed, identified, none identified in the disclosed source after review, or unknown.
Public fact, bounded example, internal research, design hypothesis, or do not publish.
The six-pass method
From story to decision record.
- DefineDecision, geography, patient segment, time horizon, unit, denominator, and safety checks.
- InventoryFiles, systems of record, owners, dates, gaps, access, and confidentiality.
- ExtractFacts into a stable required file structure with exact citations and a claim-specific support state.
- ReconcileIndependent control totals across the practice-management system, bank, tax, contract, remittance, schedule, provider, and permitted aggregate records.
- StressBase, downside, adverse-entry, staffing-loss, payer-delay, and decision-reversal cases.
- RecordConclusion, unresolved evidence, owner, deadline, human review, correction, and recheck date.
Safety checks
No score can compensate for a disqualifying fact.
DenQAI does not average unresolved lease, authority, privacy, patient, staffing, liability, reserve, or clinical-control problems into a favorable total. Facts, calculations, safety checks, and owner preferences remain separate.
Decision ruleresolve each safety check → compare natural-unit differences → stress the preferred case → request the next evidence
When evidence cannot resolve a material uncertainty, show it, price a defined scenario, restructure the deal, pause, or walk away.
AI and calculation controls
Models propose. Records and reviewers decide.
Require exact file, page, row, cell, URL, issuer, and date. An uncited answer is a lead, not evidence.
Extract before interpretation so missing data and contradictions remain visible.
Use a spreadsheet or controlled code for totals and scenarios; independently recompute control figures.
State the purpose, numerator, denominator, unit, event date, window, inclusions, exclusions, source, owner, and failure modes.
Search for double counting, period mismatches, missing components, contradictory evidence, and another plausible explanation.
Qualified professionals approve clinical, coding, legal, accounting, lending, employment, privacy, and transaction conclusions.
Inspect what controls the public product.
The corpus manifest identifies current sources, superseded rules, formula repairs, capability boundaries, and publication blocks.
Policy inference
Authority, implementation, exposure, response, outcome.
A valid law is not evidence that its operational machinery exists. An operational program is not proof that a target was exposed. Exposure is not a response, and a before/after outcome does not establish causation. DenQAI’s policy method preserves those five links and applies separate safety checks for measurement stability, baseline, lag, comparison, rival explanations, and owner translation.
Until those observations and a qualified design review exist, the site reports verified authority, verified implementation, or bounded descriptive observations—not a predictive policy effect.
Case-series method
Stories generate mechanisms. They do not estimate prevalence.
The governed owner case-series design uses within-case chronology, proposition-level source kind, support relationship and verification state, contradiction review, negative-case search, participant fact-check, and a pre-specified cross-case mechanism matrix. A small purposive pilot may report what was observed in its bounded cases; it cannot establish a national rate or a causal effect.
Recruitment remains closed until regulatory, consent, privacy, security, conflict, and publication safety checks have dated independent evidence.