Accuracy
Verify factual claims, calculations, definitions, citations, and source alignment.
Evaluation Framework
APHI workflows should not advance because the output looks polished. They advance when reviewers can inspect the evidence, risks, limitations, and results.
Safety
Check refusal rules, escalation points, uncertainty, and prevention of unsupported guidance.
Usefulness
Assess whether the output helps a real practitioner take the next step under workflow constraints.
Equity
Review missingness, subgroup effects, stigmatizing language, and potential resource allocation harms.
Source Quality
Label authoritative guidance, peer-reviewed sources, preprints, vendor claims, and secondary summaries.
Governance
Document privacy review, reviewer role, approval path, monitoring, and operational boundaries.
Readiness levels
APHI separates idea quality from deployment readiness. A workflow can be promising and still require review before any operational use.
Concept
Question and possible AI support are defined. Not tested.
Draft
Inputs, outputs, review checklist, and example cases exist.
Pilot
Tested with synthetic, historical, or partner-approved data under supervision.
Operational
Governance, monitoring, evaluation results, and escalation rules are documented.
Minimum evaluation record
Each workflow needs enough documentation for a reviewer, health department leader, or funder to understand what has been tested and what remains uncertain.
Record field
Purpose
Status
Intended use
Defines the task and user
Required
Excluded use
Prevents misuse and overreach
Required
Test cases
Exercises known failure modes
Required
Reviewer role
Assigns accountability
Required