The Nine Steps of Clinical Recommendation Readiness
How an AI system earns the right to answer a clinical question — for one patient or a whole population — Clinical Recommendation Readiness Framework (CRRF)
“I cannot determine this” is a first-class output, not a failure.
1
Understand the question
What specific clinical decision or analysis is being asked — and for what subject, one patient or a defined population? Declared explicitly, no undirected retrieval.
Unclear question or unit → no authority
2
Know what data is needed
The question defines the data model: what data and context a competent clinician would need — required, material, contextual, or irrelevant.
3
Check what we actually have
Provenance, completeness, currency, and gaps versus what is expected — of this patient's record, or of the dataset's representativeness and selection bias.
Data unfit for this question → authority capped
4
Interpret before using
Raw values become evidence only through interpretation. A potassium from a hemolyzed sample is not a potassium; a code is not a phenotype until validated at scale.
5
Judge sufficiency
Is the data sufficient for the question? Sufficient, conditionally sufficient, insufficient, or contradictory — judged against the question, not by model confidence.
Insufficient or contradictory → say so, out loud
6
See the whole picture
The data must cohere into the whole picture the question requires — a coherent patient state, or a stable, cross-stratum-consistent cohort. Incoherence caps authority.
Incoherent picture → authority capped
7
Earn the authority level
Sufficiency + coherence jointly set which of six tiers is earned — of recommendation (patient) or of inference (population). Authority is earned, never presumed.
8
Show the work
The answer ships as a package: support, counter-evidence, assumptions, provenance, confidence, potential blind spots, and the conditions that would reverse it.
9
Learn under governance
Observation, interpretation, action, and outcome are recorded separately; durable knowledge — and versioned cohort definitions — is governed and auditable, improving the next cycle.
Six tiers of earned authority
T0 · SilentStates plainly what it cannot determine — and what would change that.
T1 · Data requestMay only identify what data is missing and ask for it.
T2 · ObservationMay surface interpreted findings without judgment.
T3 · ConsiderationMay offer possibilities with explicit uncertainty.
T4 · AdvisoryMay recommend or report, with full package, for review.
T5 · ActionableFull authority to recommend or conclude — all nine steps passed cleanly.
Failing a step does not halt the system — it caps the maximum tier, and never forces a false answer. A system that cannot see the whole picture may still request data or surface observations; it may not advise.
The honest exit
At any step, “I cannot determine this — and here is what would change that” is a complete, correct answer. For a population, its analog — “this dataset cannot answer this question” — is the guardrail against confidently-wrong analysis.