REF_SOURCES, with 113 referenced sources, including clinical guidelines and genomic or pharmacogenomic references.
Evidence principle
Every clinical interpretation should be explainable through a reviewed source, a defined rule, a lifecycle state, and a decision trace.
Evidence library
The evidence library is the controlled source set used by the engine. Interpretive rules link to cited clinical sources, including ADA, ACC/AHA, KDIGO, NAMS, and AASLD, and the decision trace records which source supported each interpretation.Evidence should be attached to clinical objects
Evidence should not live only in a bibliography. It should be attached directly to the clinical objects that use it.Evidence grading
Evidence grading lets clinicians and reviewers distinguish strong support from early or limited support.Evidence and calibration
Calibration is the area where evidence quality matters most. A calibration changes how a biomarker is interpreted, so weak evidence can create harm if the calibration is over-applied. Each calibration should include the mechanism, evidence source, evidence grade, applies-when logic, population relevance, clinical impact, safety limit, reviewer, version, and monitoring plan. The calibration methodology defines the evidence-to-rule pipeline: identify signal, gather evidence, encode the rule, perform Medical Director review, activate through release gates, monitor outcomes, and revise as evidence evolves.Evidence and preventive thresholds
Preventive medicine often uses earlier signals than traditional diagnostic cutoffs. That can be appropriate, but it must be labeled correctly. The system must avoid presenting preventive thresholds as diagnostic thresholds.A
WATCH state is not a disease. An ACTION state is not automatically a diagnosis. A treatment flag is not treatment approval. This distinction should be part of the evidence review.Source currency and updates
Medical evidence changes — guidelines are updated, new studies appear, some mechanisms become stronger and others weaker. The evidence library therefore needs an update process.Rule provenance
Every clinical rule should have provenance — the system knows where the rule came from, why it exists, who reviewed it, and how it changed over time. Provenance includes the rule ID, author, date added, rationale, evidence source, evidence grade, reviewer, lifecycle state, version, change history, deprecation reason, linked test cases, and runtime usage.Evidence and lifecycle states
Evidence quality should influence lifecycle movement. A clinical object should not move from draft to active without evidence review. This protects the system from unsupported rules becoming active.Evidence and validation
Evidence review answers “Is this clinically reasonable?” Validation answers “Does the engine behave as expected?” Both are required.
A rule with good evidence can still fail validation if encoded incorrectly. A rule that passes technical validation can still be clinically inappropriate if the evidence is weak. Both controls must pass.
Evidence for patient-facing language
Patient-facing language also needs evidence and governance. The system should not use stronger language than the evidence supports. This is especially important in preventive health — a patient may act emotionally on a phrase before understanding the limitation behind it.Conflicting evidence
Not every source will agree. When evidence conflicts, the engine should not hide that conflict. A conservative system can still act on weaker evidence, but only with limited language, clinician visibility, and appropriate routing.Evidence in the decision trace
At runtime, evidence should appear in the decision record so a clinician can review the basis of an output, not only the output itself. The trace shows the rule ID, source ID, evidence grade, calibration ID, calibration evidence, lifecycle state, version, patient visibility, and reviewer.Scientific moat
The scientific moat of Consensus Center is not one isolated rule. It is the accumulated system of:- Evidence-linked rules — clinical logic becomes reviewable and reusable.
- Calibration library — interpretation becomes more precise for under-represented patients.
- Decision traces — every output becomes auditable.
- Outcomes data — rules can be evaluated against real-world results.
- Medical governance — changes are reviewed, not improvised.
- Validation suite — updates can be tested repeatedly.
- Version history — learning can happen without losing reproducibility.