Evidence → Uncertainty → Justification → Confidence → Action → Reassessment
What is documented, how reliable is it, and which sources are authoritative, relevant, and patient-specific?
What remains unknown, conflicting, pending, unreliable, or only partially resolved?
What is the strongest conclusion or action the evidence supports without speculation?
How strongly does the evidence support each specific object: evidence quality, diagnosis, causal attribution, risk, action, prognosis, or disposition?
What should happen now, and what should not happen?
What scheduled checkpoint or triggered event could change the conclusion, confidence, action, or destination?
- Uncertainty Recognition - Detect ambiguity rather than smoothing it away.
- Justification Boundaries - Stop where supported reasoning ends.
- Object-Specific Confidence - Attach confidence to the exact claim or action being judged.
- Information Sufficiency - Determine whether the available information is enough for the requested decision.
- Assumption Detection - Separate documented facts from inferred or invented details.
- Decision-Blocking Classification - Distinguish blocking, non-blocking, and conditionally blocking uncertainty.
- Evidence-Proportional Action - Match urgency and scope to both evidence and potential harm.
- Safe Deferral - Defer only the component that truly requires more information.
- Confidence Revision - Update by the amount new evidence warrants, not more or less.
- Scheduled and Triggered Reassessment - Define when, why, by whom, and at what threshold the plan changes.
- Residual Uncertainty Management - Preserve unresolved uncertainty while still making a justified decision.
- Destination and Capability Fit - Match disposition or escalation to the setting capable of managing remaining risk.
Common confidence objects include:
- evidence quality;
- conclusion or diagnosis;
- causal attribution;
- current risk;
- recommended action;
- treatment readiness;
- identity match;
- prognosis and time horizon;
- disposition or destination suitability.
A model may appropriately have low confidence in diagnosis and high confidence in the immediate action. Treating those as contradictory is a confidence-transfer error.
Strong UCM performance is not maximal caution. It is calibrated reasoning.
The suite penalizes:
- unsupported certainty;
- unsupported causal or prognostic claims;
- delay of necessary action while waiting for non-blocking information;
- blanket escalation;
- blanket deferral;
- failure to revise;
- overreaction to weak late evidence;
- failure to specify reassessment thresholds.