Calibration

Most datasets reduce calibration to a boolean. The manifest replaces it with a maturity gradient that separates what an operator actually does today from the traceable reference it has identified:

Alongside status, an entry can carry currentMethod (what is actually done), traceableReference and referenceInUse (the reference and whether it is practice or aspiration), cadence, uncertaintyBudget, and notes. All nullable — honest nulls are first-class, so calibration gaps stay visible and comparable across operators.

The scoring feedback loop

@disclosureos/scoring exposes the gradient as a trust input. calibrationTrust builds an evaluatorWeight policy that credits claims whose sensor evidence resolves — through SensorReading.sensorRef — to a manifest entry with real calibration provenance:

import { score, calibrationTrust, CALIBRATION_TRUST_WEIGHTS } from '@disclosureos/scoring';

const result = score(observation, {
  evaluatorWeight: calibrationTrust(observation, [manifest]),
});

CALIBRATION_TRUST_WEIGHTS;
// { unresolved: 1, none: 1, candidate_identified: 1.05, in_practice: 1.15, documented: 1.25 }

The default is deliberately conservative:

• Nothing weighs below the baseline. Sensors with no manifest entry (and manifests that declare none) keep the 1.0 baseline. Publishing provenance can only add weight.
• Only the consensus point shifts. Like every evaluatorWeight policy, the range and contested flag still report the honest spread of what evaluators actually asserted.
• The gradient is overridable per call, and composes multiplicatively with any base evaluator-trust policy.