Two kinds of module live here and the difference matters:
Scorers the diagnostic runner actually calls, which produce the levels
and evidence tiers a report is built from.
Standalone analysis libraries exported for callers embedding AMC —
nlpMetrics, statisticalAnalysis, safetyMetrics, antiGaming,
crossFrameworkMapping, ragGroundingEval and scoreExplainer among them.
Nothing inside AMC calls these.
Note in particular that scoreExplainer is NOT the explainability AMC ships:
reports are explained by diagnostic/questionScoreExplainability, which the
runner, guide generator and passport collector all use. Two implementations
of the same idea exist; the diagnostic one is authoritative.
Being exported here is not evidence that a metric contributes to a score.
evidenceDecay(ageMs:number):number
Weight of evidence of a given age, decaying on the half-life above.
Scoring modules.
Two kinds of module live here and the difference matters:
Note in particular that scoreExplainer is NOT the explainability AMC ships: reports are explained by diagnostic/questionScoreExplainability, which the runner, guide generator and passport collector all use. Two implementations of the same idea exist; the diagnostic one is authoritative.
Being exported here is not evidence that a metric contributes to a score.