§24's resonance score, the documented formula: the AUTHORISED CUT (which already folds in
the measured deviation, the excess threshold and the operator's amount — see
harmonicDeviationGainsDb) weighted by how sure the detector is of the voice and of the
partial's attribution.
The other two §24 inputs are GATES upstream rather than weights here: temporal stability
is binary (§13 — only a stable voice produces candidates at all), and an ambiguous
attribution is unmeasurable (§11/§54 — it never becomes a deviation, so it never scores).
Any unusable input scores zero: a NaN weight would poison every comparison it meets.
§24's resonance score, the documented formula: the AUTHORISED CUT (which already folds in the measured deviation, the excess threshold and the operator's amount — see
harmonicDeviationGainsDb) weighted by how sure the detector is of the voice and of the partial's attribution.The other two §24 inputs are GATES upstream rather than weights here: temporal stability is binary (§13 — only a stable voice produces candidates at all), and an ambiguous attribution is unmeasurable (§11/§54 — it never becomes a deviation, so it never scores). Any unusable input scores zero: a NaN weight would poison every comparison it meets.