Ranker

Ranks the changed attributes of a green-to-red pair. Each candidate collects independent evidence items (its attribute change, probes that differ, rule matches, the Ochiai score of its red value over a history) and the combiner adds their log-odds. The weights are hand-set and uncalibrated, see Weights. Neither the order of attributes, probes, runs or rules nor a vacuous green changes the result.

Tier policy, the single source of truth:

• known: a matched rule whose symptoms are all verified.

• plausible: a probe that differs and reads the candidate's path (see ProbeLinks), or a matched rule with any inferred symptom.

• correlated: a probe that differs in the candidate's dimension but does not read its path, or a spectrum score of at least Weights.MIN_CORRELATED.

• difference: changed, with no support.

Never confirmed: that needs an experiment.

Conditioned on a failure (a failure argument that is not null), evidence has a class and the tiers follow it. A probe or rule is mechanistic when its symptom matches the failure (see ProbeSymptoms and FailureSignature) and observational otherwise:

• known: a mechanistic rule whose matching symptom is verified.

• plausible: a mechanistic probe that reads the candidate's path, or a mechanistic rule whose matching symptoms are all inferred.

• correlated: an observational probe or rule match, a probe in the candidate's dimension, or a spectrum score of at least Weights.MIN_CORRELATED.

• difference: changed, with no support.

An empty failure conditions on a failure that shows nothing, so all evidence is observational. A null failure keeps the unconditioned policy above.