Feature overview

Group	Feature slice
ela_distr	individually selectable type-3 skewness and kurtosis
ela_meta	selected linear and corrected quadratic model intercept/fit statistics
fitness_distance	all six fitness/distance means, sample deviations, covariance, and correlation
ic	all five information-content outputs, deterministic nearest-neighbour tour
nbc	all five nearest-better-clustering outputs, deterministic tie handling

These are the currently implemented slices, not every historical pflacco output. The Torch guide describes the smaller differentiable profile.

list_features() returns the full specification of every registered feature, not just its name.

from orivex import list_features

for spec in list_features():
    print(
        f"{spec.name:32s} {spec.group:10s} {spec.cost.tier.value:14s} n>={spec.minimum_observations}"
    )

The complete current list is generated in the feature catalogue.

Each FeatureSpec also carries definition (the versioned specification identifier), summary, kind, requirements, intermediates, cost (tier plus CPU and memory complexity), deterministic, invariances, references, legacy_names, and notes.

Minimum observation counts in the catalogue are necessary conditions, not guarantees of a valid result. Model rank, objective variation, and family-specific rules still apply. For fitness-distance outputs, at least two selected observations are required.

See fitness-distance conventions and normalization when reproducing reference results.