AUROC

Performance metric for binary classification; used to evaluate pathogenicity prediction.

Neighborhood — ranked by edge-count

paper

method

Cross-task generalization evaluation
implements
Measuring AUROC of a probe trained on one task when evaluated on another task to assess universality.

cosine ≥ 0.65 · no typed edge

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

AUPRC Latent Activation Classifiermethod0.730
Using per-prompt average SAE latent activations and area under precision-recall curve to discriminate aligned from misaligned models
AutoMecomethod0.677
Automated benchmarking framework for evaluating LLM meta-cognition, mentioned as related work.
AITHYRAinstitute0.662
Additional institutional affiliation for Bronstein
Anthropicinstitute0.655
Lab behind Claude models and Constitutional AI training approach; represents highest baseline scores and lowest prompt lift.
5-fold Cross-Validated Logistic Regression AUCmethod0.654
Classification-based comparison of interpretation abilities across IIT metrics and Span Representation for ToM score categories.