Blind Ranking

Scoring method where responses are anonymized and shuffled; tests whether scorer rankings are real across five independent scorers

Neighborhood — ranked by edge-count

method

Koan Battery
uses
Assessment framework for measuring introspection and self-observation in LLMs; grounded in Janus's architectural theory.

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.

Blindsightconcept0.739
Example of biological degeneracy: visual responses mediated by subcortical and brainstem nuclei independent of cortex; supports multiple realizability of cognitive functions.
Blind Spotconcept0.737
refusal rateconcept0.708
The percentage of harmful requests that a model refuses to answer, a common safety metric.
Rank-one matrixconcept0.704
The simple matrix form into which VPD constrains subcomponents to enforce mechanistic simplicity.
Mind Blindnessconcept0.697
hierarchyconcept0.688
An ordering of texts via spatial cues like indentation, size, and placement, implying importance.
Probe-based ranking outperforms gradient-based and LLM-judge methods at lower costclaim0.682
Authors' claim that their approach is both more effective in reduction and cheaper than prior methods.
Cosine Similarity Ranking for Instruction Discoverymethod0.680
Method to discover new reflection-inducing instructions by ranking candidate tokens by cosine similarity to steering vectors.