Cluster bootstrap confidence intervals

Bootstrap resampling at conversation level (B=1000, 95% percentile CIs) to respect non-independence of within-conversation observations

Neighborhood — ranked by edge-count

framework

Quantitative Introspection Framework
implements
The paper's central contribution: treating LLM numeric self-report as a quantitative signal validated against probe-defined internal states with causal confirmation via steering

method

bootstrap confidence interval
related_to
Used to report uncertainty for geometry summaries and effect sizes.

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.

Wilson Score Confidence Intervalmethod0.738
Used to compute 95% confidence intervals for sticker-removal success probability
Bootstrapping of centersframework0.733
The recursive process by which centers generate life through mutual intensification, where each center's life depends on the life of others.
Behavior Clusteringconcept0.729
Grouping similar model behaviors; the unsupervised method surfaces clusters of concerning patterns.
Cluster 5 (5-cluster): SIMPLICITY AND INNER CALM, ECHOES, GRADIENTSfinding0.716
Fifth cluster of the five-cluster grouping, containing properties 14, 12, 10.
Cluster 2 (4-cluster): LOCAL SYMMETRIES, THE VOID, LEVELS OF SCALE, GOOD SHAPE, POSITIVE SPACEfinding0.714
Second cluster of the four-cluster grouping, containing properties 7, 13, 1, 6, 5.
Cluster 4 (4-cluster): SIMPLICITY AND INNER CALM, ECHOES, GRADIENTS, POSITIVE SPACEfinding0.712
Fourth cluster of the four-cluster grouping, containing properties 14, 12, 10, 5.
Bootstrap Relationconcept0.711
The mutual support among centers where each raises the others to life, with no primary elements.
Cluster 2 (5-cluster): ROUGHNESS, ALTERNATING REPETITION, GOOD SHAPEfinding0.710
Second cluster of the five-cluster grouping, containing properties 11, 4, 6.