Lower (more central) PCs of emotion feature activations are more persistent than higher-rank (noisier) PCs in both Kimi and Cogito, above variance-matched baselines.

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

• Persistence is not an artifact of probe construction because lower (more central) emotion PCs are more persistent than noisier high-rank PCsclaim0.835
  Rules out measurement artifact explanation for the persistence finding
• Emotion probe persistence correlation of 0.214 in Cogito v2.1 vs 0.099 for random vectorsfinding0.802
  Quantifies emotion feature persistence above random baseline in Cogito across 240 multi-turn conversations
• Emotion probe persistence (token-0 to token-100 correlation) in Cogito v2.1 is 0.214, compared to 0.099 for random unit vectors in 7168D space.finding0.793
  Quantitative measure of emotion feature persistence vs random baseline in Cogito
• Cogito emotion probe residual autocorrelation +0.077 above variance-matched controls (p=1.5e-27, 157/171 probes positive)finding0.792
  Demonstrates that Cogito emotion probes are persistently active beyond what is explained by their variance alone
• Emotion features in LLMs are genuinely more persistent than variance-matched random features, indicating stateful emotional encoding beyond autoregressive dynamicsclaim0.781
  Central interpretive claim of the paper supported by multiple convergent analyses
• SAE feature emotion subspace overlap correlates with persistence in Cogito: Spearman +0.413, p=4.4e-196finding0.780
  Demonstrates that SAE features more aligned with the emotion subspace are more persistent in Cogito after variance control
• SAE emotion subspace overlap correlates with variance-residualized persistence in Cogito: Spearman +0.413, p = 4.4e-196.finding0.780
  Strong positive relationship between emotion alignment and SAE feature persistence in Cogito
• Higher-activating feature intervals are systematically more interpretable than lower-activating intervals in human analysisfinding0.769
  Shows interpretability correlates with activation strength, most model effect comes from high activations