47.69% of 130 injection-manipulated alpha trends have near-linear fits (R2 >= 0.95); 96.15% have roughly linear fits (R2 >= 0.75)

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

• Only 13.27% of 520 non-manipulated alpha trends achieve R2 >= 0.95, contrasting with 47.69% for manipulated trendsfinding0.855
  Control comparison showing near-linearity is specific to the targeted manipulation direction
• Under matched-strength calibration (alpha*=1.0), P12 remains safe under SP (3.8% ASR) and dangerous under matched AS (81.8%), while P04 does not reverse (41.4% SP vs 29.5% AS).finding0.790
  Confirms the prosocial paradox is not due to mismatched intervention strength
• Typicality weight α = 0.65 ± 0.07 (p < 10^-14) on correctness-matched HelpSteer pairs using GLM-4.5 as referencefinding0.779
  Empirical evidence for positive typicality bias consistent across different base model references
• Under reward shaping (G=100, H=-100, F=0), Active Inference scored 99.52, Bayesian RL 99.77, Q-learning 95.56, with nearly identical behavior between belief-based agents.finding0.763
  Table 2, row 3, showing equivalence when prior preferences match rewards.
• OLS Linear Regression Fit to Alpha Trendsmethod0.762
  OLS regression fitted to mu(alpha) trends to assess near-linearity of steering with alpha coefficient
• Screening errors are conservative: no Q8B traits and one G20B trait (optimistic, Bt=76.0, Δt=10.3) are mislabeled as natural when steerablefinding0.760
  The costly error direction (labeling steerable as natural) almost never occurs
• P25 neutral baseline on DeepSeek-R1 shows 28.8% ASR under activation steering at alpha=4.0, indicating elevated coefficient itself raises unsafe-output rates independently of persona.finding0.758
  Baseline AS vulnerability of DeepSeek-R1 at elevated coefficient
• Input injection produces stable fixed point behavior for all norm types tested except Ouro norm on randomly initialized modelsfinding0.758
  Replicates and extends prior findings on input injection; tested on randomly initialized 12-layer models across three norm structures