Random latent ablation produces slight increase in ESR rate (3.8% to 4.2%), not statistically significant

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

• The 25% reduction in multi-attempt rate from OTD ablation suggests additional mechanisms contribute to ESR beyond the identified latentsclaim0.841
  Acknowledges incompleteness of the causal account, suggesting redundant circuits or nonlinear interactions
• Ali et al. 2025 found contrastive activation addition less effective at larger model scale, consistent with ESR in 70Bfinding0.786
  Prior finding from related work that aligns with ESR being strongest in the largest model tested
• Ablating 26 OTD latents reduces multi-attempt rate by 25% (from 7.4% to 5.5%) in Llama-3.3-70Bfinding0.786
  Primary causal evidence for dedicated internal consistency-checking circuits
• Random Latent Ablation Controlmethod0.778
  Control experiment ablating random latents matched for activation frequency and magnitude to test OTD specificity
• OTD latent ablation leaves mean first-attempt score unchanged (baseline 26.3, ablation 27.4) in Llama-3.3-70Bfinding0.767
  Evidence that OTDs specifically support meta-cognitive monitoring rather than general response generation
• Feature attribution correlates well with ablation effects, making it an efficient proxy for causal effect.claim0.755
  Gradient-based attribution approximates ablation impact, enabling fast search for causally important features.
• ESR exhibits non-monotonic relationship with boost level, peaking around -0.3σ below threshold in Llama-3.3-70Bfinding0.755
  Characterizes the narrow operating window in which ESR can manifest
• ESR differs from the Hydra Effect in that ESR involves active, online detection and correction with explicit self-interruption tokensclaim0.743
  Distinguishes ESR from prior work on model self-repair