Multi-attempt improvement rate peaks at 83% around -1.0σ below threshold in Llama-3.3-70B

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

• Llama-3.3-70B shows multi-attempt rate of 7.4% vs. ≤1.2% for all other models testedfinding0.893
  Supporting finding showing ESR is driven by both higher multi-attempt rates and comparable improvement rates
• Meta-prompting increases Llama-3.3-70B multi-attempt rate 4.3× (from 7.4% to 31.7%)finding0.849
  Demonstrates ESR can be deliberately enhanced through prompting in the largest model
• Fine-tuning Llama-3.1-8B on self-correction examples increases multi-attempt rate proportionally with training data ratiofinding0.847
  Shows behavioral pattern of self-correction is trainable in smaller models
• Multi-attempt rate peaks at 2.7% around -0.3σ below threshold in boost sweep experimentfinding0.814
  Quantitative characterization of ESR operating regime in boost level sweep
• Ablating 26 OTD latents reduces multi-attempt rate by 25% (from 7.4% to 5.5%) in Llama-3.3-70Bfinding0.812
  Primary causal evidence for dedicated internal consistency-checking circuits
• Llama-3.3-70B corrected response scores 75/100 rather than 100 due to residual steering effects (Snell's law reference)finding0.797
  Illustrative finding that ESR mitigates but does not fully eliminate steering influence
• LLaMA-3.1-8B: Sbmax = -1.896 ± 0.211, AUSN = -2.119 ± 0.198, peak layer ℓ* = 10 (median)finding0.797
  Seed-pooled geometry-only statistics (per-dev z units).
• OTD latent ablation leaves mean first-attempt score unchanged (baseline 26.3, ablation 27.4) in Llama-3.3-70Bfinding0.777
  Evidence that OTDs specifically support meta-cognitive monitoring rather than general response generation