Easy questions (acc > 80%) have average reflection rate of 25.8% for DeepSeek-R1 Llama 8b on GSM8k

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

• DeepSeek-R1 Llama 8b gains 0.16% accuracy on GSM8k with positive intervention (more reflections) at cost of ~2000 additional tokensfinding0.870
  Only model showing marginal benefit from increased reflection, at substantial token cost
• Reflection direction features achieve AUROC 0.772 vs. 0.736 for final layer baseline on deepseek-llama-8b on GSM8k correctness predictionfinding0.834
  Supports claim that uncertainty is encoded in reflection direction
• No Reflection with 'Answer' achieves accuracy .037 on gsm8k_adv for Qwen2.5-3Bfinding0.795
  Baseline accuracy when reflection is suppressed.
• DeepSeek-R1 Llama 8b accuracy on MMLU Professional Accounting drops from 56.5% at baseline to 50.1% at intervention -0.96finding0.785
  Shows smaller models are more sensitive to reflection reduction on non-math tasks
• All five judge models consistently rank Llama-3.3-70B as having substantially higher ESR rates than other modelsfinding0.771
  Cross-judge validation of the primary ESR finding across OpenAI, Alibaba, Anthropic, and Google judge models
• Llama-3.3-70B shows multi-attempt rate of 7.4% vs. ≤1.2% for all other models testedfinding0.770
  Supporting finding showing ESR is driven by both higher multi-attempt rates and comparable improvement rates
• Triggered Reflection with 'Alternatively' achieves accuracy .684 on gsm8k_adv for Gemma3-4B-ITfinding0.764
  Highest single-instruction accuracy result in the paper.
• Llama 3.1 405B shows 14% compliance gap in minimal helpful-only replication but smaller Llama and Mistral models show no gapfinding0.761
  Replication across open-weight models supports scale-emergence finding