Does the model internally maintain a form of 'consistency score' or probability mass over coherent reasoning trajectories, and how is this score modulated during reflection?

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

• a model becomes strongly confident in its final answer, but continues generating tokens without revealing its internal beliefquote0.803
  Core definitional quote for performative chain-of-thought
• As larger models develop more coherent reasoning, internal consistency pressures may generalize learned honesty to new contexts beyond the training distributionclaim0.795
  Hypothesis about scale-dependent generalization of SOO-induced honesty
• When does the model initiate reflection during its reasoning process?question0.779
  First central research question motivating ReflCtrl investigation
• Any system that persists must minimise surprisal, thereby gathering evidence for its own generative model, a process known as self-evidencing.claim0.772
  Foundational claim of the paper, defining self-evidencing.
• Coherence maximization across simultaneously active mental models is related to prediction error minimization in the FEP, but the relationship is one of compatibility rather than strict equivalenceclaim0.770
  CIMC's position on the relationship between its coherence hypothesis and Friston's FEP
• We hypothesize that explicitly instructing the model to evaluate the correctness of the given statement may change the geometry of truth directions.hypothesis0.768
  Motivating hypothesis for Section 5's investigation of prompt template effects.
• Chinese models share contemplative posture (engaging self-referentially rather than deflecting) with Claude through shared values in training data rather than trace distillation from a specific model.claim0.766
  Exploratory interpretation of Chinese model performance under contemplative prompt
• For small models, critiqued revisions yield higher harmlessness PM scores than direct revisions; for large models the difference is negligible.finding0.765
  Figure 7 comparison of critiqued vs direct revisions across model sizes.