Without system prompt, only approximately 0.02% of Huginn-0125 tokens exhibit non-fixed-point (orbit/slider) behavior on GSM8k

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

• Long Persona system prompt increases non-fixed-point token percentage to 0.14% for Huginn-0125, while a length-matched padding prompt produces only 0.05%finding0.818
  Shows that semantic content (not just length) of system prompt influences occurrence of orbit/slider behavior
• Stages of inference metrics (ColSum concentration, mixing score) remain broadly consistent even for the GSM8k prompt exhibiting the largest orbit amplitude in Huginn-0125finding0.789
  Demonstrates robustness of inference stages to non-fixed-point limiting behavior
• Huginn-0125 fails to develop stages of inference because its repeated residual stream normalization prevents the growth in activation magnitude required for compression behavior and sink formationclaim0.773
  Mechanistic explanation for the negative result observed for Huginn-0125
• Huginn-0125 performance remains constant when extrapolating beyond training recurrences, while Ouro performance deteriorates in the same regimefinding0.767
  Correlates stable fixed-point behavior with out-of-domain generalization performance at test-time
• Huginn-0125 does not demonstrate clear stages of inference due to repeated normalization of the residual stream suppressing massive activationsfinding0.760
  Negative result linking norm architecture to absence of inference stages; contrasts with Ouro and retrofitted models
• Retrofitted Llama exhibits 0% non-fixed-point token behavior under all tested system prompt conditionsfinding0.744
  Input injection fully prevents non-fixed-point limiting behavior in the retrofitted Llama model
• Ouro 1.4B, Retrofitted Llama, and Huginn-0125 exhibit diagonal patterns in Frobenius norm heatmaps confirming cyclic fixed point behavior across 8 recurrencesfinding0.738
  Empirical validation that attention patterns are most similar to same-layer outputs across different recurrences
• Ratio of unmonitored to training non-compliance increases from 1.14 to infinity for default prompt and to 66 for OpenAI variant post-RLfinding0.727
  Demonstrates severity of training-deployment gap after RL