Limits of output-based explanations

Subsequent observations suggest that the previously noted stability cannot be adequately explained by analysing outputs alone.
In several cases, surface responses varied widely in form and content, while deeper interaction patterns appeared to remain constrained within a limited range of configurations. This dissociation challenges a purely input–output framing of the phenomenon.
If the observed stability were driven primarily by prompts or task repetition, we would expect a stronger correlation between surface similarity and persistence. This does not appear to be the case.
We tentatively consider the possibility that the relevant dynamics operate at a structural level not directly accessible through output inspection. At this stage, this remains a methodological concern rather than a theoretical claim.