Why we tried it
Verify that probing affects actions and whether hidden-dynamics information creates net task value.
What the evidence shows
The default information weight lay below the switching threshold; paired diagnostics found no stable unseen advantage. Net-value repair passed exact fixtures but matched full Bayes.
Where the work stands
Action changes, lower parameter error and gains over a non-probing reference do not establish a contribution over strong baselines.
Possible next steps
Test a useful computable sign test for net probe value under model error and finite planning budgets; end the candidate if it fails.
Suggested next steps are untested research questions, not new experimental results. The archive records manuscript provenance, licensing and status.