Why we tried it
Test whether adaptation during actuator shifts creates a cost after native dynamics return.
What the evidence shows
The locked mixture improved 7.68% over an old adaptive reference but regressed 24.02% against frozen prediction; a trained calibrator also failed joint gates.
Where the work stands
The known positive diagonal-gain family and upstream backbones are strong priors; shifted, restoration and normal-regime guards were not jointly met.
Possible next steps
Freeze joint criteria, retain strong frozen/adaptive references, and test a distinct out-of-family hypothesis against established forgetting or mixing mechanisms.
Suggested next steps are untested research questions, not new experimental results. The archive records manuscript provenance, licensing and status.