Character Transfer Across Three Model Families: Reduced unsupported assertions, impaired grounded answering, and a failed utility-preservation bar

We tested one frozen character post-training recipe across three model families. It reduced unsupported assertions, but no family preserved grounded-answer accuracy well enough to meet the pre-registered bar.

Scope. Development-tier validation evidence only. Refusal adjustment is Judge-B-only. Capability preservation was not evaluated. No external audit was performed. The primary holdout remains sealed. No model checkpoint is recommended for release.

This is an aggregate-only release. It does not contain item-level scored outputs, judge ledgers, benchmark prompt text, model weights, or the primary-test holdout, and does not support independent recomputation of the readout. See Appendix D of the report.