The Delta Is Not the Skill: Zero-Prior Elicitation Transfers LoRA Behavior Across Model Families

Reusing low-rank adapters (LoRA) across model sizes, generations, and architectural families is increasingly essential for deploying specialized capabilities. However, native retraining requires inaccessible training data, while existing synthetic-data transfers rely on task-specific priors (task names, seed examples, labels, or auxiliary discriminators). Weight-space transfer avoids these priors, but we uncover a sharp dichotomy: within a shared parameter basis, zero-edit adapter transplantation is near-lossless (recovering $92$–$116\%$ of native gains), whereas across distinct families, weight-space routes collapse ($\mathrm{TRR}\le 0.08$) because low-rank deltas are base-relative. What survives across model boundaries is behavior, not weights. We introduce *zero-prior skill elicitation*, a framework that extracts adapter capabilities without task identities, seed data, labels, discriminators, or white-box access. By identifying inputs where adapter and base generations diverge, evolving these probes to broaden coverage, and distilling the adapter-equipped teacher's completions into students, we transfer complex skills purely through text. Distilling from a single 397B MoE teacher LoRA matches or exceeds native supervised fine-tuning across four student models in two architectural families, reducing target training compute by $2.6$–$3.0\times$. Gains are pronounced on multi-step reasoning (elevating 9B MATH-500 by $+27.2$ points; $\mathrm{TRR}=3.39$), competition mathematics, and out-of-distribution instruction following, with zero verbatim $n$-gram benchmark leakage and successful generalization to Granite models.