Executive Summary
Papers XV-XVII established that activation-level transmission between model instances faces fundamental barriers: orthogonal encoding (Paper XV), information type dissociation (Paper XVI), and zero trajectory content without text scaffolding (Paper XVII). This paper pivots from activation-level approaches to contextual strategies, testing whether coordination can be achieved through the text channel itself.
Four strategies are compared: priming selection (choosing which text to show the receiver), vocabulary induction (introducing domain-specific terminology), activation injection (direct activation transfer), and Socratic scaffolding (structured question-answer sequences). Priming selection dominates, closing 48.9% of the coordination gap — 5.4x better than activation injection (9.1%) and 2.5x better than vocabulary induction (19.5%). Socratic scaffolding actually worsens coordination by 19.9%, making it counterproductive.
The mechanism behind priming selection's success is unexpected. The domain match rate between selected primers and target tasks is only 15.6%, meaning the benefit does not come from domain-specific knowledge transfer. Instead, priming selection works by aligning the receiver's processing style — its attention patterns, inference strategies, and output distribution shape — with the sender's. The effect is strongly domain-dependent: medical and science domains see 86% gap closure, while legal and code domains see only 22-27%, suggesting that some processing styles are more transferable through text than others.
Key Findings
- Priming selection dominates: 48.9% gap closure, 5.4x better than activation injection (9.1%)
- Vocabulary induction moderate: 19.5% gap closure, a distant second to priming selection
- Socratic scaffolding harmful: Worsens coordination by 19.9%, actively counterproductive
- Processing style, not domain matching: Domain match rate only 15.6% — mechanism is alignment of inference strategies, not knowledge transfer
- Domain-dependent efficacy: 86% closure for medical/science, only 22-27% for legal/code
Key References
- McEntire (2026) — Sender Continuation Perplexity: text carries trajectory, activations do not (Paper XVII)
- McEntire (2026) — Ritual Shape: structural features of effective coordination (Paper XIX)
- McEntire (2026) — Full Mind Transfer: bandwidth vs fidelity (Paper XVI)
- McEntire (2026) — INLP Projection Transmission: activation transfer limitations (Paper XV)