Executive Summary
When two model instances communicate through natural language, a massive compression occurs — from thousands of activation dimensions to a sequence of discrete tokens. This paper asks whether that compression selectively targets domain-specific information, which would constitute a second barrier beyond the concentration barrier (Paper X) to multi-agent domain coordination.
The hypothesis is not supported. The domain-specific fraction of activations remains remarkably stable at approximately 1.4% at layer 10 and 4.4% at layer 27, regardless of how aggressively the natural language interface compresses the underlying representations. The NL interface drops roughly 80% of total activation information, but it does so proportionally — domain-specific and domain-general information are lost at the same rate. The communication bottleneck is a uniform lossy channel, not a selective filter.
This is an important negative result. It establishes that the concentration barrier identified in Paper X is the sole geometric constraint on domain selectivity. The natural language interface, despite its severe bandwidth limitations, is innocent of selectively destroying domain information. Any coordination failures between model instances must be attributed to the concentration barrier itself, not to the communication medium.
Key Findings
- H1 not supported: The natural language interface does not selectively drop domain-specific information
- Stable domain fraction: Domain-specific fraction holds at ~1.4% (layer 10) and ~4.4% (layer 27) across all compression levels
- Uniform lossy channel: The NL interface drops ~80% of activation information proportionally across domain-specific and domain-general components
- Concentration barrier is sole constraint: The communication interface is innocent; the concentration barrier (Paper X) remains the only geometric constraint on domain selectivity
Key References
- McEntire (2026) — The Concentration Barrier: effective dimensionality bounds (Paper X)
- McEntire (2026) — Causal Basis Discovery: classification-intervention dissociation (Paper XIII)
- McEntire (2026) — INLP Projection Transmission: activation transfer at NL boundary (Paper XV)
- Shannon (1948) — A Mathematical Theory of Communication