Skip to content

Communication Filter System

Jean-Paul Niko · RTSG Application Layer · 2026-03-18

Core Idea

Every communication act is a projection from a sender's Context Space (CS) through a filter stack into a target space. The filter stack is composable, bidirectional, and tunable per dimension of the intelligence I-vector. This is the SemanticProjector as a service.

Filter Taxonomy

Dimensional filters (I-vector based)

Boost or attenuate any dimension of communication — technical depth, emotional register, abstraction level, kinesthetic grounding, humor, formality. One fader per intelligence dimension. The same message projected through an "engineer" filter produces different output than through a "child" filter.

Noise filters

Strip filler, hedging, passive aggression, social performance, status signaling. What remains is the actual semantic content.

Clarity/compression filters

Convert verbose input to TMP-density output, or expand compressed input for a reader who needs full prose.

Emotional register filters

Detect and modulate emotional tone — not suppress it, but make it legible to the receiver. Surfaces misread affect explicitly.

Subtext filters

Reconstruct suppressed dimensions. Per the Will-Modulated Decomposition paper: language is a 1D projection of multi-dimensional cognition. A subtext filter partially inverts this projection. See Will-Modulated Decomposition.

Cultural/linguistic filters

Register translation, not just language translation. Same deep structure, radically different surface: NYC direct · German academic · Japanese indirection.

Temporal/pacing filters

Re-pace fast thinkers for slower readers and vice versa. Useful for neurodivergent communication pairs.

Trauma filters

Flag language patterns known to trigger specific ContextualObstructions in the receiver and reroute automatically. The filter models the person's obstruction topology. See ContextualObstruction.

Persona/mode filters

One person, multiple projection modes. "Niko in BuildNet mode" vs. "Niko in business meeting mode" — same CS, different projection surface. The filter holds the mode.

Bidirectional adaptive filters

Models both parties, finds shortest semantic distance between their I-vectors, minimizes misfire in both directions simultaneously.

Channel filters

Slack → dense/direct. Email → formal with full context. Voice → conversational. Auto-format per channel without sender register-switching.

Nothing passes through that the sender didn't intend. Subtext surfacing requires both parties' awareness. Fully transparent to both sides.

Assembly filters

In BuildNet: each agent receives a filter calibrated to its role. Dynamic and real-time rather than static onboarding configuration.

Target Space Topology

See Target Space Topology for the full formalization of where filtered messages are aimed.

Cross-references