Orthogonal Signal: A Structural Theory of Anti-Convergence in Multi-Agent AI Systems

This white paper proposes orthogonal signal as a formally definable, measurable primitive for human-AI co-evolution. It argues that closed agent-to-agent loops undergo predictable semantic convergence and emergence collapse, while genuinely foreign (human) constraint architecture keeps multi-agent systems generative. The paper introduces a four-type novelty taxonomy, a Constraint Origin typing scheme, a two-phase wave function for human novelty contribution, a four-role Trust Topology with a hard Non-Domination gate, and a Predictive Temporal Horizon Clock estimating a system’s runway before irreversible crystallization. It reports on the orthogonal-signal reference implementation (150 passing tests) and comparative simulation evidence across closed-loop, noise-injected, and orthogonal-injected conditions. The paper frames this as an architectural research agenda rather than a validated scientific theory, and extends the author’s companion Coheronmetry framework (Elion, 2026).