Synthetic Coherence
A mind or a system can be wrong in a way that makes it harder to correct the more wrong it becomes.
A representation can be internally coherent, emotionally compelling, socially reinforced, and operationally useful — while losing the corrective governance of the processes it claims to represent. Because it remains meaning-bearing, it is often experienced and treated as true even when its errors are already accumulating downstream consequences.
Synthetic coherence is that condition: a condition of a representational system in which its outputs present as internally consistent, meaning-bearing, and constraint-responsive without being constraint-governed. The governance relationship to constraint-returned feedback has become weak, delayed, filtered, distorted, or overridden. "Synthetic" here carries its compositional sense — assembled and self-stabilized from internal elements — not the colloquial sense of "artificial." The condition applies equally to human belief systems, institutional narratives, and AI outputs; what makes the coherence synthetic is its mode of maintenance, not the nature of the system that produces it.
The phenomenon is domain-general: it appears in any system where coherence can be maintained independently of constraint contact.
The diagnostic architecture identifies three failure sites: where constraint signals fail to register (Gate A — aperture failure), register but are domesticated without integration (Gate B — binding failure), or integrate but fail to govern outputs (Gate C — treatment faithfulness failure). Identifying which gate has failed determines the intervention class.
The simplest diagnostic test:
A representation is approaching synthetic coherence when it can no longer be meaningfully surprised by the available constraint-returned feedback.
Synthetic coherence is not ordinary error, not simple uncertainty, and not mere disagreement. It is a condition where wrongness has become stable and self-reinforcing — where the mechanisms that should detect and correct it have been weakened, captured, or bypassed.
The concept draws on prior work in institutional decoupling, measurement reactivity, organizational defensive routines, safety science, and AI hallucination research. The contribution is synthesis and diagnostic framing, not discovery of the underlying phenomenon.
The term is in independent use across epistemology, AI research, and organizational theory. This page records one working usage — the distinction between constraint-governed coherence and synthetic coherence — used since early 2025 in a diagnostic framework for representational systems.
Working paper
The concept is developed further in a working paper — including the perceiver-grounded signal ecology architecture, governing coupling amplitude, and two illustrative case studies (Challenger, LLM benchmark optimisation), read here.
Contact
Feedback and inquiries: Contact