The Un-Ganglia: procedural failure modes as a design lens

direction · 2026-06-26 · 2026-07-05

Obsessive-compulsive disorder, addiction, perseveration, Parkinson's, and Huntington's read as failure modes of the brain's action-selection circuitry, each mapped to a design rule in an owner-held procedural memory. An analogy, not a clinical model.

Direction, not a result. This page makes no medical claim, and nothing here is diagnostic, therapeutic, or clinical.

Where it comes from

Section 15.4 of Borrowable Skills as Lean Un-Ganglia Subgraphs (doi:10.5281/zenodo.21205599) starts from a well-studied picture: the basal ganglia select which action is released and compile repeated sequences into habits. The same circuitry has a catalogue of failure modes, and each one has a structural counterpart in how a trained skill’s control flow is designed.

The failure modes

Obsessive-compulsive disorder. A loop re-fires and can’t reach a stop state. A skill walk runs under a composed termination bound, per-loop caps within a recursion limit, so it can’t loop without end.
Habit capture and addiction. Over-trained habits keep driving behavior after the goal has moved on. A skill fires only when its trigger matches the live goal, and a stale skill loses vitality and drops out of dispatch.
Perseveration. Behavior stays locked into a task-set the situation has outgrown. Dispatch re-selects on every turn against the current goal, so there’s no set to get stuck in.
Parkinson’s. Under-gating impairs the initiation of an intended action. Selection is a deterministic read of the trigger index, with no stochastic initiation step.
Huntington’s. Over-release produces unselected movement. No step runs above its authored autonomy ceiling, which bounds unselected escalation.

Alzheimer’s sits apart from this list. It’s a loss of declarative memory, not of action selection, and it’s where this layer meets the first paper: the brain can lose memory while overlearned procedures are comparatively spared, and a design that keeps both memory and procedure indelible spares both.

What this direction can’t claim

The mapping is an expository analogy, and the paper says so: it carries no experimental content. It doesn’t model any of these conditions, and it doesn’t transfer to patients. Its use is narrower: it names the failure modes the design rules out, and it gives each one a counterpart that can be inspected and tested in software.