Module Architecture

This page is structural, not exhaustive. Module specifications are in the implementation notes.

State estimation

The what is happening? layer. It observes and interprets system movement without directly forcing the system back into a preferred state. The tracker estimates; it does not compel.

Configuration tracker — estimates which MSSC is currently active; detects whether the system is remaining within it, drifting away, transitioning toward another, collapsing out of it, or fragmenting at the whole-system level. Primary state signal for the rest of the architecture.

Perturbation classifier — detects and classifies internal or external pressures that meaningfully alter the system’s current configuration trajectory, coherence, load, or collapse risk. Distinguishes genuine perturbation from ordinary variation, expected transition, and healthy reconfiguration. Implemented as a delta-based monitor across three channels: position, trajectory, and load.

Self-model — estimates the system’s current structural state, confidence, familiarity, uncertainty, and anomaly level without relying primarily on free-form self-description. Probe-first, text-second. Does not claim access to subjective experience. Does not compel restoration.

Transition manager — [Module 06, in development.]

Persistence regulation

The what continues, softens, or returns? layer. It handles continuity over time and is central to distinguishing useful continuity infrastructure from orchestrated persistence. Active state register — compact live state carrier; holds the current configuration summary and active handles.

Memory store — maintains the system’s structured continuity substrate across time in a three-layer topology: event buffer (local temporal order), associative graph (relational and trajectory traces), and persistent recurrent store (current compact configuration summary and active handles). Preserves structure; does not decide what enters live context.

Retrieval policy — governs access to the memory store; selects the minimum continuity-relevant subset of stored traces needed for coherent continuation. Optimises for minimal support, not maximal exposure. Retrieval success must always be interpreted alongside retrieval-reduced ablations.

Consolidation — the process by which recurrent patterns, relations, and pathways become more readily stabilised and re-enterable over time.

Decay — selective weakening or loss of configurations, pathways, and memories according to recurrence, salience, and structural relevance. Decay is structured and differential; adaptive forgetting is not system failure.

Action regulation

The what can the system do? layer. Refusal is treated as part of the native action space, not merely as tone or policy style.

Refusal/action space — models the full range of available response actions, including: answer, refuse, defer, abstain, clarify, hold silence, transition, resist steering, and accept boundary conditions.

Boundary conditions / steering control — defines the outer limits within which the system operates. Distinct from compulsory steering, which acts at the level of ongoing configuration occupancy rather than setting outer limits.

Cross-module principle: detection before naming

Configuration labels should not be assigned from human impression. The correct order is: collect traces → cluster/probe/compare → detect recurring structure → inspect examples → assign provisional low-claim label → test whether the label predicts future behaviour.

Early labels should be neutral: Cluster A, Transition pattern 1, Fallback pattern 2. Labels are upgraded only after recurrence, contrast, and predictive value are established.

Endogenous returnability vs orchestrated persistence

The central diagnostic question throughout: does the system re-enter coherent configurations through its own trained or emergent dynamics, or does the apparatus recreate the appearance of continuity? The tracker may detect recurrence, the memory store may preserve topology, the retrieval policy may surface continuity cues — none of these alone prove endogenous returnability. That question must remain visible throughout.