Path Authority And Interrupts

Claim Type: mechanism_hypothesis
Scope: Path authority, interruptibility, and commitment pressure
Depends On: ARC-005, ARC-003, ARC-008
Status: provisional
Claim ID: MECH-005


Source: docs/processed/legacy_tree/architecture/path_authority_and_interrupts.md

Norepinephrine: Path Authority, Interruptibility, and Commitment Pressure

(Proposed file: path_authority_and_interrupts.md)

Overview

In REE, all temporal depths (τ bands) operate continuously and coherently. There is no switching on or off of τ bands.

Instead, a norepinephrine-like control signal (ν) modulates: • which predictions and paths are allowed to influence commitment, • how interruptible the current trajectory is, and • how strongly post-commit errors drive learning and reorganisation.

ν is therefore a path-authority and commitment-pressure signal, not a clock selector and not a learning signal.

Core Claim (Normative)

Norepinephrine does not select temporal depth. It modulates the authority, interruptibility, and commitment pressure of paths derived from nested τ-scale predictions.

All τ bands run; ν controls what must now matter.

The Imperative to Commit

REE assumes an irreversible temporal arrow: • Time advances. • The system must act or continue acting. • Inaction is itself a commitment to trajectory persistence.

Therefore: • Commitment is unavoidable. • The question is not whether to commit, but how tightly the system is bound to its current path, and how costly deviation is allowed to be.

ν exists to regulate this pressure.

Functional Definition

Let: • P_{current} = currently committed trajectory • {P_i} = candidate future extensions (θ-scale paths) • \nu = norepinephrine-like signal • \phi = control mode • \tau bands = nested predictors producing constraints and errors

ν modulates path authority, not prediction generation.

What ν Does (Precisely)

  1. Path Authority Weighting

ν controls how strongly the system privileges: • continuation of the current path, • recently validated past paths, • newly generated alternative paths.

High ν: • recent path endings dominate • deviation is costly • commitment becomes “sticky”

Low ν: • alternative paths gain voice • branching is tolerated • commitment is more exploratory

Formally (conceptual):

Authority(P_i) \propto A(P_i) \cdot g(\nu, context)

Where A(P_i) is coherence-based viability, not reward.

  1. Interruptibility and Reorientation

ν controls whether local prediction errors (γ/β) are allowed to: • remain local updates, or • trigger θ-scale path reevaluation.

High ν: • small mismatches can force reorientation • rapid abandonment of failing paths

Low ν: • local errors are absorbed • the system “stays the course”

This is not learning; it is allocation of attention and urgency.

  1. Commitment Pressure and Post-Commit Error Salience

Once a commitment is made: • prediction errors after commitment carry special status • these errors are uniquely informative about: • model adequacy • action viability • environmental volatility

ν modulates how much post-commit error propagates upward: • High ν → post-commit errors strongly drive θ-level restructuring • Low ν → post-commit errors are damped, treated as noise

This captures a crucial biological asymmetry:

Errors after action matter more than errors before action.

What ν Explicitly Does Not Do

Architectural prohibitions: • ν does not update precision registers (πτ) • ν does not encode prediction error • ν does not select τ bands • ν does not determine value or reward

Any architecture where stress or surprise directly rewrites confidence is already collapsed.

Interaction with Phase (φ)

ν is permitted to influence φ transitions.

Examples: • sustained high ν → φ forced into TASK / ALERT • sustained low ν → φ permits DMN-like simulation • extreme ν → φ may force abort / reset

φ changes alter which τ×ρ combinations are eligible for commitment, but τ bands themselves continue running.

Relationship to Nested τ Bands (Clarified)

Key invariant:

γ, β, θ, and δ predictions coexist continuously.

ν does not: • enable or disable these predictors.

ν does: • weight their influence on: • path extension, • commitment persistence, • escalation of error.

γ always updates perception. β always updates affordances. θ always generates candidate paths. δ always constrains identity and narrative.

ν decides how hard the system must care, right now.

Failure Modes (Interpretive) • Hypervigilance: ν chronically high → constant reorientation, no stability • Perseveration: ν chronically low → failure to abandon incoherent paths • Burnout: ν flattened → weak post-commit learning • Chaotic switching: ν incorrectly coupled to precision

Design Rationale

Fragmenting: • precision (dopamine-like, τ-scoped) • path authority (norepinephrine-like, urgency-scoped)

allows REE to: • act decisively without freezing, • learn from consequences without panic, • imagine freely without compulsive enactment.

This is a control architecture, not an optimisation trick.

Summary • τ bands are nested and always active • Commitment is inevitable as time advances • ν modulates urgency, interruptibility, and path authority • Post-commit errors are privileged learning signals • ν never directly rewrites confidence or belief —

Open Questions

None noted in preserved sources.

  • MECH-005

References / Source Fragments

  • docs/processed/legacy_tree/architecture/path_authority_and_interrupts.md

REE is developed by Daniel Golden (Latent Fields). Apache 2.0.