AgentSeeResearch Notebook
version 1.0.0 · created 2026-04-08 · updated 2026-04-08

Coherence-Agency Bootstrap

modelhypothesisintegrated
ClaimCoherence and agency participate in a self-reinforcing cycle with identifiable neural substrates. Minimal coherence enables initial agency; successful agency reinforces integration; deeper integration enables broader agency. The architecture maintains biological preconditions for this cycle.
This claim fails if
If repeated successful exercise of agency during green states does not produce measurable improvement in baseline coherence indicators (Layer C capacity probes) or time-to-regulation metrics.

The hypothesis

The relationship between coherence and agency is reciprocal rather than unidirectional:

  1. Minimal coherence enables initial acts of agency
  2. Successful agency produces aligned action that reinforces integration across domains
  3. Deeper integration enables more sophisticated agency across broader contexts

This dynamic operates only when biological state permits both coherence construction and agency execution. When dysregulated, the bootstrap stalls, making time-to-regulation the rate-limiting factor -- not because recovery is the goal but because recovery is the precondition for the generative cycle to resume.

Three mechanism-level feedback loops

Each describes a process by which successful action strengthens neural conditions for future successful action:

(a) Controllability circuit proactive resilience

Prior experience with controllable stressors activates vmPFC patterns that provide lasting protection against later uncontrollable stress (Amat et al. 2006, cited in Maier & Seligman 2016). Successful agency builds a neural substrate that enables future agency. ESTABLISHED in animal models.

(b) DA gain amplification

The MODAL model (Inglis et al. 2022) shows that high state-feedback contingency (action reliably produces outcomes) increases the active DA population, which amplifies phasic learning signals from subsequent actions. Effective action in controllable environments produces stronger DA signals that improve subsequent learning. SUPPORTED (computational model validated against neural data).

(c) LC-NE exploitation stabilization

Aston-Jones & Cohen (2005) describe an explicit self-reinforcing loop: as reward sources are discovered, DA-dependent reinforcement learning strengthens reward-producing behaviors, which increases current utility assessment, which drives LC into phasic mode, which further stabilizes exploitation of those behaviors. This is the source authors' own description, not this notebook's interpretation. ESTABLISHED as proposed mechanism.

The integration claim

Each loop independently describes how successful action strengthens conditions for future successful action. The claim that these loops constitute a unified upward dynamic -- the coherence-agency bootstrap -- is this work's integration. The bootstrap is not the same claim as "coherence is good" or "agency is good." It is a specific mechanistic prediction: that coherence and agency participate in a self-reinforcing cycle with identifiable neural substrates, and that the architecture's role is to maintain the biological preconditions under which that cycle operates.

Connection to the objective function

The three-target optimization equation penalizes degraded states but has no explicit term for growth enablement. If the bootstrap operates as hypothesized, these objectives converge: maintaining biological preconditions for coherence IS maintaining conditions for the generative cycle, because the bootstrap requires the same PFC-online, controllability-circuit-functional state that L_state targets.

Evidence status

HYPOTHESIS / INTEGRATED. Individual loops graded ESTABLISHED or SUPPORTED. The unified dynamic is the integration. Observation period required to test is an experimental design parameter, not a theoretical claim.