Plant Model

The gap

The architecture requires a dynamical model of how psychophysiological state responds to inputs -- the "plant model" in control theory. This model would specify: state variables, measurement model, dynamical equations, and input-output relationships. The catecholamine-PFC literature provides mechanism but not a state-space model suitable for real-time estimation.

Partial progress

Two component models provide partial coverage:

1. MODAL model (Inglis et al. 2022): Spiking-level computational implementation of DA gain control dynamics (vSub -> NAcc -> VP -> VTA pathway). Validated against neural data. State variables: NAcc membrane potentials and bifurcation thresholds, VP firing rates, VTA active population size. Input-output: modulating variable -> DA gain on RPE.
2. Controllability-hopelessness model (Karvelis & Diaconescu 2024): Active inference model of how controllability and hopelessness interact through LC-NE, amygdala, and vmPFC-DRN-amygdala circuits. Computational targets: learning rate and belief decay.

Together these bound the problem: the plant model must integrate at minimum (a) DA population gain dynamics, (b) NE arousal/gain dynamics (LC phasic/tonic modes), (c) serotonergic controllability gating (vmPFC-DRN), and (d) their interactions, including LC-to-DRN projections coupling NE and 5-HT systems.

Additional complexity

The human's own state estimation is itself state-dependent. The anterior insula transforms bodily signals into conscious awareness (interoceptive processing), and interoceptive accuracy degrades under the same stress conditions described in the biological mechanisms (Wang et al. 2019). The plant model must account for the fact that the human's self-model is itself a state variable that degrades under the conditions the architecture detects.

Note: maladaptive DMN hyperconnectivity (rumination) represents a degradation pattern distinct from PFC shutdown -- partially functional PFC trapped in self-referential loops rather than offline PFC. May eventually require distinct actuator logic.

What's needed

Collaboration with computational neuroscientists who work on state-space models of neuromodulatory dynamics. The gap has narrowed from "no models exist" to "component models exist but have not been integrated or connected to peripheral measurement."

Kill condition coupling

If the plant model proves impossible with consumer sensors, kill condition KC1 is triggered.