State-Conditioned Gating
Principle
Actuator selection must be conditioned on the human's estimated neurobiological regime. The same input (observation, question, suggestion, silence) produces different outcomes depending on whether the human's PFC is online or offline, whether the controllability circuit is functional or suppressed, and whether arousal is on the ascending or descending limb of the inverted-U.
Derivation
The catecholamine-PFC dynamics (Arnsten 2009, 2015) establish that PFC function follows an inverted-U curve with catecholamine levels. On the ascending limb, moderate arousal enhances executive function. Past the apex, PFC shifts from reflective to reflexive mode. The same information presented to a person on the ascending limb (enhances processing) versus past the apex (adds uncontrollable cognitive load) produces opposite effects.
The controllability circuit (Maier & Seligman 2016) adds a gating dimension: the vmPFC-DRN pathway detects instrumental contingency. Interventions that provide detectable action-consequence contingency activate this pathway. Interventions that add demands without contingency do not.
Clinical precedent exists. In motivational interviewing, balanced ambivalence exploration during unresolved ambivalence produces worse outcomes than selective evocation (Miller & Rose 2015; Magill et al. 2018, meta-analysis of 36 studies, N=3,025). The same technique (reflection) helps or harms depending on the client's state.
Engineering implication
The architecture requires a regime estimation function and a gating function that maps estimated regime to allowed/prohibited actuator classes. The illustrative regime table (notebook Section 1.10) proposes five regimes: G0 (green-stable), G1 (green-fragile), R0 (red), R1 (red-severe), U (uncertain). Each regime has a whitelist and a blacklist. The U regime defaults to null or minimal action.
This principle is the bridge from the neuroscience (catecholamine dynamics are state-dependent) to the design requirement (actuator selection must be state-conditioned). It underwrites prediction P1 (backfire regime exists).
Discrete vs. continuous
The regime table is one design choice. An alternative is continuous confidence-weighted state estimation with graduated actuator intensity. The discrete approximation is chosen for engineering tractability and testability. Experiment E1 tests regime classification; if it succeeds, a later iteration could move to continuous estimation. The graduated structure of the regime table (four regimes plus uncertainty) is an approximation of a continuous space, not a claim that underlying states are discrete.