Homeostatic Override / Imminent-Death Drive Amplification Literature Pull
Paste this into a new Claude Code session as your opening message. Created: 2026-04-22 Origin: V3-EXQ-471 fishtank visualization showed a catatonic-locked agent whose energy depleted to zero with no homeostatic override. SD-036 (GABAergic decay) resolves the harm-stream lock but does not address the missing override authority that should let hunger override threat under survival demand.
Prompt
/lit-pull Homeostatic-need brain circuits and how they amplify or override other drive systems under imminent-death conditions – specifically the question of how hunger / energy depletion / dehydration acquires authority to override threat-driven avoidance behaviour. The architectural question: in V3 we have separate streams for threat (z_harm), goal (z_goal), drive (energy/homeostatic), and an arbitration mechanism (mode selection via SD-032a SalienceCoordinator). When a hazard event locks the agent into avoid mode and energy then runs to zero, no mechanism currently reverses the lock – avoid still wins arbitration even when starvation is imminent. Subjectively, hunger does not just compete with threat; it sharpens or amplifies multiple drives simultaneously (the dread stream may itself intensify, the fast-interrupt becomes more sensitive, goal pursuit becomes more compulsive). This suggests a broadcast neuromodulator that gain-modulates several streams from one homeostatic source, rather than a competing scalar.
Target claims to inform / extend:
- SD-012 (drive-modulated goal seeding – currently has no override authority over z_harm-driven avoid mode)
- SD-036 (GABAergic cross-stream decay regulator, registered 2026-04-22 – the decay layer is one half of the catatonia rescue; this lit-pull addresses the other half: what reverses the mode lock when survival demands it)
- MECH-279 (PAG GABAergic freeze-gate – freeze exit may be normal-route via decay or override-route via homeostatic emergency)
- SD-032a (SalienceCoordinator – target for whatever mechanism provides override authority)
- Possible new SD/MECH for the override mechanism itself
Specific neurobiological systems to cover:
- Lateral hypothalamus (LH) – the canonical hunger/feeding circuit. Specifically:
- LH GABAergic and glutamatergic projections to VTA, PAG, and PFC
- How LH activity overrides defensive behaviour in foraging-under-threat paradigms
- Burnett et al 2019 (LH circuits and feeding); Sternson lab work on AgRP/POMC
- Orexin / hypocretin (LHA) – broadcast arousal/motivation neuromodulator strongly implicated in the kind of multi-stream gain modulation the architectural question requires:
- Orexin projections to VTA, LC, raphe, BLA, PFC, PAG
- Berthoud, Saper, Sakurai work on orexin as a homeostasis-arousal coupler
- How orexin loss (narcolepsy) produces specific failures in motivated behaviour under metabolic challenge – the human knockout
- Lateral parabrachial nucleus (LPB) and the brainstem sensory-aversive coupling:
- How interoceptive distress signals (hunger, thirst, hypoxia, hypoglycaemia) are routed through LPB to higher centres
- LPB CGRP neurons and threat-vs-need integration (Palmiter lab)
- Ventromedial hypothalamus (VMH) and defensive-vs-feeding switching:
- VMH SF1 neurons in defensive behaviour and how they interact with LH feeding circuits
- Lin et al, Anderson lab work on VMH switching
- Insula and interoceptive amplification under metabolic stress:
- How insular processing of interoceptive signals scales with deprivation level
- Craig 2009 + extensions; whether this maps onto SD-032c (AIC-analog) gain modulation under drive
- The clinical and ethological literature on starvation overriding fear:
- Foraging-under-predation models (Lima & Dill 1990 and successors)
- Clinical: anorexia nervosa as a failure of this override (drive does not acquire override authority – explanatory question for the architecture)
- Clinical: hyperphagic conditions (Prader-Willi, hypothalamic damage) as pathological excess of override
- The phenomenology of severe hypoglycaemia / hypoxia: which streams intensify vs which suppress
Architectural questions the lit-pull should help answer
-
Single broadcast vs multi-target competition. Does a single homeostatic neuromodulator (orexin is the leading candidate) gain-modulate multiple streams simultaneously, or is the override mediated by point-to-point projections that each carry their own copy of the homeostatic signal? The former predicts a regulator-layer architecture (parallel to SD-036’s GABA layer); the latter predicts per-target wiring without a regulator.
-
Threshold vs continuous. Does override authority appear gradually as drive accumulates, or does it cross a threshold and snap on? The latter would be architecturally cleaner and matches the clinical phenomenology of “I just couldn’t keep refusing food / stop drinking water.” The former matches the neuroeconomics literature on graded value.
-
Symmetry across drives. Is the override mechanism the same for hunger, thirst, hypoxia, sleep deprivation, etc., or does each homeostatic system have its own override pathway? Architectural parsimony favours shared; neurobiological evidence may favour partly distinct (e.g. CO2-sensing for hypoxia is anatomically separable from glucose-sensing for hunger).
-
Does the override AMPLIFY z_harm or REPLACE its arbitration weight? The user observation suggests amplification – hunger sharpens dread, not replaces it. This would predict that under starvation the agent becomes more threat-sensitive, not less, but also more willing to act despite threat (because both signals are amplified and the action threshold is the difference). This is a substantively different architecture from “drive overrides threat” and the lit-pull should be alert to which model the biology supports.
-
What is the bridge to z_goal seeding (SD-012)? Currently SD-012 requires benefit_exposure > 0 to seed z_goal. Under override, does drive directly seed z_goal (skipping benefit_exposure), or does override force the agent into exploratory action that creates benefit_exposure, which then seeds z_goal via the normal pathway?
Output structure
Standard targeted_review_*/ format. Suggested directory: evidence/literature/targeted_review_homeostatic_override/
Per-paper records as usual. After the pull, write a short synthesis note flagging:
- Which architectural question(s) each paper addresses
- Which of the five questions above remain underdetermined and need additional pulls
- Whether the evidence supports registering a new SD/MECH cluster for the override mechanism (and if so, draft proposed claim text)
Estimated scope: ~10-15 papers, single session.
Notes for the agent doing the pull
- The user is a consultant psychiatrist who deals clinically with both ends of the override-failure spectrum (anorexia nervosa = override-deficient; hyperphagia = override-excessive). Use precise clinical terminology.
- The architectural reading explicitly expects the override to be a broadcast neuromodulator – be alert to evidence that contradicts this expectation, not just evidence that supports it.
- Connect to existing REE memory entries on homeostasis and drive: SD-012 (drive seeding), MECH-186/187/188 (serotonergic regulatory layer for the goal pipeline), SD-036 (GABAergic decay regulator). The hypothesised override layer is candidate third regulatory layer alongside 5-HT and GABA, plausibly orexin-mediated.
- The exemplar that motivated this pull is V3-EXQ-471 (fishtank visualization, 2026-04-21 to 2026-04-22). Worth re-reading the trace summary in
psychiatric_failure_modes.md“Catatonia, Subtype II” before starting.