SD-037: Broadcast Override Regulator (orexin-analog)

Claim ID: SD-037 (parent), MECH-280 / MECH-281 / MECH-282 (children) Subject: broadcast.override_regulator Status: IMPLEMENTED 2026-04-25 (substrate-readiness PASS V3-EXQ-483b 2026-05-08; Tier-1 evidence V3-EXQ-483c queued) Registered: 2026-04-22 Origin: Lit-pull synthesis (homeostatic override under sustained threat), 2026-04-22, motivated by V3-EXQ-471 catatonic lock-in. Source: targeted_review_homeostatic_override/SYNTHESIS.md. Depends on: SD-036, SD-012, SD-032a, MECH-279

Problem

SD-036 (GABAergic cross-stream decay) addresses one half of the V3-EXQ-471 catatonic lock: in the absence of input, harm streams now relax toward baseline. But the lit-pull on homeostatic override is convergent that decay alone is not how biology breaks harm-stream lock-in under sustained threat. The biology uses a dedicated override authority — a small, broadly-projecting gain modulator (canonically orexin/hypocretin) that integrates internal homeostatic state and threat context and reweights downstream arbitration without silencing the upstream signals.

The convergent picture across the lit-pull (SYNTHESIS.md §1, §4, §5):

z_harm continues to be computed and broadcast even when override fires (Burnett 2016, de Araujo Salgado 2023). The mouse still perceives the predator; it just acts on hunger.
The override is implemented downstream as reweighting at commit gates, not upstream as silencing of harm computation (Bjorness/Greene 2024; Marino 2020).
The override is recruited specifically under high effort, sustained threat, or metabolic demand — it is not a continuous baseline modulator (James/Aston-Jones 2017).
The bridge between drive_level and z_goal seeding (SD-012) is gated by this override authority. Without it, drive accumulates but does not gain authority to seed goal-mode behaviour under threat (Marino 2020 peri-LC; cataplexy in narcolepsy as the decoupling failure mode, Bassetti 2019).

The architectural commitment SD-037 makes is that the V3 regulatory stack needs a third layer, parallel to and dissociable from:

5-HT layer (MECH-186/187/188) — goal-pipeline gain, slow tonic modulation.
GABA layer (SD-036) — cross-stream decay; tonic + phasic relaxation when input drops below threshold.
Orexin-analog layer (SD-037, this doc) — broadcast override; reweights z_harm at downstream commit gates and gates SD-012 drive→z_goal seeding under metabolic-demand + threat conditions.

These three layers are mutually dissociable in failure: 5-HT depletion ≠ GABA decay loss ≠ orexin loss, and clinical phenomenology bears this out (depression vs benzodiazepine withdrawal vs narcolepsy/cataplexy).

Mechanism

SD-037 introduces a scalar (or low-dimensional vector) override_signal computed from drive_level (SD-012), sustained-threat context (rolling window of z_harm magnitude above threshold), and a recruitment threshold. When override_signal exceeds its recruitment threshold:

Reweight z_harm at commit gates — the weight of z_harm in commit-gate threshold computations (SD-032a SalienceCoordinator arbitration, MECH-279 PAG freeze-gate threshold) is scaled down. z_harm itself is not modified — the decay layer (SD-036) remains responsible for that — but its arbitration influence drops.
Gate drive→z_goal seeding (the SD-012 bridge) — when override_signal is below recruitment threshold, drive_level cannot seed z_goal even if benefit_exposure is non-zero. Above threshold, drive_level seeds z_goal directly. This is the architectural answer to the homeostatic-override question: drive does not have to compete with threat in arbitration; under override conditions it acquires authority.
Threshold-like recruitment, not continuous — the regulator is not a continuous baseline modulator. It accumulates evidence (drive accumulating, threat sustained) until a recruitment threshold crosses, then fires. This matches James/Aston-Jones 2017 and the sigmoid-like behavioural switch in de Araujo Salgado 2023.

Failure-mode predictions

The lit-pull surfaced two clean failure-mode signatures the regulator must reproduce:

Saturated override (PWS-analog): override_signal pinned high → compulsive drive pursuit despite negative outcomes, hyperphagic-equivalent (Brown 2022).
Lost override (narcolepsy/cataplexy-analog): override_signal absent → z_harm and drive_level both compute normally but fail to integrate into coordinated behaviour; emotional/drive input fails to recruit motor coupling (Bassetti 2019).

These failure modes are diagnostic for distinguishing SD-037 dysfunction from SD-036 (GABA decay) dysfunction or from MECH-186/187/188 (5-HT) dysfunction — they look clinically different.

Child mechanisms

MECH-280: LH-PAG override projection

Statement: A direct projection from LH (orexin and/or peri-LC GABA analog) to PAG modulates the GABAergic freeze-gate (MECH-279) under metabolic demand. The projection does not silence the freeze gate; it raises its threshold (theta_freeze scales with override_signal). This produces the behavioural transition from freeze to active foraging under sustained hunger + threat.

Substrate prediction: MECH-279 theta_freeze and exit_threshold parameters become functions of override_signal: theta_freeze_effective = theta_freeze_base * (1 + alpha_override * override_signal).

Evidence anchors: Marino 2020 (LH GABA → peri-LC for foraging engagement); de Araujo Salgado 2023 (sustained hunger overrides predator avoidance); Wang/Lin 2015 (PAG as downstream arbiter of broadcast threat signals).

MECH-281: Orexin gain modulation of drive-coupled arousal

Statement: An orexin-analog modulator gates the coupling between drive_level and arousal-relevant downstream targets (PFC for SD-033a deliberation, BLA/CeA for SD-035 amygdala arbitration, beta-gate for MECH-090). Loss of this coupling produces a cataplexy-analog failure: emotional/drive input fails to recruit motor activation.

Substrate prediction: override_signal scales the gain on drive→action-pipeline projections (drive→z_goal weight, drive→beta_gate sensitivity). Lesion test: setting override_signal to constant zero should produce decoupling between drive accumulation and any motor/action consequence — drive computes, threat computes, but they fail to integrate at action selection.

Evidence anchors: Bjorness/Greene 2024 (orexin as coupling signal); James/Aston-Jones 2017 (orexin recruitment under high effort/demand); Bassetti 2019 (narcolepsy as coupling-loss clinical model).

MECH-286: Override-gated sleep-state transition (wake-stability axis)

Statement: SD-037 override_signal also gates entry to and exit from offline (sleep) mode. High override (sustained drive + threat under metabolic demand) suppresses sleep-mode onset; low override + accumulated MECH-284 V_s-residual staleness + permissive threat context permits onset and recruits MECH-285. This is the recruitment authority for the offline arm of the V_s bidirectional cluster — without MECH-286, MECH-285 produces a priority list with no transition mechanism to consume it. This is the V3 architectural account of orexin’s canonical role as wake-stability gate (Saper/Scammell flip-flop), which MECH-281 understates by framing cataplexy purely as a motor-coupling failure.

Substrate prediction: sleep-mode entry requires three joint conditions: override_signal < theta_sleep_permit AND MECH-284_staleness > theta_sleep_recruit AND threat_context permissive. Lesion test: pinning override_signal high under conditions that would normally permit sleep should produce sleep-onset failure (hyperarousal-insomnia phenotype). Saturation test: setting override_signal near zero should produce uncontrolled sleep intrusion (narcolepsy phenotype, distinct from MECH-281 cataplexy but produced by the same lesion — predicting clinical co-occurrence).

Substrate implementation (MECH-286): IMPLEMENTED 2026-05-21. ree_core/sleep/sleep_onset_gate.py (evaluate_sleep_onset_permit) wired into SleepLoopManager._run_cycle before run_sleep_cycle. Config: REEConfig.use_mech286_sleep_onset_gate (default False), mech286_theta_sleep_permit (0.5), mech286_theta_sleep_recruit (0.3), mech286_threat_tonic_threshold (0.4). Diagnostics on blocked or completed cycles: mech286_sleep_permitted, mech286_override_signal, mech286_staleness_max, mech286_z_harm_a_norm. Validation: V3-EXQ-599 queued.

Tertiary prediction (schema-repair starvation): chronic hyperarousal sustained over many cycles should produce stable accumulation of MECH-284 staleness with no clearance, observable as deteriorating schema fit on novel inputs even with intact waking computation. This is the architectural mechanism behind sleep-deprivation cognitive decline and a contributor to PTSD chronicity (paired with MECH-285 priority loss + MECH-094 hypothesis-tag dysfunction as the rumination architecture). See psychiatric_failure_modes.md “Hyperarousal Insomnia and Schema-Repair Starvation”.

Evidence anchors: Saper/Scammell flip-flop literature (orexin as wake-stability gate against VLPO sleep-active circuits); orexin neuron silence during sleep, firing during wake; Bassetti 2019 (sleep-state instability as core narcolepsy deficit, cataplexy as separable downstream symptom); chronic insomnia in PTSD and severe melancholic depression; sleep deprivation as established psychosis precipitant; orexin-receptor antagonist clinical signal in chronic insomnia (suvorexant, lemborexant).

MECH-282: LPB interoceptive routing into harm-arbitration

Status: IMPLEMENTED 2026-05-21 (substrate; validation V3-EXQ-600 queued).

Statement: A lateral-parabrachial-nucleus analog (LPB) routes interoceptive distress signals (visceral malaise, taste-aversion analogs, hypoxia, hypoglycemia) into the harm-arbitration network via PBN-CGRP → CeA → PAG. This is the upstream broadcast arm of the threat-side architecture, parallel to VMHdm’s predator-side broadcast. Both feed downstream arbiters where SD-037’s override_signal reweights them.

Substrate prediction: A separate input channel into the z_harm computation that carries interoceptive/metabolic-distress signal, dissociable from external-threat z_harm. Under SD-037 override conditions, the interoceptive harm-channel is not suppressed (hunger-as-pain remains felt), but its weight in arbitration becomes the input that drives the override_signal recruitment threshold from below.

Evidence anchors: Palmiter 2018 (PBN CGRP general alarm review); Pyeon 2025 (PBN CGRP active defense modulation); Strigo/Craig 2016 (insular common-currency framing).

Architecture context

SD-037 sits at the third regulator layer of the V3 control stack. Stream computation happens in SD-010/SD-011/SD-012 (sensory and drive); decay happens in SD-036 (tonic relaxation); arbitration happens in SD-032a (SalienceCoordinator) and at downstream commit gates (MECH-279 freeze, beta-gate). SD-037 is the regulator that intervenes between stream computation and arbitration — it does not modify the streams themselves but modifies the weights with which they are read at commit gates and gates the drive→goal bridge.

The relationship to SD-036 is critical and easy to confuse:

	SD-036 (GABA decay)	SD-037 (orexin override)
Acts on	The streams themselves (z_harm_s, z_harm_a, z_beta)	The arbitration weights and the SD-012 bridge
Trigger	Absence of input below threshold	Drive accumulation + sustained threat above recruitment threshold
Effect on z_harm	Decays its magnitude	Leaves its magnitude untouched
Effect on commit gate	Indirect (smaller z_harm crosses threshold less)	Direct (changes the threshold itself)
Failure mode	Pinned streams; catatonic lock under input absence	Decoupled drive and arousal; no override under sustained threat with metabolic demand
Clinical analog	Catatonia subtype II / benzo deficiency	Narcolepsy/cataplexy + PWS hyperphagia
Recovery from V3-EXQ-471 lock	Removes the pinned z_harm	Even with z_harm=0, would let drive seed z_goal under metabolic demand

Both are required. SD-036 unsticks the stream; SD-037 grants drive the authority to drive behaviour once unstuck. EXQ-471 lock-in needs both to be unconditionally broken. This is the headline finding of the lit-pull synthesis (§”Bottom line”).

What this SD enables

Resolution of V3-EXQ-471 at architectural depth — not just unsticking the lock via decay, but providing the override authority that lets metabolic demand drive z_goal seeding under threat.
Bridge between SD-012 and the goal pipeline — the missing piece that prevents drive_level alone from seeding z_goal in V3 (the EXQ-085 cluster’s failure was partly this).
Diagnostic dissociation of three regulator-layer failures — SD-036, SD-037, and MECH-186/187/188 each predict distinct clinical/behavioural failure phenotypes.
Foundation for compulsive-pursuit and decoupling failure modes — PWS-analog hyperphagia and narcolepsy-analog decoupling become testable in V3.

Validation experiment (deferred)

A full SD-037 validation experiment is not yet queued — implementation is gated by EXQ-475 results (whether SD-036 decay alone produces partial recovery or no recovery on the EXQ-471 lock). Once EXQ-475 results land, the SD-037 validation should be a factorial:

4 arms: {SD-036 OFF, SD-037 OFF}, {SD-036 ON, SD-037 OFF}, {SD-036 OFF, SD-037 ON}, {SD-036 ON, SD-037 ON}.
Same conditions as EXQ-471.
Predicted: arms 2 and 3 show partial recovery (each addresses one half); arm 4 shows full recovery; arm 1 reproduces the EXQ-471 lock.

This dissociation experiment is the empirical test of the SYNTHESIS.md §”Bottom line” prediction.

SD-036 — paired regulator (GABA decay); the stream-side complement to SD-037’s arbitration-side intervention.
SD-012 — homeostatic.drive_modulated_goal_seeding; SD-037 is the gating authority for the SD-012 bridge.
SD-032a — SalienceCoordinator; consumes override_signal as a weight modulator.
MECH-279 — PAG freeze-gate; consumes override_signal as a theta_freeze modulator (the MECH-280 projection).
MECH-186/187/188 — 5-HT regulatory layer; first regulator layer in the stack. SD-036 is the second; SD-037 is the third.
MECH-269 — hippocampal verisimilitude anchor selection; the override_signal recruitment threshold can plausibly be informed by V_s (anchored, high-V_s harm = known threat → over-ridable; novel low-V_s harm = take seriously, do not override). Cross-claim coupling deferred until EXQ-475 + SD-037 implementation.