Depressive Network Regimes and Repair

Status: anchor doc for the clinical-depression intake claims (RA-001, MECH-367, Q-061). Registered: 2026-06-09. Off the V3 / GAP-7 critical path (V4/V5 + out-of-domain). Source intake: evidence/planning/thought_intake_2026-06-06_clinical_depression_network_connectivity_reversal.md Raw thought: docs/thoughts/2026-06-06_clinical_depression_network_connectivity_reversal.md

Terminology guardrail (load-bearing): “long-term depression” throughout this doc means long-term CLINICAL depressive illness, NOT synaptic long-term depression (LTD).

This doc holds the three claims reaped from the 2026-06-06 clinical-depression thought intake. It is an anchor/compass doc, not a substrate design: none of these is a V3 implementation target.

ra-001

Claim (research_anchor, out_of_domain): Long-term clinical depression involves a chronicity-dependent reversal of large-scale network coupling. Past a ~24-month chronicity threshold, the relationship between symptom severity and Central-Executive-Network (CEN) <-> Default-Mode-Network (DMN, precuneus) functional connectivity reverses sign:

  • Non-chronic: stronger CEN-DMN coupling at low severity, weaker at high severity.
  • Chronic: connectivity strengthens as symptoms worsen, co-occurring with entrenched rumination.

The depressive brain-state is duration-indexed, not a single static regime.

Status of the source. Press item: Neuroscience News, Long-Term Depression Reverses Brain Network Connectivity. Primary peer-reviewed paper TRACED + verified (2026-06-09): Zanao T., Salvan P., Razza L. B., et al., “Chronicity moderates the impact of severity on central executive-default mode network functional interactions in depression,” Scientific Reports (2026), DOI 10.1038/s41598-026-40364-2 (preprint medRxiv 10.64898/2026.01.28.26345027) – 46 patients, chronicity = episode > 24 months, HAM-D severity; CEN-precuneus(DMN) connectivity sign-moderated by chronicity. The 24-month threshold and precuneus locus are paper-verified.

Precision on “reverses”. This is a cross-sectional reversal of the connectivity-vs-severity relationship between non-chronic and chronic cohorts. It is not a demonstration of therapeutic reversibility (that is the open question Q-061).

Why out_of_domain. The test domain is a clinical fMRI cohort; no REE substrate at any level distinguishes named large-scale networks (CEN/DMN). Registered as a research_anchor (the RA- prefix; claim_type sanctioned by v4_spec.md “Out-of-domain claims … belong as research_anchor / literature_synthesis claim types”). It grounds the existing REE depression cluster (MECH-088, INV-034, MECH-124, Q-021, MECH-082/086) and motivates MECH-367 and Q-061. It contradicts nothing in the registry.

mech-367

Claim (mechanism_hypothesis, substrate_conditional, V4/V5): The depressive failure mode is carried by a multi-axis network-regime vector, not a scalar low-mood / low-reward value:

depressive_regime_risk = f(future_trajectory_access,
                           goal_stream_coupling,
                           residue_load / self_weighting,
                           rumination_loop_gain,
                           action_threshold / energy_budget,
                           social_affordance_access,
                           offline_repair_quality)

The depressive state is defined by the joint configuration of these coupling/access axes and is self-maintaining; a scalar mood/reward value is an insufficient representation.

What is new vs what REE already owns. REE already owns each axis individually:

Axis Existing REE owner
future_trajectory_access INV-034; MECH-082 / MECH-086 (locked-in avoidant brain)
goal_stream_coupling INV-034 (depressive attractor = goal-maintenance failure, EXQ-237a)
residue_load / self_weighting MECH-124 (residue-field harm dominance)
rumination_loop_gain MECH-124 (self-amplifying replay loop)
action_threshold / energy_budget Q-021 (pure-avoidance behavioural flatness / anhedonia)
social_affordance_access (scattered; nearest: social-coupling salience)
offline_repair_quality MECH-124 / MECH-123 (consolidation can deepen or repair)

What is new is the composition into one weighted vulnerability vector whose joint configuration is the regime. The empirical anchor RA-001 (duration-dependent coupling reversal) motivates treating the regime as a coupling-structure vector rather than a scalar.

Why substrate_conditional / V4. The axes exist piecemeal in V3 but the joint-vector read-out is not built, and there is no named large-scale-network coupling substrate. Off the V3 / GAP-7 critical path. Do not build in V3 until routed by experiment.

Failure-mode precision (do not blur). Keep the depression failure mode distinct from:

  • Psychosis = NA collapse of E1/E2 constraint + DA aberrant salience on noisy input (MECH-088 psychosis profile; tag-MISassignment MECH-094/MECH-115).
  • Confabulation = tag-LOSS / simulation-real source-monitoring failure (MECH-094).

“Rumination” here is the depressive recurrent-loop sense (MECH-124), not psychotic content-intrusion.

q-061

Question (open_question, answer_state): Is a maladaptive depressive network-regime (MECH-367) reversible – reconfigurable by REE’s existing repair machinery (offline integration / MECH-124 replay rebalancing, decommitment from ruminative loops, re-access to social affordances, goal-stream re-coupling) – or does chronicity lock the regime past a threshold (RA-001) such that the repair routes that work early fail or reverse-sign late?

The sharp sub-question is duration-dependence: do the offline-integration / decommitment / social-re-access routes that reconfigure an early depressive regime still work late, or does the MECH-124 self-amplifying consolidation loop plus the RA-001 coupling reversal make late repair require a different mechanism?

Status. A question, not an assertion. Resolvable in REE only once the MECH-367 vector substrate exists (V4/V5). Off the V3 critical path – do not queue a V3 experiment against it. Hold the anti-overclaim guard: do not assert that depression is “merely a reversible software state.”

Not registered here

  • Candidate D – the depression-analogue AI cognitive-failure mode (persistent low action-initiation despite available goals; negative-evaluation loops; future-trajectory suppression; excessive failure-generalisation) – belongs in the cross-repo Latent-Fields/ai-cognitive-failure-taxonomy, not in claims.yaml, and is handled with explicit anti-anthropomorphic guardrails (the analogue is a network-state failure in trajectory generation and action readiness, NOT mood/sadness). The REE EXT- series (LLM external failure modes) is the in-registry sibling pattern, should that taxonomy ever fold back.

REE is developed by Daniel Golden (Latent Fields). Apache 2.0.