Computational Psychiatry through Second-Order Cybernetics

A clinical application library for the Sounio programming language that formalizes psychiatric phenomena as computable properties of cybernetic systems.

Current computational psychiatry (active inference, RL models, Bayesian brain) is first-order: the model observes the patient but never models itself observing. The therapist is invisible. Treatment resistance is a hidden parameter. Metacognitive dysfunction is hand-waved.

This library is second-order: the therapist IS an observer whose drift IS part of the clinical system. Treatment resistance isn't mysterious — it's drift / trust, computable from session dynamics.

sounio-psychiatry/
├── src/
│   ├── core/           # Cognitive process graph, closure detection
│   │   ├── clinical_state.sio    # ClinicalState struct + constructors
│   │   ├── closure.sio           # DFS cycle detection on cognitive graph
│   │   └── metacognition.sio     # Observer-based self-monitoring model
│   ├── models/         # Specific disorder formalizations
│   │   ├── anxiety.sio           # Ashby variety deficit
│   │   ├── depression.sio        # Rumination as budget exhaustion
│   │   ├── dissociation.sio      # Process deactivation
│   │   ├── psychosis.sio         # Closure loss
│   │   ├── double_bind.sio       # Bateson 1956
│   │   ├── schismogenesis.sio    # Bateson 1936
│   │   ├── alliance.sio          # Therapeutic alliance (Pask)
│   │   ├── resistance.sio        # Treatment resistance
│   │   ├── homeostasis.sio       # Symptom-as-regulator
│   │   └── crisis.sio            # Sedenion encoding + ZDP
│   ├── bridges/        # Connections to sounio-cybernetics
│   │   ├── observer_bridge.sio   # Observer → ClinicalState
│   │   ├── eigenform_bridge.sio  # Self-model as eigenform
│   │   └── variety_bridge.sio    # Ashby's Law integration
│   └── validation/     # Clinical outcome measurement
│       ├── rci.sio               # Reliable Change Index
│       ├── assessment.sio        # Epistemic assessment fusion
│       └── trajectory.sio        # Treatment trajectory tracking
├── tests/
│   ├── unit/           # Per-function invariant tests
│   ├── integration/    # Cross-module scenario tests
│   └── clinical/       # Tests against published meta-analyses
├── examples/
│   ├── healthy_baseline.sio
│   ├── anxiety_treatment.sio
│   ├── therapy_session.sio
│   ├── double_bind_family.sio
│   └── full_trajectory.sio
├── data/               # Clinical dataset interfaces (NOT raw data)
│   ├── abide/          # ABIDE-I phenotypic data loader
│   ├── adhd200/        # ADHD-200 loader
│   └── ukbiobank/      # UK Biobank mental health loader
├── paper/              # Computational Psychiatry journal submission
├── docs/               # Clinical user guide
└── scripts/            # Dataset download + preprocessing

• sounio >= 1.0.0-beta.4 (the language)
• sounio-cybernetics (9 foundational modules: distinction, eigenform, observer, autopoiesis, coupling, conversation, variety, bateson, languaging)

# Clone
git clone https://github.com/sounio-lang/sounio-psychiatry
cd sounio-psychiatry

# Set stdlib path to include cybernetics
export SOUNIO_STDLIB_PATH=/path/to/sounio/stdlib

# Run the healthy baseline example
souc run examples/healthy_baseline.sio

# Run the full treatment trajectory
souc run examples/full_trajectory.sio

# Run clinical validation tests
souc run tests/clinical/meta_analysis_h1.sio

The model makes structural predictions testable against published evidence:

• ABIDE-I: Map phenotypic data → cognitive process graph → test closure/anxiety predictions
• ADHD-200: Test executive function disruption → closure loss prediction

• Partner with clinical site for therapy session data
• Code sessions as (therapist_observation, patient_response) pairs
• Compare model alliance trajectory with Working Alliance Inventory

• Calibrate quantitative outputs against clinical scales (GAD-7, PHQ-9, WAI)
• Establish normative ranges for anxiety_deficit, treatment_resistance, metacognitive_capacity
• Pre-register replication study

This library formalizes theories, not diagnoses. It is a research tool, not a clinical decision support system.

• No individual clinical decisions should be based on model outputs
• All clinical validation requires IRB approval
• Raw clinical data is never stored in this repository
• The data/ directory contains only loader scripts and phenotypic metadata schemas
• Model predictions are structural (qualitative pattern) not quantitative (specific scores)

If you use this library in research, please cite:

@software{sounio_psychiatry_2026,
  title = {sounio-psychiatry: Computational Psychiatry through Second-Order Cybernetics},
  url = {https://github.com/sounio-lang/sounio-psychiatry},
  year = {2026},
  note = {Requires sounio >= 1.0.0-beta.4 and sounio-cybernetics}
}

We welcome contributions from:

• Psychiatrists / clinical psychologists — clinical validation, hypothesis refinement
• Systems theorists — model extensions, new disorder formalizations
• Data scientists — dataset integration, statistical validation
• Sounio developers — core library improvements

See CONTRIBUTING.md for guidelines.

MIT. See LICENSE.