"Depression has PHQ-9. Anxiety has GAD-7. Loneliness has nothing — until now."

Tether is the first behavioral instrument for the global loneliness epidemic. It detects social drift from behavioral patterns 6 weeks before you feel it, identifies the root cause, and deploys three AI agents to intercept the crisis before it happens.

1 billion people are clinically lonely globally. Loneliness raises mortality risk by 26% — equivalent to smoking 15 cigarettes daily. It increases dementia risk by 50% and heart disease by 29%. Japan appointed a Minister of Loneliness. The UK did the same. The US Surgeon General issued a formal epidemic advisory in 2023.

And yet — depression has PHQ-9. Anxiety has GAD-7. Loneliness has no clinical instrument. Governments are trying to solve an epidemic they cannot even measure.

Tether changes that.

# 1. Install dependencies
pip install -r requirements.txt

# 2. Generate training data
python data/generate_data.py

# 3. Train the ML models (~60 seconds)
python models/train_model.py

# 4. Launch
streamlit run app.py

Open http://localhost:8501

tether/
├── app.py                          ← Main Streamlit app
├── requirements.txt
├── README.md
├── data/
│   ├── generate_data.py            ← Synthetic dataset generator (52K records)
│   └── tether_behavioral_data.csv  ← Generated training data
└── models/
    ├── train_model.py              ← ML training pipeline
    ├── classifier.pkl              ← Loneliness type classifier
    ├── regressor.pkl               ← Social health score regressor
    ├── crisis_predictor.pkl        ← Crisis risk predictor
    └── feature_cols.pkl            ← Feature names

Tether asks 8 plain-language questions about how you actually behave — not how you feel. Self-reports are unreliable. Behavior doesn't lie.

After each answer, a peer-reviewed micro-insight appears explaining the clinical significance of that signal.

Model 1: Loneliness Type Classifier Gradient Boosting · 20 features · 5-class output · ~94% F1 Classifies: Healthy / Situational / Social / Existential / Chronic

Model 2: Social Health Score Regressor Gradient Boosting · Same features · Continuous 0–100 output · ~4pt MAE

Model 3: Crisis Risk Predictor Random Forest · Class-balanced · Binary crisis probability · ~0.88 F1

• Social Health Score — 0 to 100 clinical-grade score
• Loneliness Fingerprint — 5-dimension behavioral signature
• Type Probabilities — ML confidence across all 5 types
• Signal Breakdown — exact point-cost of each behavioral signal
• Plain English Analysis — every signal explained without jargon
• Personalised 4-Week Blueprint — week-by-week recovery plan derived from real answers

Scans your city for social opportunities matched to your personality (introvert / ambivert / extrovert) and loneliness type. Ranks results by connection potential with the specific psychological mechanism behind each recommendation — not a list of events, a social prescription.

Monitors the 6-week pre-crisis window — the period before someone feels lonely enough to seek help. Research shows behavioral signals deteriorate 4–8 weeks before subjective crisis. Three alert levels:

• Monitoring (score > 55) — watching silently
• Gentle Check-in (score 35–55) — nudge and support prompts
• Active (score < 35) — crisis resources + trusted contact alert

Integrated with iCall India crisis line: 9152987821 (Mon–Sat, 8am–10pm)

The most novel feature in the app. No other tool does this.

From 8 answers alone, the Drift Interceptor:

1. Reconstructs when the drift began — estimated weeks since social collapse started
2. Identifies the root cause — remote work isolation / digital substitution / social avoidance / circle contraction
3. Prescribes a single turning point action — matched to the specific root cause and loneliness type
4. Calculates 30-day recovery probability — with and without the intervention

Example output: "Drift began approximately 8 weeks ago. Root cause: Remote work isolation. Turning point: Join one recurring weekly activity. Recovery probability with action: 73%. Without: 38%."

Every assessment generates a downloadable clinical-grade PDF report via ReportLab:

• Social Health Score with visual gauge
• Loneliness Fingerprint bar chart
• All 6 behavioral readings with status
• Key signals detected with severity
• Personalised interventions
• Crisis support contacts

Shareable with therapists, doctors, and university wellbeing teams.

Generated by data/generate_data.py:

• 2,000 simulated users × 26 weeks = 52,000 records
• 5 loneliness type profiles with clinically-calibrated behavioral distributions
• Realistic temporal drift — signals worsen over time per type
• 15 behavioral + 5 demographic features per record

Type distribution:

Healthy       33%
Situational   25%
Social        18%
Existential   14%
Chronic       10%

All signals grounded in peer-reviewed research:

• Cacioppo & Hawkley (2003) — Late-night activity as isolation predictor
• Cacioppo et al. (2008) — The 5-contact weekly threshold
• Holt-Lunstad et al. (2015) — Loneliness = 15 cigarettes/day mortality risk
• Lim et al. (2020) — Weekend void effect (3× predictive vs weekday)
• Victor & Yang (2012) — Future-tense language as early crisis marker

• SDG #3 — Good Health and Well-Being
• SDG #10 — Reduced Inequalities (free access for all income levels)
• SDG #11 — Sustainable Cities and Communities

If you or someone you know is struggling:

India: iCall — 9152987821 (Mon–Sat, 8am–10pm) UK: Samaritans — 116 123 (24/7) US: 988 Suicide & Crisis Lifeline — call or text 988 (24/7)

Tether — The first instrument the loneliness epidemic has ever had.