discover

Discover — What drives your best and worst nights

Automated correlation discovery across 60 days. Finds patterns you didn't know to look for.

Steps

Pull 60 days and run correlation discovery:

python3 << 'PYEOF'
import sys, os, json
sys.path.insert(0, os.path.join(os.environ.get('CLAUDE_PLUGIN_ROOT', '.'), 'lib'))
from fetch import fetch_biometrics
from dataclasses import asdict
from metrics import (compute_correlation_discovery, compute_gut_score_correlations,
                     compute_caffeine_sleep_coupling, compute_food_item_effects,
                     compute_food_hr_sensitivity, compute_disruption_classification)

data = fetch_biometrics(days=60)
d = asdict(data)

# --- CORRELATION DISCOVERY ---
disc = compute_correlation_discovery(d)
print(f"DISCOVERY REPORT — {disc.get('n_days_analyzed', 0)} days analyzed")
print()

# Best/worst night profiles
best = disc.get('best_nights')
worst = disc.get('worst_nights')
if best and worst:
    print(f"YOUR BEST SLEEP NIGHTS (avg score {best['avg_score']}):")
    if best.get('common_factors'):
        print(f"  Common factors: {', '.join(best['common_factors'])}")
    print(f"YOUR WORST SLEEP NIGHTS (avg score {worst['avg_score']}):")
    if worst.get('common_factors'):
        print(f"  Common factors: {', '.join(worst['common_factors'])}")
    print()

# Top correlations
correlations = disc.get('correlations', [])
if correlations:
    print("TOP CORRELATIONS:")
    for c in correlations[:10]:
        sign = "+" if c['direction'] == 'positive' else "-"
        print(f"  {sign}  {c['feature']} → {c['outcome']} r={c['r']} (n={c['n']})")
    print()

# --- CAFFEINE COUPLING ---
meals = d.get('meals', [])
sleep = d.get('sleep', [])
if meals:
    caf = compute_caffeine_sleep_coupling(meals, sleep)
    if caf.get('n_days_with_caffeine', 0) >= 5:
        print("CAFFEINE → SLEEP:")
        print(f"  Avg daily: {caf.get('daily_avg_mg', 0)}mg ({caf['n_days_with_caffeine']} days with caffeine)")
        corr = caf.get('correlations', {})
        if corr.get('onset_latency'):
            print(f"  Caffeine × onset latency: r={corr['onset_latency']}")
        if corr.get('deep_sleep'):
            print(f"  Caffeine × deep sleep: r={corr['deep_sleep']}")
        hvl = caf.get('high_vs_low', {})
        if hvl.get('onset_latency'):
            h = hvl['onset_latency']
            print(f"  High caffeine days: onset {round(h['high_caffeine_avg']/60, 1)}min vs low: {round(h['low_caffeine_avg']/60, 1)}min")
        print()

# --- FOOD EFFECTS ---
    fe = compute_food_item_effects(meals, sleep)
    if fe.get('n_foods_analyzed', 0) > 0:
        print(f"FOOD EFFECTS ({fe['n_foods_analyzed']} foods analyzed):")
        if fe.get('best_foods'):
            print(f"  Best for sleep: {', '.join(fe['best_foods'][:3])}")
        if fe.get('worst_foods'):
            print(f"  Worst for sleep: {', '.join(fe['worst_foods'][:3])}")

        for name, effect in list(fe.get('food_effects', {}).items())[:5]:
            metrics = effect.get('metrics', {})
            n = effect.get('n_nights', 0)
            effects_str = ', '.join(f"{m}: d={v['cohens_d']} ({v['direction']})"
                                    for m, v in metrics.items())
            print(f"  {name} (n={n}): {effects_str}")
        print()

# --- FOOD HR SENSITIVITY ---
    heartrate = d.get('heartrate', [])
    fhs = compute_food_hr_sensitivity(meals, heartrate)
    if fhs.get('n_foods_analyzed', 0) > 0:
        print(f"FOOD HR SENSITIVITY ({fhs['n_foods_analyzed']} foods, avg recovery {fhs['overall_avg_recovery_min']}min):")
        if fhs.get('flagged_foods'):
            print(f"  Slow recovery foods: {', '.join(fhs['flagged_foods'])}")
        for name, stats in sorted(fhs.get('foods', {}).items(), key=lambda x: -x[1]['avg_recovery_min'])[:5]:
            flag = " ⚠" if stats['flag'] == 'slow_recovery' else ""
            print(f"  {name}: {stats['avg_recovery_min']}min avg recovery (n={stats['n_meals']}){flag}")
        print()

# --- DISRUPTION HISTORY ---
    disrupt = compute_disruption_classification(d.get('sleep', []), d.get('readiness', []), d.get('spo2', []))
    if disrupt.get('n_disruptions', 0) > 0:
        by_type = disrupt.get('by_type', {})
        print(f"DISRUPTION EVENTS ({disrupt['n_disruptions']} detected):")
        for dtype, count in by_type.items():
            if count:
                print(f"  {dtype}: {count}")
        for e in disrupt.get('events', [])[-5:]:
            print(f"  {e['day']}: {e['classification']} ({e['recovery_shape']}-shape, {e.get('days_to_recovery', '?')}d recovery)")
        print()

# --- GUT SCORE PATTERNS (if Suna connected) ---
gut_scores = d.get('gut_scores', [])
if gut_scores:
    gc = compute_gut_score_correlations(gut_scores, sleep)
    print(f"GUT SCORE PATTERNS ({gc.get('n_days', 0)} days):")
    print(f"  Avg score: {gc.get('avg_score', 0)}")

    corr = gc.get('correlations', {})
    for k, v in corr.items():
        print(f"  {k}: r={v}")

    if gc.get('best_day') and gc.get('worst_day'):
        dow = gc.get('by_day_of_week', {})
        print(f"  Best day: {gc['best_day']} ({dow.get(gc['best_day'], '?')})")
        print(f"  Worst day: {gc['worst_day']} ({dow.get(gc['worst_day'], '?')})")

PYEOF