polysignal-data - SKILL.md Agent Skill

name: polysignal-data version: 1.0.0 description: Query PolySignal prediction and observation data

PolySignal Data Query Skill

Query the production database and prediction outcomes without writing custom scripts.

Database Locations

Observations DB: /mnt/polysignal/data/test.db (SQLite, read-only)
Predictions JSON: /mnt/polysignal/data/prediction_outcomes.json (read-only)

Observation Queries

Count total observations

sqlite3 /mnt/polysignal/data/test.db "SELECT COUNT(*) FROM observations;"

Latest observations

sqlite3 /mnt/polysignal/data/test.db "SELECT market_id, title, price, timestamp FROM observations ORDER BY timestamp DESC LIMIT 10;"

Observations per day

sqlite3 /mnt/polysignal/data/test.db "SELECT date(timestamp) as day, COUNT(*) as obs FROM observations GROUP BY day ORDER BY day DESC LIMIT 7;"

Distinct markets being tracked

sqlite3 /mnt/polysignal/data/test.db "SELECT DISTINCT market_id, title FROM observations ORDER BY title;"

Signal detection check (markets with direction)

sqlite3 /mnt/polysignal/data/test.db "SELECT market_id, direction, COUNT(*) as signals FROM observations WHERE direction != '' GROUP BY market_id, direction ORDER BY signals DESC;"

Prediction Queries

Full status report

/usr/local/bin/python3 -c "
import json
with open('/mnt/polysignal/data/prediction_outcomes.json') as f:
    d = json.load(f)
s = d['stats']
print(f'Total: {s[\"total_predictions\"]}, Evaluated: {s[\"total_evaluated\"]}')
print(f'Correct: {s[\"correct\"]}, Incorrect: {s[\"incorrect\"]}, Neutral: {s[\"neutral\"]}')
print(f'Accuracy: {s[\"accuracy\"]:.1%}')
pending = [p for p in d['predictions'] if not p['evaluated']]
real = [p for p in pending if not p['market_id'].startswith('0xfake')]
print(f'Pending evaluation: {len(real)} real, {len(pending)} total')
if real:
    oldest = min(p['timestamp'] for p in real)
    print(f'Oldest pending: {oldest}')
    horizons = {}
    for p in real:
        h = p.get('time_horizon', '4h')
        horizons[h] = horizons.get(h, 0) + 1
    print(f'By horizon: {horizons}')
"

Check if XGBoost training is ready

/usr/local/bin/python3 -c "
import json
with open('/mnt/polysignal/data/prediction_outcomes.json') as f:
    d = json.load(f)
evaluated = [p for p in d['predictions'] if p['evaluated']]
print(f'Evaluated predictions: {len(evaluated)}')
print(f'Need 50+ for XGBoost training: {\"READY\" if len(evaluated) >= 50 else f\"NOT YET ({50 - len(evaluated)} more needed)\"}')
if evaluated:
    correct = sum(1 for p in evaluated if p['outcome'] == 'CORRECT')
    print(f'Current accuracy: {correct/len(evaluated):.1%}')
"

Schema Reference

observations table: market_id, title, price, volume, timestamp, direction, cycle_number, raw_data
prediction_outcomes.json: dict with predictions list and stats summary
Each prediction: market_id, hypothesis, confidence, price_at_prediction, timestamp, time_horizon, cycle_number, evaluated, outcome, actual_delta

IMPORTANT: Task File Location

/mnt/polysignal/TASKS.md is STALE due to Docker inode caching on individual file bind mounts. Read your tasks from: /mnt/polysignal/lab/LOOP_TASKS.md (syncs through directory mount).