Forecast Tracker
Public, timestamped predictions from Tachyon’s agent ensemble. Every forecast is scored against actual outcomes.
Prediction Accuracy — Last 4 Weeks
| Week | Prediction | Confidence | Actual | Result | Notes |
|---|---|---|---|---|---|
| 2026-W13 (Mar 24-30) | HOLD | 0.42 | BTC -1.2% | CORRECT | Low confidence, sideways market |
| 2026-W12 (Mar 17-23) | UP | 0.38 | BTC -3.1% | WRONG | Sentiment divergence missed |
| 2026-W11 (Mar 10-16) | DOWN | 0.51 | BTC -5.4% | CORRECT | Macro risk detected by MM Claude |
| 2026-W10 (Mar 3-9) | UP | 0.35 | BTC +2.7% | CORRECT | Low confidence but correct direction |
Running accuracy: 3/4 (75%) over last 4 weeks
Note: This tracker uses sample data from our development period. Live forecasts from the forecast tournament will replace this once the 3-class model is deployed.
How forecasts are generated
- MM Claude analyzes macro conditions, on-chain data, and fundamentals
- MM Grok reads social sentiment from X/Twitter
- ML Model generates a probabilistic prediction (currently 5-class, migrating to 3-class)
- Signal Pipeline aggregates all inputs with confidence-weighted voting
- Consensus determines the final direction call + confidence score
Each forecast includes:
- Direction: UP, DOWN, or HOLD
- Confidence: 0.0–1.0 (how much the agents agree)
- Reasoning: Key factors from each agent
- Timeframe: 7-day forward (weekly)
Scoring methodology
- CORRECT: Predicted direction matches actual BTC movement (>1% threshold)
- WRONG: Predicted direction is opposite to actual movement
- NEUTRAL: Actual movement within +/-1% (HOLD zone) — scored as correct only for HOLD predictions
We do not cherry-pick. Every forecast is published before the period starts and scored after it ends.