Your market.
One clear signal.
A calm, read-only view of model health, risk and market conditions—designed for your iPhone.
What matters now
Predictions are being evaluated without placing trades.
Recent observations
Learning intelligence
Calibrated XGBoost V2
Three calibrated classifiers using leakage-safe, past-only features.
- Model set
- 20260724T204215Z
- Features
- 40
- Symbols
- 10
- Archive rows
- 4,301
Horizon quality
Purged walk-forwardWalk-forward drawdown above limit.
Failed return, positive-fold and drawdown gates.
Collapsed fold and below-random balanced accuracy.
Deep-learning readiness
TCN challenger3 of 6 checks pass.
History3.9 / 30 days
Examples / horizon3,802 / 25,000
Minority class809 / 2,500
Symbols10 / 10
Market regimes5 / 4
Feature completeness100% / 75%
Training datasetView details
Classes by horizon
Regime coverage
First-touch volatility-barrier labels at 15, 30 and 60 minutes; five-minute samples; horizon-sized purge gaps.
Feature system40 inputs
Price & volatility · 14
Returns, volatility, EMA, RSI, ATR, candle shape and volume.
Time · 4
Cyclical hour and minute encodings.
BTC & regime · 10
BTC state, relative returns, breadth and regimes.
Order book · 3
Imbalance, spread and freshness.
Futures · 7
Funding, open interest, ratios and freshness.
Coverage · 2
Liquidations and data completeness.
Safety & validationAlways enforced
✓ Read-only input and isolated shadow storage
✓ No exchange or live-order imports
✓ Fees 0.08% + slippage 0.04% included
✓ Collapse and non-overlap checks
✓ Execution disabled