Best models · family = up
most-optimal model per asset · ranked by validation backtest score (Sharpe / Calmar / drawdown / return) · click a row to loadAsset profile · —
marketcap · fundamentals · upcoming events · volume
loading…
Earnings & calendar
loading…
Volume
loading…
Recent news (sentiment proxy)
- loading…
Explore — asset · target · model · backtest
Iterate below to discover better-performing variants for the selected asset.
Top picks above already optimise risk-adjusted return on the validation window;
AUC alone is not a reliable proxy for trading PnL — Sharpe, Calmar, drawdown
and total return matter more.
Price & trades
Equity curve
Backtest stats
Trades
Monthly performance — backtest window
Strategy comparison — tiered concurrent vs single-position
Same model + same backtest window, two execution policies.Most stable models & best month-to-month returns
Most stable (Sharpe-like)
Best single month
Best compounded return
Stability = mean monthly return / (std of monthly returns + downside-month penalty).
Higher = smoother month-to-month equity curve.
Regression heads (
logret) are excluded — they don't emit a confidence
for the tiered sizer.
Signal emergence — sizing tiers
Tiers: p<0.20 strong short · 10% @ cat lev ·
0.20–0.30 weak short · 5% @ 5x ·
0.30–0.70 neutral ·
0.70–0.80 weak long · 5% @ cat lev ·
p>0.80 strong long · 10% @ cat lev
Engineered features
Targets (last 400 rows of window)
Recent predictions — last 30 days
Yellow line = model probability · grey = close price ·
circles = realised outcome · click a point to drill down.
Prediction detail
click a point on the prediction chart above…top 20 feature drivers (|z-score|)
Time-series predictions — all horizons
One subplot per horizon. Shows probability/prediction vs. realised target on the
selected window. Useful to judge if the model keeps its edge as the horizon grows.
Feature engineering — check pipeline
Feature distribution
Hourly price & live spot — crypto
Source: CryptoCompare histohour (raw schema:
volumefrom /
volumeto normalised to USD turnover). Yellow dot = latest
live spot from /dashboard/live_price.
Hourly predictions vs realised — crypto
Probability path on hourly bars from the same model registry, scoped under
models/<SYMBOL>/1h/<family>/h<n>/.
Hourly backtest performance — crypto
Data manifest & sanity
System utilization · host monitor
polling…
CPU—
Memory—
Swap—
Disk—
Load (1m / 5m / 15m)—
Uptime—
Net ↑ / ↓—
Cores—
Source:
GET /api/system · refreshes every 5 s ·
process list shows top-10 by RSS.