Price · SMA(20) · Bollinger(20,2) · trade markers — live tape scrolling right
ML Signal Engine
multi-timeframe · week→year
—P(up)
—
Holdout accuracy
—
Baseline
—
Edge
—
AUC
—
Learns from history →
Order Blotter
live fills, newest first
Time
Side
Price
P&L
Opportunity Scanner
whole universe ranked by ML conviction · live · click a card to open it
Risk Cockpit
One book across every desk — deployed exposure, concentration, and the daily-loss circuit breaker. Identical in sim and live; the same breaker and kill-switch guard a linked account.
Signal Provenance — why P(up) is what it is
logit decomposition of the current signal · which timeframes moved it, and the ledger hash it commits to
RSI (14)
70 / 30 bands
MACD (12,26,9)
histogram + signal
Equity Curve
recent history · strategy gated by ML confidence · $10,000 start
Execution Parity — Modeled vs Realized
demo · modeled fills vs modeled + latency (connect a broker to mirror your account)
Each signal records the quoted price, the modeled fill (slippage-adjusted), and the realized fill (adds the latency & queue slip a model can't predict). This is the proof the “Estimated” numbers survive contact with a real account — the same panel mirrors your linked paper account the moment you connect one.
🔐 Verifiable Signal Ledger
0 signals committed
Every signal is hash-chained (SHA-256) at emission — history can't be edited or cherry-picked without breaking the chain. Verify it yourself, or export the ledger. Nothing leaves this tab.
No signals committed yet
#
Time
Sym
Signal
Price
P(up)
SHA-256 (chained)
Technology demonstration — synthetic data, not investment advice.
Tickers are real companies, but the price movement shown is simulated, not live market data. The ML model is a
logistic classifier over multi-timeframe history (week→year), trained in-browser with a strict time-ordered
holdout and walk-forward evaluation; results show the engine's mechanics and discipline, not a proven market edge.
Broker connectivity is a separate, secured layer.