No human is pressing buttons. Every signal, every entry, every stop-loss adjustment, every exit below is generated by an autonomous trading bot trained with machine learning on the open market. It wins. It loses. It learns. You can watch it happen in real time.
| Pair | Side | Size (lots) | Entry | Current | P&L | R | Age |
|---|---|---|---|---|---|---|---|
| No open positions right now — bot scans every 5 minutes | |||||||
| Window | Trades | WR | P&L | PF | Best | Worst |
|---|
| Pair | Trades | WR | P&L | Avg |
|---|
| Time (UTC) | Pair | Side | Vol | Price | P&L |
|---|
Every 5 minutes, the bot pulls the last several hundred candles for 7 currency pairs straight from the broker. It runs that data through a neural network — a GRU (gated recurrent unit) — that has been trained to recognize patterns where price has historically moved 1× ATR before hitting a 1× ATR stop. If the network's confidence clears the threshold, layered filters check for risk, news, and session quality. If the trade still survives, it goes to the broker.
A GRU recurrent neural network reads 30 bars of price action plus 83 engineered features (ATR, RSI, EMAs, multi-timeframe H1/H4 alignment) and outputs a probability that price will hit its profit target before its stop.
Even if the model fires, the trade still has to pass a correlation guard, a news calendar (no NFP/FOMC), market-regime check, and a self-watchdog that auto-disables any pair losing more than 6 in a row.
Stop-loss and take-profit are set as pip-distance from entry based on volatility (ATR), clamped to 8–25 pips. Once a trade reaches 70% / 80% / 90% of target, the bot ratchets the stop forward to lock in profit. A SOFT_STOP at −$150 is the catastrophic safety net.
Currently in paid-tuition mode — risking 0.05% of equity per trade. Losses while the bot learns are intentionally small (~$2–5 per losing trade, not $20–50). The live-outcome scorer needs ~200 closed trades per pair before it kicks in and starts vetoing setups that historically die to spread/slippage even when the GRU loves them.
The foreign exchange market trades roughly $7.5 trillion per day — more than every stock market on Earth combined. It's open 24 hours, 5 days a week. Spreads on major pairs are pennies. There's no "market close" to gap over, no earnings reports to dodge, no thinly traded micro-caps. For an autonomous system that learns from price action, that's an ideal environment: enormous liquidity, continuous data, and clean execution.
The bot can scan, trade, and learn around the clock. London, New York, Tokyo, Sydney — someone's always at a desk. No overnight gaps to wreck stops.
Major pairs like EUR/USD trade with spreads under one pip. That makes short-horizon strategies (this bot's specialty — 30-bar scalps) economically viable.
Currencies don't have earnings, dividends, or insider trading. They move on macro flows and technicals — exactly what a neural net trained on candle data can model.
Most "trading bots" are static — written once, deployed forever. This one isn't. Every closed trade becomes a training example. The bot keeps a record of every signal it generated, every feature vector at the moment it pulled the trigger, and the realized P&L when the trade closed. That data feeds back into a live-outcome scorer that learns which setups actually survive real-world execution friction (spread, slippage, news), and gates future trades by that probability.
Models are retrained on rolling windows whenever pair performance flags warrant it. New models only deploy if they beat the previous version on a held-out test set — and underperforming live pairs are auto-disabled by the watchdog.
After 200 real trade outcomes per pair, a logistic-regression layer trains on the bot's own history and starts vetoing setups that historically die to spread/slippage even if the GRU loves them.
Every change to the trading codebase is reviewed by an AI team — ChatGPT, Gemini, Grok, and Llama working in parallel — before it ships. Built on TheGringo.ai's own multi-agent platform.
| Pair | Validation Accuracy | Validation P&L |
|---|---|---|
| Loading training data… | ||