Walk-forward, no look-ahead
Every gate is trained strictly on data preceding its evaluation window. Expanding multi-fold and two-half splits — all results out-of-sample.
Grexie Smart Grids
Grid trading that knows when to get out of the way.
Grid bots harvest mean reversion beautifully in ranging markets — and blow up in trends. Smart Grids add a learned trend-onset gate and a squeeze re-entry state machine that stand the grid down before a trend wrecks it, then re-enter once the market re-consolidates.
Free paper & demo trading. Live trading unlocks with a subscription.
The grid edge
A grid bot is a ladder of resting orders
Place buy orders below the price and sell orders above it on a geometric ladder around an anchor. As price oscillates, the grid buys low and sells high, recycling inventory and booking the spread on every swing. In a ranging, mean-reverting market it earns a steady positive cash flow per oscillation.
The failure mode
In a trend, a naked grid accumulates a loss it can't escape
When price walks away from the anchor, the grid keeps adding one-sided inventory whose mark-to-market loss grows without bound — until a stop fires or the position is liquidated. On 1-minute alt-coin perpetuals, naive grids showed a median maximum drawdown near −90%, with many instruments fully blown up and routine liquidation at leverage ≥ 2×. That is the lived failure the strategy must avoid.
The fix
Detect the trend, stand down, re-enter on the squeeze
Smart Grids don't predict price direction — a low-signal problem. They classify regime: a logistic trend-onset gate flags impending large moves of either sign and stands the grid down, flattening inventory. The grid re-enters only when the market squeezes — moving averages converge and go flat — re-anchoring on the new range instead of chasing the trend it just dodged.
- Worst-half drawdown reduced on 26 of 26 instruments tested
- Every naked blow-up (ALGO, DOGE, ZEC, WLD) rescued into controlled, positive territory
- The edge survives a 10×-fee / 30%-missed-fill abuse matrix
- Return scales predictably with leverage; the gate generalises far beyond its tuning set
Validated to be hostile to its own conclusions
The headline isn't “it made money in a backtest”
It's that the relative structure survives walk-forward validation, parameter sweeps, and a deliberately abusive fee-and-fill stress matrix. The comparative conclusions — not a single cherry-picked curve — are the bankable output.
Parameter survival
Results must hold across a grid of spacing, stop, leverage, and timeframe — not at a single tuned point — guarding against knife-edge overfitting.
Adversarial fee & fill abuse
A fee × fill-rate matrix to 20bp maker (≈10× realistic) and 70% fill rate. The median stays positive with 97% of instruments positive in both halves.
Honesty metrics
Time-in-market, fee drag, gate pass-rate, and log-return per 1,000 fills — a compounding-neutral efficiency measure reported alongside return.
Production-grade, not a notebook
One engine, driven from your desk or your phone
The signals daemon runs on a Linux or macOS host you control and owns the trading runtime. The native macOS and iOS apps are clients: they hold a single full-duplex line-JSON connection to the daemon's control socket, proxied over SSH and Tailscale, and multiplex every request and live subscription over it.
Honest about its own optimism
The only sufficient test is live forward execution
A backtest's frictionless fills flatter absolute returns. So the system measures exactly how the simulation flatters itself — and ships the antidote: a production-grade paper-trading harness that takes real fills with real queue position and reconciles them against the backtest baseline every day. Prove it yourself before you fund it.
Start free
Run Smart Grids in paper and demo at no cost on the macOS and iOS apps. Unlock live trading with a subscription when you're ready.
Contact us