Project Redline Building a Trading System From Scratch#

Step 1 Research and Backtesting (HardEdge)#

Before real money is involved, a strategy needs to be tested on historical data. That’s what HardEdge does.

You describe a strategy in a config file, what indicators to use, when to enter, how to manage risk, and HardEdge replays years of market data bar by bar to see how it would have performed. It handles all the messy stuff: slippage, commissions, partial fills, session boundaries, multi-timeframe logic.

The results are honest. It doesn’t curve-fit, it doesn’t hide losing trades, and it doesn’t let you fool yourself.

IMPORTANT

Important is that i do not use the whole dataset to train the model, i only use a small portion of the data set to train the model and then i use the rest of the dataset to test the model and will never tweak and test the model again on the same data set, because the more you tweak and test the model the more it will overfit and the more it will not show any real edge.

NOTE

HardEdge also runs parameter optimisation sweeps, testing thousands of combinations in parallel and rendering the results as a 3D surface. The goal isn’t finding the single “best” setting. It’s finding settings that work across a wide range, not just at one lucky point. but usually because of overfitting i do step 2 and only after that will I optimise the parameters

Step 2 Does It Actually Have an Edge? (MCPT)#

Backtests can lie. A strategy that looks profitable might just have gotten lucky on that specific sequence of bars. MCPT (Monte Carlo Permutation Testing) is how I catch that.

The idea is simple: run the same strategy on thousands of randomly shuffled versions of the price data. If it performs about the same on shuffled data as on real data, there’s no edge. If the real result is clearly better than almost every random run, there’s a genuine signal.

IMPORTANT

Monte Carlo Permutation Test is a statistical test that is very complex for beginners and this explanition above is an overly simplified version of it, so i will not go into the details of the test, but now you understand the idea of Monte Carlo Permutation Testing.

I run 1000 statistical tests before I even consider a strategy worth developing further (that means i go back to optimise the parameters and run this test again). Most ideas fail here. That’s the point.

IMPORTANT

Passing MCPT doesn’t mean a strategy will make money going forward. It just means the historical result is unlikely to be noise. That’s a necessary condition, not a guarantee.

Step 3 Pattern Discovery (DataMining)#

Sometimes the question isn’t whether a strategy works, it’s whether there are any patterns worth building a strategy around in the first place.

DataMining uses a genetic algorithm to evolve candlestick pattern hypotheses directly from price data. It starts with random guesses, tests them, keeps the better ones, mutates and combines them, and repeats for hundreds of generations. Same idea as biological evolution, just applied to finding market structure.

The best patterns get validated on data the algorithm never saw during evolution. If they survive that, they’re worth looking at.

Step 4 Live Execution (Sentinel)#

Once a strategy is validated, Sentinel takes it live. This is the piece that actually places orders in the real market.

Sentinel is written in Rust, chosen specifically because speed and reliability matter here more than anywhere else. It connects to either NinjaTrader 8 (over a c# bridge) or Interactive Brokers (with their api) depending on what you’re running. The exact same config file used in backtesting gets loaded by the live engine, so what you tested is what gets traded. No translation, no drift.

IMPORTANT

Important is that I need to code the indicators in both Python and Rust, and the Strategy loader too, only the config file is the same json file in both components, and I don’t want to use some bride for that, This is the best solution to keep them seperate but still use the same config file.

NOTE

Sentinel runs a background health monitor, logs UTC timestamps on every trade signal, and handles order state with thread-safe logic. Live trading is where bugs become expensive, so there’s no room for sloppy code here.

Step 5 Everything Else (RedlineEngine)#

All the monitoring and review lives in RedlineEngine, a native Swift app for macOS and iOS. It has three tabs.

The first tab is performance analytics: equity curve, drawdown, win rate, expectancy. Everything I need to see whether a strategy is running clean or starting to fall apart. Both HardEdge and Sentinel will send the data to this tab.

The second tab is a visual strategy editor. Instead of writing JSON config files by hand, you build strategies through a UI. It generates the same file format HardEdge and Sentinel use, so there’s no disconnect between what you designed and what actually runs.

The third tab is trade review. For those who are familiar with Funded Firms, this tab would be familiar to you, with the calendar view, the summary stats, and the actual individual trades.

IMPORTANT

The reason it all lives in one native app instead of a bunch of browser tabs is the same reason I built everything else myself, I want one place that does exactly what I need, nothing more, nothing less.