It Fired
February 16, 2026. NinjaTrader 8, Sim101 account. Market opens. The Pyramidal Coherence Strategy — ten months of nights, weekends, and a few too many cold cups of coffee — places its first live trade.
I watched the order fill and felt something I hadn't expected: not excitement, but calibrated suspicion. Ten months of backtests said this thing worked. Five and a half years of walk-forward out-of-sample testing said every year was profitable. Sharpe ratio: 4.29. Max drawdown: $459 on a single contract. Win rate: 62.7%.
But live is different. Live is the proof that matters.
This is the story of how we got there — and why getting there took ten months instead of ten weeks.
Why a Structural Engineer Builds Trading Algos at Night
I'm a structural engineer by training and trade. My day job is stress analysis — figuring out how loads propagate through systems, where failure modes hide, where the math says one thing and the material says another.
It turns out that's exactly the right brain for quantitative trading.
Markets aren't random. They're noisy, nonstationary, adversarial — but they're not random. They have structure. Regimes. Stress states. And if you're willing to study that structure the way an engineer studies a load path, you can find real edge. Not guaranteed edge. Not infinite edge. But real, measurable, reproducible edge.
The target market: MES futures. Micro E-mini S&P 500. $5 per point, small enough to test intelligently, liquid enough to matter. The question I kept coming back to wasn't "how do I predict direction?" — that dead end cost me about three months — but "when does the market behave predictably?" That reframe changed everything.
The Failures First
Let me be honest about the wreckage before the breakthrough.
The Hilbert Transform disaster.
Early on, I built a regime classifier around the Hilbert Transform — a signal processing tool that estimates instantaneous frequency and phase in price data. The backtest results were extraordinary. Then I found the bug: lookahead contamination. The indicator was reading future bars. When I fixed it, the edge evaporated completely. Three weeks of work, gone. Lesson: validate your data pipeline before you trust your results. Always.
The direction prediction dead end.
Like most algo traders starting out, I spent months trying to predict whether the next candle would be up or down. I built classifiers. I tested indicators. I read papers. I got nowhere useful. Directional prediction in liquid futures markets is a losing game — the competition is too strong and the signal-to-noise ratio is brutal. The breakthrough came when I stopped asking "which way?" and started asking "what kind of market is this right now?"
The overfitting trap.
I can't count the number of "systems" I built that looked great on in-sample data and fell apart on out-of-sample. The fix wasn't fewer parameters — it was a rigorous walk-forward testing framework and a rule: no discovery counts unless it validates on data the model never saw. Everything in PCS has been tested that way. 5.5 years of rolling windows, all years profitable.
The Research Philosophy
Feynman had a rule: if you can't explain something simply, you don't understand it. I adopted that for this project. Every indicator, every filter, every regime classifier — I had to be able to explain exactly why it should work, not just that it did work in a backtest.
That discipline is what separated PCS from the systems that came before it. We weren't curve-fitting patterns onto noise. We were building a model of market behavior from first principles and testing whether the market confirmed it.
The core bet — the one that took ten months to validate — was this: markets have distinct behavioral regimes, those regimes are detectable in advance, and different strategies perform radically differently across regimes. If you can map the regime in real time, you can deploy the right strategy at the right time.
The final system — PCS — is a two-layer adaptive architecture. A meta-controller reads macro stress. A regime multiplexer reads micro conditions. Together they route capital across four strategies, each optimized for a specific market state.
Sharpe 4.29. Calmar 4.13. Max drawdown $459.
But the path there included one discovery that changed everything — something we were looking at as a filter and turned out to be the key to the whole system.