Author: admin

Setup B survived 11.75 years of backtesting and killed my AI filter. Now it runs on a VPS with default inputs — and I promised not to touch it for 6 months.

This is Part 4 of the honest validation series. In Part 1, a coding bug inflated my results. In Part 2, I published the real backtest: 819 trades, +196R, profit factor 1.37. In Part 3, I trained an AI filter — and the walk-forward test killed it. What remains is the simplest thing of all: pure rules, running on a server, with me forbidden from interfering.

Today, Monday October 12, the EA went live.

What exactly is running

The EA is called FibGann_BreakoutEA_v2. It trades one setup and one setup only — Setup B, the M15 breakout, the only pattern that survived the full validation pipeline. Everything else (Setup A, the ML layer) is gone, not hidden.

Its rules, frozen in code:

• Stop loss: 1.5× ATR(14) on M15 — one and a half average ranges, structural, no guessing.
• Entry: the M15 close beyond the consolidation zone, no M5 confirmation trigger (kept as a documented backup variant, not shipped).
• TP1 at 3R: close half, move stop to breakeven. TP2 at 6R: let the rest run.
• Fail-breakout exit: if a M15 candle closes back inside the zone, the breakout failed — exit. Time exit: 24 hours, no lingering.
• One trade at a time, max 3 per day, 7–23 server time, spread filter at 500 points. No grid, no martingale, no averaging down.
• Risk: 1% of equity per trade, lot size auto-computed from the stop distance, broker minimums, and a margin check.

The v2 addition is two daily guards: stop opening new trades at +10% of the day’s starting balance, and close everything at −5%. Honest admission: with 1% risk and at most 3 trades a day, −5% is nearly unreachable (three full losers is −3%), and +10% only happens on a freak day. They are safety rails, not profit drivers. I publish them anyway, because secret settings are how honest systems die quietly.

What I expect — written down before looking

Expectations, set in advance, because that’s the discipline of this series:

• ~30–70 trades per year. Roughly one a week. There is nothing to watch daily.
• Winrate ~24%. Three out of four trades lose. The edge lives in the winners being 3–6× the losers.
• Profit factor ~1.2–1.4, below the 1.37 backtest. Real ticks mean real spread, real slippage, real requotes. A forward test that beats its backtest is usually a red flag, not a celebration.
• The first week is macro-heavy: US CPI Wednesday, PPI and retail sales Thursday. If the EA takes losses around those releases — spread widening, spikes, stops hit — that is the week, not the system.

If I had to bet one number: roughly +0.1 to +0.15R per trade after costs. Modest. Boring. That’s the point.

The promise

Six months, untouched. No parameter changes. No manual closes. No “just this once” after a losing streak, and no victory lap after a winning month. The backtest and the forward test must speak for themselves, on the same settings.

Why publish a promise in public? Because the hardest part of systematic trading is not the code — it’s the operator. Every EA graveyard I know was dug by someone who “improved” a live system mid-drawdown. The public promise is my insurance against myself.

Why this is the end of the research phase

The honest-quant posture is: every layer you add must earn its place. The AI filter didn’t. Setup A didn’t. Tweaks with names like “CONSOL=8” looked good until spread and slippage ate them. What survived is one setup, one risk model, one promise.

I closed new-system research deliberately. From here, the only updates are the honest monthly check-ins: trades taken, R accumulated, what the macro weeks cost, and whether the forward numbers still resemble the backtest. If they stop resembling it, I will say so — that’s what Part 2’s full disclosure was for.

---

This is Part 4 of 4. We vet tokens. We vet systems. We vet ourselves.

Start from the beginning: Part 1: the bug that lied to me. The numbers: Part 2: 819 trades, every number public. The dead AI: Part 3: the autopsy.

Not financial advice. Research and education only.

After 819 trades of pure rules, the tempting next step was AI. I built the full ML filter — then let walk-forward testing kill it.

In Part 2, I published the complete backtest of Setup B: +196R over 11.75 years, profit factor 1.37, every number public. The obvious next move was to bolt on machine learning and squeeze out more. So I did it properly — 10 clean features, no leakage, walk-forward validation — and the models told me, to my face, that they had nothing.

This is the autopsy of an honest negative result.

The setup: built to not fool myself

The ML filter’s job was simple: look at each Setup B signal and predict whether it would win. The design rules were strict:

• 10 features, all known at signal time: direction, distance from EMA in ATR units, Stochastic K/D, reward-to-risk to first target, leg size in ATR units, hour and day-of-week encoded as sine/cosine. Nothing a trader couldn’t see on the chart at that moment.
• Aggressively removed anything from the future: the outcome, the R-multiple, how long the trade lasted, the maximum favorable/adverse excursion — all banned. Also removed three features that turned out to be constants for Setup B (they’d only add noise dressed as signal).
• Label: 1 if the trade won (195 of 819 = 23.8%), 0 otherwise.
• Validation: 5-fold TimeSeriesSplit — train on the past, test on the future, never shuffled. 680 out-of-sample predictions pooled.

If the features carried no real information, this design would show it. That’s the point of doing it this way.

The result: two models, one verdict

• Logistic Regression: walk-forward AUC 0.514 — no signal
• HistGradientBoosting: walk-forward AUC 0.523 — no signal

A coin flip scores 0.50. My models scored 0.51–0.52. Per fold, the AUC swung between 0.43 and 0.60 — pure noise around the coin line, not a stable edge hiding in one regime.

The features simply could not tell winners from losers.

The illusion I almost fell for

Here is the dangerous part. Slice the predictions by confidence threshold and one number looks delicious: gradient boosting at threshold 0.3 gives 194 trades, profit factor 1.54 — better than the 1.37 baseline.

I almost wrote the “AI improves my system” post. Then I checked where the threshold came from: it was picked on the evaluation set itself. Choose the cutoff after seeing the answers, and any random model can look skilled. That’s not forecasting — that’s the threshold-selection illusion.

Discipline means deciding your pass criteria before you look.

The kill rule — set in advance, honored in public

Before training, I wrote down the bar a model had to clear to ship:

• Profit factor ≥ 1.3, and beats the 1.37 baseline
• At least 100 trades
• Walk-forward AUC ≥ 0.55

The models failed the AUC bar (0.523 < 0.55). So the training pipeline — built to enforce this automatically — refused to export the model. No ONNX file was created. Nothing shipped. No human override.

That automation matters: in the moment, with a 1.54 staring at you, a human would have rationalized. The script didn’t.

What survived the autopsy

A negative result is still a result — and this one had real value:

1. A reusable pipeline: when I have a longer broker feed, I re-run one command and get a fresh verdict instantly.
2. A verified export path: Logistic Regression exported to ONNX and reproduced identical results (error ~6e-08). Gradient boosting hit a converter incompatibility — documented so I don’t step on it again.
3. An honest saving: I now know that bolting ML onto this rule system adds complexity with zero skill. I won’t spend months operating a useless layer — or worse, trading one.

And the likely reason for the failure is instructive: my features describe each signal in isolation. They carry no market-regime context — no sense of whether gold is trending, choppy, or sitting under a news event. If I ever retry ML, that’s where the new features have to come from, on a longer broker-native feed.

The point

Adding AI to a good rule-based system did not make it better. It made it more complicated and no more predictive. So Setup B stays exactly as it is — pure rules, profit factor 1.37, no AI.

Publishing a negative result is easy to promise and hard to do. But “my AI filter failed” is exactly the kind of sentence this series exists to print.

---

This is Part 3 of a 4-part series. In Part 4, the EA went live on a VPS — and I promised not to touch it for 6 months.

Missed the numbers? Read Part 2: 819 trades, every number public. Start at the beginning: Part 1: the bug that lied to me.