Blog

The lab, open

AI & Method · Pillar

Building trading systems with AI: the complete working method

Oracle mode vs colleague mode, the 5-rule contract, the real prompts I use for specs, implementation and adversarial review, and the true timings: one day for an upgrade that halved a drawdown.

Read the article →
Python · Binance API

Quant research with Python and Binance API feeds

The complete pipeline: downloading and auditing data (REST klines and websocket), building leak-free features, testing with in-sample/out-of-sample and triple barrier. With starter code.

Read the article →
MQL5

MQL5 done right: the mistakes that cost real money

New bar vs every tick, normalized volumes, failing orders, server-side state, idempotent OnTradeTransaction: the reasoned list of where most accounts die — and where AI-generated code fails most often.

Read the article →
Backtesting & Overfitting

The backtest is an adversary: overfitting and how to defend yourself

The five doors of self-deception, my most instructive mistake (fixed-point thresholds on gold that tripled) and the minimum defense protocol: single-vote IS/OOS, robustness, walk-forward, relative units.

Read the article →
Case study · EA

Kimera RTO_SMC: from idea to EA, and 590 trades under interrogation

How a discretionary reading becomes measurable thresholds, the 5-module architecture of a professional EA, and the trade-by-trade analysis that cut drawdown by 47% without touching the entries.

Read the article →
Debunking

Anatomy of a deception: my +641% dismantled in 5 minutes

NeuralPath wins 81% of the time across nine thousand trades. It's also mathematically doomed. The analyst's five minutes, applicable to any public Myfxbook page — including mine.

Read the article →
HFT · Binance

Bitcoin latency arbitrage: when the edge is the structure

Two prices for the same asset, an exploitable delay, three rewrites and a −$127.02 failure documented line by line. The cost-gated v6: it only enters if the edge exceeds 2× live-measured costs.

Read the article →
Research · Negative result

The value of NO: a complete negative-result research project

210,240 BTC bars, four independent tests, one verdict: 5-minute price action has no tradable edge at retail costs. Why publishing a failure is worth more than a hundred pretty backtests.

Read the article →
Prop firms

Prop firms: the game inside the game

What you really buy with a challenge (an option, not capital), the math of drawdown constraints, and the playbook from 50+ managed challenges: you pick the system for the constraints, not the other way round.

Read the article →
Indicators · VWAP

VWAP: what it is, why it matters, how to actually use it

The volume-weighted average price is the benchmark of institutional execution: objective context, not prophecy. With standard deviation and bands: statistical levels for TP and SL.

Read the article →