Hey traders.
Last time I said the backtesting video was next. It's up:
https://youtu.be/ZPsoSZpaeNg
The short version. I had a strategy that made money on one contract in a backtest. I fixed one line in the pipeline that reads the data, and the same strategy lost money.
The bug was a 10-minute confluence on a 1-minute chart. Every 1-minute bar was reading the 10-minute candle that was still forming, and that candle already knew how it would close. So every entry could see the future. It cost me about a week of testing and a week of trading before I found it. Before that one, I'd spent hours chasing why the open looked quieter than it should, and the answer was a clock: my data was in UTC and I trade in Pacific time.
Neither problem was in the strategy. Both were in the plumbing. So when a backtest looks great and live looks nothing like it, I check the pipeline before I touch the strategy.
The habit that catches most of these: when a backtest looks good, I ask Claude Code to prove me wrong. One time it found a thin edge that broke under slippage, a check I hadn't thought to run.
I turned that into a free Claude Code skill you can point at your own backtest. It maps the pipeline, checks every indicator for lookahead, and never edits your files:
https://github.com/drewautomates/optd-starter/blob/main/.claude/skills/bias-check/SKILL.md
The video also has the prompt I'd start from to build the pipeline, and why I don't run my research in Sierra Chart's replay. Link here:
One question: which part of your backtest do you trust the least right now? Hit reply and tell me the step. That's where the next video comes from.
— Drew