// blog
From the loop
How the thing actually works, written from the code rather than around it. Every claim in these posts quotes the file it came from, so you can check it rather than take our word.
RSS feed451: Unavailable for legal reasons
Every chart, screener and crypto quote in production was dead, and had been since launch. It worked on every machine we owned. The cause was geography, and the fix was one small service in the right country.
Read →
The backtest that read a quarter of the data
A request for 200,000 bars returned 54,210 of them and reported no error. The metrics were all computed correctly — on 27.6% of the sample. Here is the bug, why four reviews missed it, and the general shape of it.
Read →
09:30 is not on the hour
The first hourly bar of every US trading session covered thirty minutes and was labelled as sixty. The 4-hour chart changed shape twice a year with no change in the data. Both came from one line of arithmetic.
Read →
Why your journal is your edge
You cannot measure an edge you never wrote down. Why manual fills and bot fills belong in one table, and what breaks when they don't.
Read →
Backtesting beyond win rate
Win rate is the least useful number a backtest produces. What the other 23 are for, and the one our own Monte Carlo refuses to pretend it can tell you.
Read →
Kill switches and circuit breakers
A bot executes exactly what you told it to, including the parts you did not think through. Inside the two independent checks that stop one, and the honest limits of both.
Read →