Bitácora
Working notes from the studio — trading systems run on real capital, engineering decisions made alone, and the consulting playbooks behind the products. Documented honestly, losses included.
- trading
Two dollars and thirty-six cents
A strategy crossed its kill line by about one percent. The whole question of whether any rule in this system means anything came down to whether we honoured that.
- trading
Six wins, no losses, and that was the problem
A strategy with a perfect record was retired today. Not despite the perfect record — because of it.
- trading
The rules that could not win
We ran a thirty-day experiment to test a strategy. It failed. Then we did the arithmetic and found the rules we'd written could never have succeeded — in any market, in any month. Thirty days to learn what two lines of maths would have said on day one.
- trading
Fifteen minutes behind the market
The first real loss, a green trade that turned red, and the discovery — made while losing — that every trade for two months had been priced against a market that no longer existed.
- trading
Sixty days of patience
We wrote this post's outline in June with a note: 'DO NOT write early — the data finishes the post.' The gates have now spoken: one graduate, two kills, one extension, two clean records. This is what they said.
- trading
The profitable strategy we refuse to deploy
A demo book up double digits in fifty days, three objections that each died on evidence, and the one that survived: a real strategy can still be undeployable — for infrastructure reasons.
- trading
The first rejection was a gift
Day two of live trading: the machine fired its first real order, the broker said no, and the investigation uncovered a ghost that paper trading had been silently feeding us for months.
- trading
Why I'm building a trading company in public
Eleven years ago a workshop taught me that most ideas deserve to die quickly. It took me this long to apply that lesson to my own code — and to bet on myself the way I've always bet on my employers.
- trading
The night before the first real dollar
Getting money into a US brokerage from Argentina, the $500 probe that answered what support wouldn't, a deposit that impersonated a profit, and why go-live morning is designed to be boring.
- trading
The ninety-nine percent
Thirty-two strategy families audited in one afternoon: one forgotten winner, nineteen graves, and the discovery that the discard rate isn't the failure rate of the company — it's the product.
- trading
The fast seat was never the problem
7,549 arbitrage windows a day, a quarter of them reachable with plain Python from a $6 server — and 16,370 simulated trades proving that latency was never the constraint. The toll was always bigger than the dime.
- trading
The 82% win rate that lost 71%
The #1 strategy in an open catalog by composite robustness score — six tests, weighted battery, real engineering. On long out-of-sample data it lost 71%. The killer wasn't fees: it was geometry the robustness score couldn't see.
- trading
The graveyard of 25 trading bots
An AI surveyed a decade of my project folders and found 25 trading systems. Not one died of a bad idea — they all died of invisibility. This is the autopsy, and the fix.