It started with a losing streak
Not a dramatic one. The ordinary kind. A different ticker every time, and underneath it the same mistake: a good plan in the morning, a gut-feel entry by ten.
You know the rest. The loss you promise to review and never do. The red morning that becomes a revenge trade by lunch. Knowing the pattern and repeating it anyway, because knowing is not the same as seeing it in your own fills.
What I wanted was one desk that keeps score: a journal that names the habit bleeding the account, and a read on the next trade graded against real outcomes instead of felt. Nobody sold that desk, so in early 2026, in Geneva, I started building it.
First we built it for one trader. Now we're building it for all of them. The place you go to know your trading inside out, and the best trade prediction engine ever built.
Built for traders, run on three rules
Trading is a business full of big promises. Here, nothing ships as a promise: it ships with a number attached. Three rules keep those numbers honest.
Models train on old data and get graded only on days they have never seen. Memorising history earns nothing here.
Commissions and slippage are subtracted before a single number counts. The bar is beating the cost of trading, not zero.
Public numbers include every loss. Experiments get tagged and excluded, so the figure you see is one you can check.
What shipped, and when
Tap any release for the longer version. The early entries describe an engine that was already running, so the real start is a little older than the paperwork.
In private
Nobody was watching, which turned out to be the useful part.
Doors open
The part where other people got to see it.
The second model
A different idea about what a good trade looks like.
Next up is the sizing model: which single level to act on, and how big. Watch it trade while we build it.
12 releases listed, oldest first.
Trade with the thing that keeps score
Trade Journaling is $60 a year with a 14-day free trial. Trade with AI is $25 a month with a 7-day free trial, journal included.