Judex reviews what you own, identifies new opportunities, and presents clear, evidence-backed recommendations. Every conclusion comes with a complete decision trail, from the data considered to the outcome measured.
Two engines watch daily. They recommend; you decide. Execution follows only your verdict — and every recommendation lands on a scorecard neither of you can retouch.
It judges every holding you already own against a five-rung risk ladder — rungs chosen before the loss, when nobody was upset. On a drop it never re-asks "is this good?"; it asks only which rung you're on, and sizes the move.
It scans a hundred names on one doctrine, through three gates. Most days its best work is the list of things it did not offer you — damage is refused, only a real dip in a strong business is passed on to you, sized and stamped.
Every ask ends at one screen — the action, the size, the reason, and the price it saw. Three first-class outcomes. There is no wrong button; there is only a recorded one.
Any recommendation, months later, reopens with the exact data it saw, the exact rules file it was judged under, and the exact code version that ran — threaded end to end by a single run id. Auditable like a docket, not a dashboard.
Most tools want your trust. Judex is built to be audited.
Once a month: what was asked, what you did, and what both were worth. Every call carries two independent grades — and the advice you ignored is scored as counterfactual P&L, so following it and ignoring it are both measured, honestly.
Was the system right — graded forward from the suggested price, whether or not you took it.
Did your verdict favour you — graded from your actual fill. Skipping a right recommendation cost you; skipping a wrong one saved you.
Every account, position and price is pulled — and any trade you placed by hand is matched back to the recommendation it answered.
The Control Tower rules on what you hold; the Opportunity Finder rules on what's worth owning. Both write down their reasons before you see them.
A brief, an opportunity note, and — when there's a move — one approval. You deliver the verdict; execution follows it, never precedes it.
Ask runs on your own AI key — OpenAI, Anthropic or Gemini — and answers only from your data: why a signal fired, what a trace means, what's waiting for your approval. It can even stage an action for you.
In an agent world, the one action that matters still needs your yes.
Judex already runs every day on a real-money account. Before we talk price, we want a small group putting it through real decisions — telling us where it's sharp, where it isn't, and what a verdict should feel like. No cost, no commitment.
No — you approve every order. On your agent account, approving is the execution, but only on your yes. On your main account, Judex never holds the pen at all; you place the trade, and a reconciler matches your real fill.
No. It's deterministic decision support — rules decide, Judex explains. It builds a case from evidence; the verdict, and the responsibility, stay yours.
That's a first-class outcome — "let it pass." It's recorded as silence, and the scorecard still measures, honestly, what ignoring it was worth.
A brokerage account and about ten minutes a day. You'll be asked for a verdict most days — if you'd rather not be asked, this isn't your tool, and that's fine.
iūdex — the one who says what is right