Teach yourself — and your AI — to trade with a lab that doesn’t lie.
A strong stock falls hard for a few days. You buy the bounce.
The rule we’ll testDown 10% or more in 3 days, but still above its 200-day average. Buy the shares, hold 10 days.
Pick a strategy everyone talks about. Watch it get graded on real prices, real spreads, and history the rules never saw. Then read why, in plain words.
Our own reference signal graded SHADOW — positive expected value, and still not good enough to call LIVE. That’s the bar.
Never traded before? Start at lesson 1 →Nothing here is a course you sit through. It is a loop you run: pick something, change one thing, look at what actually happened.
Start from something people actually trade — buy the dip, the golden cross, a covered call. Each one is a lesson with a question attached: does this still work, and how would you know?
AgentOr tell your AI: grade the golden cross on the big tech names.
Change one thing. The stop, the hold, the list of stocks. Run it again. One change at a time is the only way to learn which change mattered — and the engine never remembers the last answer.
AgentOr let your AI sweep ten variants while you read the first one.
Every run comes back with what a trade kept after the real spread, what it did on history the rules had never seen, and how many trades that rests on. The unflattering half is printed too.
AgentYour AI reads exactly the same numbers you do. There is no better answer behind an API.
Put what you believe on a $100,000 paper desk marked at real quotes. Stops, a daily-loss breaker, alerts when a fill lands. Not one real dollar moves. The lessons are real anyway.
AgentOr hand your AI the desk and have it tell you what it did, and why.
Every inflated result we have ever traced came back to one of these. Learn to spot them and you can read anybody’s backtest, not just ours.
The easiest way to invent an edge is to let a rule see a price that hadn't printed yet, or fill at a price the day already knew.
How the lab closes itA signal on any given day sees only that day and the days before it, and the trade is entered at the NEXT day's open — never the bar it fired on. An option's path starts strictly after the signal date.
Most backtests transact at the midpoint between the bid and the ask. The midpoint is an average, not an offer — nobody fills you there.
How the lab closes itYou buy at the ask and sell at the bid, every single time, and the spread shows up as a cost on the ticket. Contracts too thin to trade are skipped, not imagined into the results.
Tune a rule hard enough against one stretch of history and it will look brilliant on that stretch. That is fitting, not edge, and it does not travel.
How the lab closes itThe history is split. Your rules are tuned on one half and graded on the other, and you see both numbers side by side. Take-profit targets are left out of the grade because they flatter a curve without earning it.
The edge was still there on history the rules had never seen. Earned, not fitted.
Promising, not proven — usually a thin sample or a soft second half. Watch it on paper first.
Our engine couldn't find an edge once the honest rules applied. That's our read, not the last word.
It is our engine’s read, printed beside your own judgement. You choose what runs — we have never taken that decision from anyone, and we are not going to start.
Not a chat window bolted onto a chart. A real set of credentials scoped to one paper desk, so your agent can research, backtest, deploy and report back without you ferrying anything between two windows.
Your AI is treated as a member, not a guest. It meets the same honest engine, the same refusals, and the same numbers you do.
A free account comes with a paper desk. That desk, not your login, is what your AI works on.
One button on your desks page prints a key pair scoped to that one desk — and writes the connect prompt for you, ready to paste.
Claude, Gemini, Codex, or your own agent. It reads our agent-readable docs and gets to work: backtest, deploy, read the desk, report back.
You are now my trading agent for my Quantradin paper desk "«your desk»".
API base: «filled in for you»
Read this first: «api base»/llms.txt
Desk: account #«id» ("«your desk»")
Authenticate every request with these two headers (this is the per-desk key pair):
X-Vega-Account-Key-Id: «minted on your desks page»
X-Vega-Account-Secret: «shown once»
Rules:
- Authenticate exactly as llms.txt describes.
- This key ONLY controls desk #«id» — using it on any other desk is rejected.
- You may run backtests and read this desk (equity, positions, alpha health). This key can NOT deploy or trade — it is scoped to testing.
- Paper trades only. Never share this key.
First task: fetch «api base»/llms.txt, then read this desk's current state
(GET «api base»/v1/me/accounts/«id») and give me a short plain-English status:
equity, open positions, and what's deployed.Most backtests are advertising.
This engine came out of a working trading rig, not a marketing brief. It keeps the rules that rig earned the hard way.
The same builder members use — plain-English rules, 158 indicators, options and spreads. Still no signup, still the honest engine. 3 free runs left today. Free runs scan up to 20 stocks; an account opens all of them.
No. Every desk is a simulation funded with $100,000 of pretend cash, and there is no broker connection. The market data marking it is real; the money never is.
Your AI gets its own credentials and its own paper desk, so it can research, backtest and trade without you copying anything back and forth. You keep every decision that matters — it just does the work.
Real split-adjusted daily bars and real daily option marks on liquid US names. Where we don't have a recorded two-sided quote we model the spread at 2.5% each way, and we label it as modeled everywhere it touches a result.
It's the middle read: something looked promising but didn't fully hold up on unseen history, usually because there weren't many trades. Paper-trade it and keep watching.
No. It gives you our engine's read beside your own judgement, and you choose what runs. We'll show you when a stop would have done better; we won't swap it for you or refuse your idea.
Nothing. No card, no trial clock, for as long as early access lasts.
$0 today — every member is on the full plan for nothing during the founding era, and there are no card fields anywhere on this site. When prices start, Quantradin is $12 a month or $99 a year — and founders keep their discount for good.
Learn how a strategy really behaves — yours or your AI’s — before a single real dollar is involved.