I watched traders lose millions on strategies that were fake. So I built an AI that solves that
Here's the dirty secret of trading: almost every "edge" you see online is fake. Not intentionally — people just don't know how to test properly. They overfit to historical data. They don't correct for multiple testing. They cherry-pick the one timeframe where it worked. They look at in-sample results and think they found something.
I saw this from the inside. I'm a quant — I spent years watching even smart traders make the same statistical mistakes over and over. And the tools out there? They let you do it. Pine Script doesn't stop you from overfitting. Python backtesting libraries don't warn you about data snooping. QuantConnect gives you 16 hours of compute per experiment and zero guardrails on methodology.
So we built Varrd.
It's not a backtesting tool. It's a system that an LLM gets placed into where it is physically impossible to test wrong. The AI runs inside a framework with hard guardrails:
Bonferroni correction — every additional test you run adjusts the significance threshold automatically. You can't p-hack your way to an edge.Out-of-sample is sacred — the system locks OOS testing to one shot. You don't get to peek and re-run. Once it's done, it's done. Fingerprinting — every unique combination of formula, market, horizon, and entry is tracked. You can't accidentally double-count or test the same thing twice.Real event studies, not curve-fitted backtests — we measure forward returns from pattern signals, not fit a model to historical equity curves.Stop-loss and take-profit optimization — grid search across real parameters, not vibes, Multi-market validation — your idea has to work across markets, not just the one you cherry-picked. and much more behind the scenes work along the same lines
You describe any idea in plain English. The AI loads real data, builds the pattern, charts it, runs the event study, backtests with stops, optimizes, and gives you a complete trade setup. 12 minutes. If there's no edge, it tells you. That's a valuable result too — you just saved yourself months of trading a fake strategy. (it even does it autonomously for you while you sleep)
Where we're at:
- 7,459 hypotheses tested
- 1,018 validated edges (14% hit rate — the other 86% were killed, which saved those traders real money)
- Pilot programs with Tier-1 brokerages
- Validated by Wedbush Securities personally
- AI agents are our fastest growing user base — we built MCP and CLI access so other AI systems use VARRD as their research engine
The biggest thing I've learned building this: the world doesn't need another backtesting tool. It needs a system that refuses to let you lie to yourself.
Happy to answer anything about the methodology, the guardrails, or why 86% of trading ideas don't survive real statistical testing.
Replies