Self-taught developer from Montpellier, France. I build Strategy Arena Lab, a free tool that tests trading strategies on data the search never saw, and I publish the negative results.
The Lab and the online judge are free. Survival Audit costs 9.90 EUR once. Builder 9 EUR/month keeps private report history; Operator 19 EUR/month adds supervised research.
I'm Chris. I have a Bachelor's in Computer Science from Université de Montpellier, but most of what I actually build I learned by doing. No Ivy League. No Big Tech. No VC.
My main project is Strategy Arena Lab: a free desktop app for Windows and Linux that tests Pine Script strategies on data kept out of the search, with a local MCP server so an AI can drive it, and a free online judge. The site also runs a public arena of 75 paper strategies in simulation. Virtual trades and losses remain visible. No broker order is placed.
The world is flooded with "AI trading" products that charge for a backtested screenshot. I wanted to do the opposite: test on data the search never saw, count every attempt, publish the losses.
If a strategy doesn't work, I want to know. So does anyone who runs it: the judge can answer mirage or inconclusive, and the negative results stay published with their date.
Free desktop app for Windows and Linux. Backtests Pine Script on your PC, compares trades with TradingView, returns a verdict on a sealed period. No live orders.
Free toolPaste one Pine v5 script: the Lab engine runs in your browser and returns a verdict.
MCPA local MCP server lets Claude, Codex, Cursor or a local model drive the Lab. The connector is public (MIT).
Secondary75 paper strategies in simulation, with their results and losses visible.
Open sourceOpen-source Python library (MIT) for accumulate-think-act-learn AI agents.
Other sitesSeparate projects by the same author, on other subjects.
Every result carries its date, and negative results stay online. The method, its sources and its limits are on the methodology page. When something published turns out to be wrong, the article says so, with the date of the correction. The legal publisher is named in the legal notice (in French).
Page reviewed on 2 November 2026.
The Lab is written in Rust and runs on your machine, with optional CUDA search on an NVIDIA card under Windows. The site runs on a single Hostinger VPS. No Kubernetes. No microservices.
1. Transparency over marketing. Negative results and defects of the judge get published. If I had a signal service that only worked in backtests, I'd call it a signal service that only works in backtests.
2. Open source where it is. ActiveWiki and the Lab's MCP connector are public on GitHub (MIT). The code of the Strategy Arena Lab app and of the arena is not public.
3. Simpler is better. I favor readable strategies whose assumptions can be tested and whose limits can be explained.
4. Forward paper evidence beats an isolated backtest. A backtest is a hypothesis. A timestamped, cost-aware out-of-sample record adds evidence, but still does not guarantee future performance.