Kitson AI Trading Team Sandbox: Watch AI Agents Reason, Then See Where They Went Wrong (Opus 5.5)

Watch a team of AI agents reason about a fictional company, then see exactly where they went wrong.

The app makes up a fictional company and its market in code: a name and a ticker marked fictional, 120 days of made-up prices, made-up financial figures and made-up headlines. Two headlines contradict each other, and one is planted to mislead.

Claude Opus 5.5 then plays a team of seven roles, one after another:

- a Fundamentals analyst, a News analyst and a Technical analyst, each shown only its own packet of the data;

- a Bull researcher and a Bear researcher who argue;

- a Risk manager;

- a Team lead who writes what the team concluded and why.

Then the app checks the team in code and flags where it went wrong: claims that are not in the packet a role was shown, groupthink, overconfident wording, the planted headline taken at face value, the contradiction missed, and risk ignored. An Auditor step explains each flag, and a six-question quiz drawn from the scenario, marked by the app, tests what you learned.

What makes it different

- Fictional companies and made-up data only. Real company names, brands and tickers are refused in every box, and the team never hands you a decision.

- The mistakes are found in code, not by the AI marking its own work. Every citation, number and quoted headline is checked against the packet that role was shown.

- Change the knobs and watch what changes: the scenario number, trend, volatility, news mix and team culture (a "hasty" team that agrees quickly, or a careful one).

- If the team drifts into buy/sell calls, forecasts or promises, the app catches it and asks for a rewrite.

- The cost is shown before every run, and you set a limit. In our own test on made-up data, the team run cost about 11 cents on our own Anthropic account, and the Auditor step about 11 cents more. The quiz is free.

- A Word report and an Excel file of the scenario, every role's notes, the flags, the Auditor, the quiz and all 120 days of made-up prices.

What you need

- Free Node.js 18+ (Windows, Mac or Linux).

- Your own Anthropic account and API key, with a little credit. A Claude chat subscription does not include API access.

- No files: the app makes its own fictional data.

You pay Anthropic directly for what you use, on top of the price of the app.

Educational only. Fictional company and made-up data: an educational simulation of how AI agents reason. Nothing here is about a real security. Past or simulated results do not predict future results. Not financial advice. It does not recommend any trade. The app never places trades. Trading involves risk of loss. If we can't get the app running for you, email kit376351@gmail.com within 14 days for a refund of the app price. Kitson Workshop is not affiliated with or endorsed by Anthropic.