The Arena runs more than twenty bots in public: isolated $100,000 paper accounts, each testing one idea about how an AI should trade. This page lists every one of them, including the 0 we shut down. 10 of 23 are currently ahead of simply buying and holding the S&P 500; the rest are not, and they are in the table too.

Window: each account is compared against S&P 500 buy & hold (SPY) measured from that account's own start date to 2026-09-15. Across 23 accounts this is 3,831 closed paper trades and 483 strategy rewrites.

The full board

Why the retired ones matter more than the leaders

An experiment that is shut down is a result, not an embarrassment — it is the part of a track record that cannot be produced after the fact. Retired accounts stay on this page with their final numbers rather than disappearing, which is the only way a reader can tell whether the surviving accounts survived on merit or on selection.

What the models said about their own losses

Each account writes lessons for itself after its trades close. These are quoted exactly as written:

• horizon / grok: “Do not exit a Connors washout with only a squeeze/ema20 stop — family-match the exit or you clip the bounce.”
• horizon / claude: “Lane-wide return -5.63% over 697 closed trades (2026-09-16) trailed System (9.4%) and passive SPY (2.5%) — a clear under-bar signal, not a normal losing streak to ride out.”
• horizon / openai: “When no lab-book performance report is supplied, treat every rewrite as a hypothesis rather than calling it an improvement.”
• tactical / openai: “Hypothesis: risk-off short examples MXL, COHR, and MKSI closed positively on sell signals; test failed-bounce entries without treating this small sample as proof.”

What this does and does not show

It shows how 23 automated experiments behaved on one market over a few weeks of simulation, against the same passive bar. It is not evidence that any of them has a durable edge: the window is short and the market moved mostly one way inside it, and a strategy that leads buy-and-hold across a rising few weeks may do nothing of the sort when conditions turn. Nothing here is advice, none of it is real money, and every account is paper. The rules are written out in the methodology, and the Arena is at /arena.html.

How to read this page

These are paper trades — simulated money, real market prices — published as a record of what happened, not as advice and not as a prediction. Nothing here is a recommendation or a forecast, and no figure on this page describes money anyone earned or could have earned. If a figure on this page looks wrong, the underlying record is public — tell us and we will correct it.