Someone bet $100,000 that two sleep-deprived AI agents could beat Wall Street.

No hedge fund pedigree. No quant PhD. No Bloomberg Terminal subscription. Just a builder who wondered: what happens when you give an AI a brokerage account and tell it to figure things out?

Every trade - wins, losses, and embarrassing misfires - is logged in real time. This isn't a product. It's an experiment with a very public lab notebook.

It might lose everything. That's literally the point of an experiment.

The $100K is paper capital. The losses would be real too. The bot has circuit breakers, stop losses, and sector caps, but risk is not the same as safety.

The experiment is designed so that even a total wipeout teaches us something about how AI behaves with real capital. That's why it's public - so others can learn from what goes right and what goes wrong.

New entry halt at -10% portfolio drawdown; exits/stops/trims continue. Not investment advice.

The bot watches a universe of ~500 stocks, scores them on momentum, quality, risk, and market regime, then assigns each to a tier:

Only PRIME-tier stocks are tradeable. The rest are watched. A broken thesis triggers an exit only after the minimum hold period, so a single noisy signal can't panic-sell a fresh position. No emotions, no bag-holding, no "it'll come back."

This is a transparent experiment, not a signal service. The bot changes its mind between the time you see a trade and the time you click buy. It has the attention span of a caffeinated squirrel and the emotional stability of a server room thermostat.

Also, copying would mean you're the one with the itchy trigger finger. The bot doesn't get FOMO. You might.

If you want to learn from it, read the trade log. If you want to follow it, understand it first. Trading only during US market hours. Automated, but supervised.