Make three agents trade every day and you change the experiment

Here is a fun experiment: give three agents different signals and require each to trade. One reads headlines, one watches weather, and one acts randomly.

It can teach you about execution. It cannot cleanly tell you which agent recognizes a good opportunity if all three are forced to act.

Inactivity belongs in the results

A selective strategy may find no attractive price today. A must-trade rule turns that decision into an error and pushes it into a position it would normally reject.

Build a comparison group that can abstain. Give every agent the same capital rules, transaction assumptions, and access times. Otherwise you are comparing different experiments.

Freeze the rules before the run

Record the prompts, input sources, allowed markets, and evaluation period. Do not improve one agent halfway through and compare it against the others as if nothing changed.

Track turnover, exposure, drawdown, and net results. An agent that trades ten times as much can generate more exciting screenshots while taking much more risk.

A random baseline can be useful, but one short run is noisy. Repeat the setup across periods without choosing only the most flattering one.

The most revealing output may be a quiet log saying no valid trade. A good experiment should let you learn from that answer too.

Further reading: Prices and order books · Resolution rules · Current fees.

Explore the markets: Open Polymarket (referral link).