Four bot strategies, four very different failure modes

People say arbitrage as if it describes one job. In practice, four bots can use that label while doing completely different things.

One watches prices. Another reads news. A third studies resolution language. The fourth posts quotes and waits. Their risks barely look alike.

Start with the source of the edge

A latency strategy assumes one venue reacts later than another. Its test is the gap between a timestamped observation and an obtainable fill. A fast chart alone does not prove anything.

A news strategy assumes fresh information changes the event probability. It needs source verification and a record of when the news became public. Old headlines rediscovered by a bot can be expensive.

A settlement strategy depends on interpreting a contract more accurately. It needs the actual rules and an account of ambiguous cases. Confidence in an interpretation does not eliminate a dispute.

Market making depends on providing quotes that compensate for inventory risk and being selected by better-informed traders. A collected spread can be followed by a larger mark-to-market loss.

Match the test to the claim

Do not score all four by win rate. Track execution delay for the first, calibration for the second, rule errors for the third, and inventory-adjusted returns for the fourth.

The label tells you almost nothing. Ask what must remain true for the strategy to work. Then look for the exact event that breaks that assumption.

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

Explore the markets: Open Polymarket (referral link).