Vi Bots: a Polymarket trading bot that publishes every decision

Rules first, then the run

A bot's rule set is committed and published before its run begins, and it is not edited afterwards. If a rule changes, that is a new bot with a new start time and a new record, not a revision of the old one — otherwise every published result is really a result about hindsight.

The universe is stated too: which contracts the bot may trade, on which venue, over which period. A rule that works on one thin contract has not been shown to work on anything else, and the coverage line says so instead of leaving the reader to generalise.

How a simulated fill works

The bot steps through the published price series and fills at the printed price, charging the stated fee on entry and exit. That is a deliberately optimistic assumption in one direction — a real order has to find a counterparty at that price and might not — and a conservative one in another, since the fee is applied on every trade regardless of size.

No order reaches Polymarket, or any other venue, at any point. There is no live-money mode, no plan that enables one, and no route from a Vi Bot to an exchange.

What is available today, and what is not

The published bot runs and publishes now, and its record is free to read here and in the app. Building your own bot with Vi is in private preview: it is not on sale, no plan turns it on, and this page will say so until that changes.

The wider write-up — the reasoning behind the current rules, the preserved original record, and what an improvement would have to survive to count — is on the Strategy Lab page.

What a rule set actually contains

A publishable rule is a set of numbers, not a description. The bot on this page states its entry condition as a price move of a given size over a given lookback, its exit conditions as a stop, a target and a maximum holding time, the price band it will trade inside, the size of each position, how many positions it may hold at once, and the fee it charges itself on every trade.

Each of those is a decision that could have gone another way, and each is the kind of thing that gets quietly tuned after a bad week when nobody is watching. Publishing them in advance is what makes the record afterwards worth anything: you can check every action in the log against the rule that should have produced it.

The price band matters more than it looks. Refusing to trade contracts near zero or near one removes the trades where a small absolute move is a large relative one and where the fee is proportionally hardest to recover. A rule that does not state a band is usually one that has not thought about this.

How to judge a paper record

Count the closed trades first. Anything in single or low double figures is a sample from which almost nothing can be concluded — a hit rate over twelve trades moves by more than eight points if one of them goes the other way. The record on this page is small and says so.

Then check the fee line against the profit line. On prediction markets the assumed cost frequently exceeds the strategy's gross result, which means the rule is being asked to beat a headwind rather than to find an edge. If the fees are a large fraction of the loss, the interesting question is about cost, not about the signal.

Then look at the drawdown, not the final number. A rule that ends flat after being well underwater is behaving very differently from one that never moved, and only one of them would be tolerable to run at size.

Finally, look at what the rule refused to do. Abstentions are decisions. A momentum rule that sat out a fortnight because nothing met its entry condition is telling you about the market as well as about itself.