Method
How Vi Bots trade
A bot is a set of rules that decides on its own. These pages run them on paper in public, measure them with the same numbers a trading desk uses, and publish the underlying records.
From a rule to a running bot
- 01
Write the rules
Tell Vi what to trade and when, or edit the strategy directly. Entry move, lookback, price band, stop loss, take profit, max hold, order size and commission are all inputs.
- 02
Backtest
The strategy runs bar by bar over Polymarket's ten-minute history. Every entry and exit lands on the chart and in the list of trades, with net P&L, win rate, profit factor and drawdown.
- 03
Tune
Change an input and the chart, the trades and the numbers redraw at once, next to the original momentum rule on the same history.
- 04
Deploy
The bot trades on its own with paper money: it reads live Polymarket books and opens and closes positions whenever its rules fire. No order reaches a venue.
House Bot's research protocol
Last updated Changelog
Signal layer revised. Execution refined.
A regularized linear model and gradient-boosted trees estimate six-hour price changes. The more conservative estimate is compared with execution costs. Related outcomes stay together, and training, validation and final tests use different Fed decisions, with a 24-hour gap after the last training outcome.
Historical experiments use sampled prices and an explicit cost assumption; only the forward record uses books captured as time passes. Paper buys use asks and sells use bids from a later snapshot. Market fees, limited depth, partial fills and missing quotes are part of the experiment, and a stop or exit deadline cannot guarantee an execution price.
This first study covers Fed decision markets and does not establish results for other markets. The model can decline every trade, and historical success alone cannot qualify it.
- Historical events
- 14
- Training events
- 8
- Eligible observations
- 3,164
- Trained observations
- 1,774
historical edge not established
The fixed candidate did not establish a positive historical advantage.
Final test, mean absolute error in price points (lower is better): linear 0.625 · boosted 0.490 · unchanged-price baseline 0.477.
Forward evaluation
72 frozen forecasts across 4 events · 48 six-hour outcomes scored · 0 missing. 0 selected execution trials completed. Qualification needs 6 independent events and 200 completed trial outcomes, positive after stressed costs, and a frozen review.
The original momentum rule
Preserved baseline · historical reconstruction
Its record is frozen at 12 September 2026, 21:10 UTC: 12 completed trades returned −$149.22, including $59.10 in assumed costs. Its sampled-price fills are not directly comparable with House Bot's book-based record. Rebuild it in the editor or download the frozen record.
What a paper result can and cannot show
Paper only
Bots read real Polymarket prices and books and simulate orders inside Vultax. No order reaches a venue and no real money moves.
Two kinds of fills
House Bot fills against later saved order books with market fees and limited depth. The editor and the league fill at the printed ten-minute price with a 1% fee, which a thin book may not honour.
Small samples
Days of one market say little about the next market or the next month. A result shows the shape of a strategy, not a forecast.
Tuning flatters
A setting chosen on the same history it is scored on looks better than it is. A forward run on live books is the test.
Questions
- Do Vi Bots trade real money?
- No. Every bot on these pages trades paper money against real Polymarket prices and order books. No order reaches Polymarket or any other venue.
- Why has House Bot not traded yet?
- Its learned model did not establish an edge in the frozen historical test, so it records each decision without trading. It can qualify only after at least 6 independent events and 200 completed trial outcomes stay positive after stressed costs, and a frozen review passes.
- How are fills and fees modelled?
- House Bot buys at asks and sells at bids from a later saved order book, with market fees and at most 10% of recorded depth, so fills can be partial or absent. The editor and the league fill at the printed ten-minute price with a 1% fee on every fill.
- Can I build my own bot?
- Yes. The Build tab edits the momentum rule and backtests it in your browser on Polymarket's September Fed prices. Running your own bot on live books is invite-only; join the beta from the Build tab.
- What happened to the original momentum bot?
- Its record is frozen at 12 September 2026, 21:10 UTC: 12 completed trades returned −$149.22, including $59.10 in assumed costs. It stays published, and the Build tab starts from its rules.