How it works

The combination of three configurations: the wallet pick (who to trust), the entry pick (what to copy from those wallets) and the stake rule (how much to bet). Each block can be reused across bots, so you can create two bots with the same wallet pick and different stake rules — which isolates the effect of the stake rule.

Because that way each theory is measured without contaminating the others. If a strict bot and a high-volume one run in parallel, the difference between them answers whether strictness paid off. The alternative — changing the criterion and comparing against the past — mixes periods, opponents and luck.

No. Each bot has its own bankroll and its own cash ledger, so one experiment cannot starve another. When the cash limit is on, that bot cannot commit more than its bankroll plus realised P/L. The combined view sums the independent experiments; it is not one shared real-money account.

It narrows what the screens show to a single bot, or sums them all. Switching bots there changes what you are looking at, never what any bot does.

In the configuration, under 'Markets traded', every market starts checked. Unchecking one makes the bot ignore bets in that niche — that is how we created the bot that avoids minute-by-minute crypto.

Polymarket quotes prices (0.30 means a 30% chance). Decimal odds say the same thing more directly for a bettor: odd 3.33 returns $3.33 per dollar when it wins. The whole dashboard converts to odds; no price is shown.

Each dot is a wallet. The grey line is break-even — there the hit rate exactly equals the probability implied by the odd it pays. The further above the line, the more the wallet earns per odd. Blue dots are the wallets that passed the criteria.

If we bet ten times at odd 2.00, the market predicts five wins. Getting seven is beating the market; getting three is losing to it. Profit on its own cannot tell you which happened, because one long shot landing pays a lot while proving nothing.

That is how many bets the statistics require to tell an 8% edge apart from pure chance, at the average odd we trade. Below that, both a profit and a loss can be noise. Twice in this project a reading based on fewer than 25 bets flipped completely later on.

Every time a selection rule changes, the bets that follow carry a new marker. Summing results from different rules makes the average meaningless, so vintages appear separately on the Results screen. That is why you no longer need to wipe history when changing a criterion.

Each wallet's verdict was issued under the ruler in force at the time. Change the ruler and every verdict expires, sending the whole base back into the queue — automatically. That is why it pays to batch criteria changes: doing them right after a scan restarts costs little; doing them at the end costs the entire scan.

Because the idea that a big bet means confidence was tested and did not hold. Comparing above-median against below-median bets within each wallet across 138,000 bets, the difference was −0.0006 — and in half the wallets the small bets returned more. Filtering by size only cost volume.

In Flat mode, the configured percentage is the planned stake on every eligible bet. It is not changed by a floor, ceiling, long-odd taper or Kelly. Available bankroll, per-market exposure and order-book liquidity remain execution safeguards and may make the actual fill smaller. Variable controls only apply in By wallet conviction mode.

We exit in the same proportion it sold, measured against its entire position in that market — not against the single trade. A big wallet builds a position across dozens of small buys, so the trade that fires the signal may be $1 while it holds thousands.

Because they lose to the market. Across 76,000 sells, the sold token lost 36.7% of the time against 37.5% predicted by the odds — an edge of −0.0079, with z −6.58. It makes sense: selling is taking profit or cutting a loss, not expressing a view.

Yes. Two tracked wallets backing the same side is the same view arriving twice, and multiplying the bet multiplies the loss when it is wrong. Once the signal is identified we size once, enter once, and the position only moves when the wallet that originated it sells.

No. No order is sent and no wallet of ours is used. The odds and liquidity are real, captured at the instant of each signal, but the capital is play money.