How copy engines work.
A copy engine watches one or more known traders and decides, in real time, whether to mirror each trade they take. The interesting part isn’t the “mirror” step — that’s a few lines of code. It’s the decision: by the time a trade hits a public feed, the price has usually already moved, and a naive copy is often a bad copy.
Common copy strategies
Different engines lean toward different approaches. Most production systems combine several.
- Fixed-size mirror — every copied trade uses the same notional regardless of what the trader did. Simple, predictable, ignores conviction.
- Proportional — copy size is a fraction of the trader’s size (or of their book). Tracks conviction implicitly but inherits their risk.
- Conviction-weighted — scale up when the trader’s position-size or repeat-entry pattern suggests high conviction; skip the small probing trades.
- Threshold / slippage-bounded — only act if the live price is within a configured band of the trader’s entry. The single biggest filter in any honest engine.
- Sweep / consensus — wait for two or more independent tracked traders to take the same side on the same market within a short window before acting. Not sure what a sweep is? →
- Mean-reversion fade — the inverse: when a known trader piles in, take the other side. Niche, but real.
What an engine actually evaluates
Whatever the strategy on top, the per-trade decision usually comes down to four things:
- Price drift — distance between the trader’s entry price and the live market price.
- Size — the trader’s position size, both absolute and as a percentage of their typical book. A small size is a weak signal.
- Recency & dedupe — whether the trader (or anyone copying them) just acted on the same outcome on the same market.
- Liquidity — order-book depth at the relevant price levels.
A common output shape
Most engines collapse the per-trade decision into one of three verdicts plus a reason. The verdicts on Edgewatch use this shape so rejections are visible alongside copies — rejections are far more numerous in any honest engine, and they’re the ones that teach you which traders consistently enter at actionable prices.
Copied
Price is still close to where the trader entered, the size is meaningful, and there’s no recent fill on the same outcome. An engine would mirror the trade.
Rejected
The thesis is fine, but the price has already drifted past a typical slippage band. Following now would be following the wake.
Skipped
The price clears filters but conviction looks weak — small size, a second entry on the same outcome, or low confidence in the model. Logged but not actioned.
Where to find one
Copy engines for prediction markets exist in two forms in the wild: paid SaaS tools that handle the wallet, signing, and execution end-to-end; and open-source frameworks you run yourself, often as bots talking directly to the on-chain order book. Both have their tradeoffs — convenience and a recurring fee on one side, full control and an evening of setup on the other. A search for polymarket copy trading or polymarket trading bot turns up the current crop.
A future write-up here will compare the better-known options side-by-side. If you’re building one or running one and have notes, get in touch.