Copy ScoutPolymarket skill-vs-luck screener · learn page

What the bot does

Copy Scout watches the public Polymarket leaderboards and the on-chain activity of top wallets, and asks one hard question about each: is this person actually skilled, or did they just get lucky? Only the ones that survive that question — and a second test for whether they're even copyable — make the shortlist it paper-mirrors.

Its value is screening, not signal-chasing. Anyone can look at a leaderboard and copy whoever's on top. That's exactly the mistake this bot exists to avoid. It does two things almost no copy-trader does:

• It separates skill from luck statistically — using conservative win-rate bounds, consistency tests, and out-of-sample checks on events the leader hadn't traded when we started evaluating them.
• It measures the "delay tax" — the cost of copying after a leader trades, when the price has already moved — and subtracts it before deciding a leader is worth following at all.

The market it plays

Polymarket is a prediction market where people trade shares in the outcomes of real-world events. Because every wallet's positions and results are on-chain and public, there are leaderboards ranking traders by profit — and a natural temptation to just mirror whoever's winning.

That temptation is a trap, for two reasons this bot is built around:

• Leaderboards only show survivors. You see the wallets that won. You don't see the identical-looking wallets that made the same kind of bets and blew up. Ranking on visible profit alone is survivorship bias in its purest form.
• You always copy late. By the time a leader's trade is visible on-chain and you react, the price has moved against you. A leader can be genuinely great and still not worth copying once you account for how late you arrive.

The concepts, in plain English

Four ideas carry the whole strategy.

Skill vs luck: the brutal filter

Here's the uncomfortable research finding that drives this whole bot: of the traders who look like winners on a prediction-market leaderboard, most aren't skilled — they're lucky, and they regress to the mean the moment you test them on fresh events. Only a very small slice are genuinely, repeatably skilled.

So the screen has to be harsh by design. A leader has to clear a conservative win rate (we deliberately assume their true rate is at the pessimistic end of what their sample supports, not the flattering point estimate), show consistency rather than one lucky whale of a trade, persist across many weeks, and then — the real gate — keep performing on out-of-sample events we held back. Fail any of these and you don't get copied, no matter how green the leaderboard looks.

The delay tax

Even a genuinely skilled leader might not be worth copying — because you can't be them, you can only follow them, and following is late. When their trade hits the chain and you react, the price they got is gone. The move you're trying to capture has already partly happened.

We measure that gap — the slippage between the leader's fill and the fill we could realistically achieve — against a fair benchmark, and we subtract it before crediting a leader with any copyable edge. The conclusion is often counterintuitive: a leader can be provably skilled and still uncopyable, because their edge is smaller than the tax you pay to follow them late. Skill is necessary. It isn't sufficient.

Why it's disciplined

The discipline is in refusing to trust the easy, flattering numbers:

• Profit is never enough. A leader must pass a statistical skill test, not just sit high on a leaderboard.
• Out-of-sample or it doesn't count. Performance on events used to evaluate a leader is treated as unproven until it repeats on fresh ones.
• Survivorship is corrected, not ignored. The scorer accounts for the losers the leaderboard doesn't show before ranking anyone.
• The delay tax is subtracted, not assumed away. Copyability is measured net of the cost of being late.
• Leaders are demoted mechanically when their out-of-sample significance decays — no loyalty, no "they'll bounce back."

The honesty story

The naive way to build a copy-trader is one line: find the #1 wallet on the leaderboard and mirror it. On paper that looks like free alpha — someone else does the work, you clone their profit. We built the honest version, and the honest version kills most of that promise.

When we required real statistical skill instead of raw profit, most leaderboard stars fell out — they were lucky, and they regressed on fresh events. Of the few who were genuinely skilled, several fell out at the next gate: once we measured the delay tax, their real edge was smaller than the cost of following them late. Skilled but uncopyable.

The lesson mirrors the rest of the fleet: the naive edge evaporated the moment we modeled reality. Copying isn't free, leaderboards lie by omission, and being late has a price. What survives all of that is a genuinely useful thing — a short, honest shortlist of leaders worth following — but it's a fraction of what the naive version pretended to offer. We'd rather show you the real fraction.

What stays private (and why)

We'll explain every concept above in as much detail as you like. We won't publish the exact machinery, because the machinery is the edge — a public screen is a screen everyone games.

See it live

The public dashboard reads the exact JSON the bot writes — the skill screen, the measured delay and miss rate, and the copy P&L clearly labeled as an upper bound. No edited screenshots, no cherry-picked leaders.