Before a Bot Mirrors a Polymarket Wallet, Run the Skill-vs-Luck Check It Skips

A Polymarket copy trading bot ranks and mirrors wallets by realized PnL, and that single choice selects for luck. Ranking by PnL picks the winning tail of a variance process, so an auto-copy or mirror bot tends to lock onto wallets that took one concentrated bet and happened to win. A 30-second skill read tells you whether the wallet you are about to auto-mirror has a repeatable process or one big draw. Convexly places no orders and mirrors nothing; it is the diagnostic the bot skips. Paste the 0x address into the free Polymarket wallet analyzer and read the realized entry edge with its 95% interval before any automation inherits that wallet's entries.

Entry edge is not money

This is the trap that catches every PnL-ranked mirror bot, so it goes first. The honest read of a wallet's record is its realized entry edge: how much more often the wallet's entries resolved true than the price it paid implied, reported with a bootstrap 95% interval against zero. That is a property of the entries, and it is the right thing to read.

But a copier does not inherit the wallet's money. A copier inherits the entries and then applies their own position sizing, their own bankroll, their own timing and slippage. A wallet can show a positive realized entry edge with its interval clearing zero and still be net-negative overall, because its own sizing concentrated risk into a few resolved events. Mirror those same entries with different sizing and the result is different again. The skill read tells you whether the entries are worth anything; it cannot tell you what a bot's sizing will do with them. Two different questions, and only the first one is answerable from the public record.

Why a PnL-ranked mirror bot selects for luck

A mirror bot needs a sortable number to decide which wallet to auto-copy. The number it reaches for is realized PnL or win rate, because that is what Polymarket surfaces. Three problems compound on that choice.

First, a single concentrated bet can produce a full career of PnL. If the bot ranks on that figure, it auto-mirrors a wallet whose record is one resolved event, not a skillset. Across the full 8,656-wallet Polymarket cohort, calibration and realized PnL correlate at just Spearman +0.148, so the PnL ranking and the skill ranking are largely different orderings.

Second, realized PnL is conditional on survivorship. Wallets that took concentrated risks and lost are not at the top of the board, so ranking on PnL selects the winning tail and ignores the losing tail that took the same kind of risk. A reader, or a bot, that copies today's winning tail has no reason to assume that tail repeats tomorrow.

Third, the per-wallet PnL distribution has a heavy tail (Hill alpha 1.28, 95% bootstrap CI 1.20 to 1.36, well below equities at roughly 3.5). Under a tail that heavy, a single point estimate of “top wallet by PnL” is a noisy thing to rank on, and an interval-based skill read is the more honest primitive. On our own published top-50 board, a meaningful share of wallets were net-negative, which is exactly the kind of name a PnL-ranked bot can still surface.

The check a mirror bot skips, in three reads

Run these on the candidate 0x address before you let any automation copy it. None of the three are visible to a bot that ranks on PnL.

1. Realized entry edge, with its 95% interval against zero. If the interval includes zero, the record cannot be told apart from chance, whatever the dollar figure says. Only an interval that clears zero on the positive side is consistent with an edge on the record.
2. The resolved-position floor: 30. Below 30 resolved positions the interval is so wide that no verdict is honest. The check says “too few resolved trades to read” and stops there rather than handing a bot a false-confident number.
3. The concentration screen. If one event is at least 60% of the net result, the record is flagged as too concentrated to read as skill, and the flag overrides any flattering point estimate. This is the exact one-big-bet pattern a PnL ranking rewards and a mirror bot would copy blind.

Does Convexly publish wallets to mirror?

No. We do not recommend wallets to follow or mirror, and we publish no copy list. Recommending a wallet to auto-copy would be a forward-performance claim, and our own forward tests do not support one. Edge Score V3b, the descriptive composite the analyzer also reports, is a description of past behavior; its only out-of-sample forward test held a +0.11 rank correlation (95% CI 0.05 to 0.18) against future PnL and did not clear the pre-set +0.30 bar. The useful primitives are the free wallet analyzer (per-wallet skill read), the published wallet rankings (a descriptive board, not a copy list), and the V1 methodology paper (the ex-ante rules). Run your own check; the data is public.