Who Audits Prediction-Market Trader Skill?

Convexly is an independent audit layer for prediction markets. Point it at a Polymarket wallet and it measures, on that wallet's public resolved positions, whether the record reflects skill or luck. It returns a four-state read, the realized entry edge with its bootstrap 95% confidence interval and the count of resolved positions behind it, and a concentration screen, all computed on one fixed, published method. Convexly takes no venue-derived revenue, so the read does not depend on the venue or on what the trader chooses to show. The free Polymarket wallet analyzer computes it on any public 0x address; first wallet free, no signup, more wallets free with an account.

What an independent audit means, versus self-reported PnL

A self-reported PnL is a number the trader chooses to show. A screenshot can be cropped, and a dollar figure can be driven by one lucky event. The venue that hosts the trades and the trader who posts the screenshot both have an interest in how the record looks.

An independent audit is a third party applying one fixed, published method to the full on-chain resolved record, with no venue-derived revenue and nothing to sell on the outcome. The output is not a headline dollar total; it is the realized entry edge reported with its bootstrap 95% confidence interval and the count of resolved positions it was computed on. Because the method is reproducible on any public 0x address, two people running it get the same read, and the read follows the interval rather than the screenshot.

The four-state read

The audit resolves a record into exactly one of four honest states, computed from the realized entry edge with its bootstrap 95% confidence interval and the count of resolved positions behind it:

1. The interval clears zero on the positive side. The realized entry edge, with its bootstrap 95% confidence interval and its resolved-position count, sits entirely above zero. The record is consistent with an edge (uncorrected for multiple comparisons).
2. The edge is not separable from chance. The 95% confidence interval around the realized entry edge includes zero on the resolved-position count available, so the record cannot be told apart from luck, whatever the dollar figure says.
3. Too few resolved trades to read. Below a floor of 30 resolved positions the interval is so wide that no verdict is honest, and the read stops there rather than guess.
4. Too concentrated to read. If a single event is at least 60% of the net result, the record cannot be separated from one outcome. The concentration flag overrides any flattering point estimate, and more volume does not clear it.

Where Edge Score fits, and where it does not

The analyzer also reports an Edge Score V3b percentile against a frozen 8,656-wallet reference cohort. Edge Score V3b is a descriptive in-sample composite 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. So treat the score as a description of how a record looks next to its peers, and treat the realized entry edge with its interval and resolved-position count as the skill-or-luck read. The two answer different questions, and only the second one carries its own uncertainty.

Diagnostics on public on-chain records, not investment advice; a past read is not a forecast, and no read is a signal to copy any trader.