About PredCheck

PredCheck analyses one slice of Polymarket: short-term crypto markets. That means the 5-minute, 15-minute, hourly and four-hourly Up or Down markets on supported assets such as BTC, ETH, SOL, XRP, DOGE, BNB, HYPE and ZEC.

Sports, politics, elections, equities, FX and longer-dated crypto markets are out of scope. A market only enters the dataset when its underlying asset is on the supported list and its price window matches one of the four supported timeframes, so out-of-scope trading cannot affect any figure on this site.

The unit of analysis is a round: everything one trader did inside one short-term market. A trader who buys, partially sells, takes the opposite side and re-enters inside a single 5-minute BTC market has traded one round, not six trades.

This matters because short-term traders fire many orders per market. Counting fills as trades makes win rate and average return meaningless. Counting rounds makes them comparable across very different trading styles.

Trade PnL is PredCheck’s own calculation, built from raw fills rather than taken from anywhere else. It is the gross trading result — everything received from sales and settlement, minus everything paid in — less the fees each market’s published schedule implies. That last subtraction is the point of the number. A gross profit figure is the easy one to publish, and on these markets it is the misleading one, because fees here are a first-order cost rather than a footnote.

The schedule is rate × size × price × (1 − price), which peaks at $1.75 per 100 shares at even odds and falls away towards the extremes — so a trader working the middle of the book pays several times what one trading at 90 cents pays on the same volume. The arithmetic is exact; what public data does not disclose is which side of a fill was the aggressor, and only the aggressor pays. PredCheck starts by charging every fill as if it were, then hands back the maker share: the venue funds its maker rebate from 20% of taker fees and splits it by each maker’s fee-equivalent, so a wallet’s rebate is a direct reading of how much of its flow was maker flow. Where a wallet’s rebates plainly come from markets outside the ones measured here, no split can be read off them and the upper bound stands. Either way the fee is labelled Estimated.

Rebates and rewards are settled income, but the venue credits them to an account rather than to a market, and no payout says which market earned it. So PredCheck reports two figures rather than one. Trade PnL is the trading alone and never includes them. Overall PnL adds them, and is what the account took home — the figure to compare against the venue’s own profile. Their halves do not span the same markets: the trading is short-term crypto only, the incentives are account-wide, and the page says so wherever Overall PnL appears. Payouts also arrive in batches rather than round by round, so a short window can show a trading loss with the rebate that covers it not yet paid.

A trader whose payout feed has not been read yet reads Unavailable rather than zero, and no Overall PnL is stated at all. Referral income is paid by the venue but PredCheck does not read that feed, so it is always Unavailable and never counted as zero.

Where a trader acquired inventory by minting complete sets rather than buying on the book, PredCheck charges the real 1 USDC cost of each set. Without that, a market maker’s sales would look like pure profit.

Copy Score runs from 0 to 100 and measures how practically followable a trader’s record has been over a window — the rolling day, week or month, matching the period filter. It weighs net profit, the average edge per round and how statistically solid that edge is, the share of trading days that closed profitable, rounds per day, and drawdown relative to profit. Win rate is deliberately left out: on a binary market it is the easiest number to fake with a small sample, and a trader who manages risk well is profitable at 50%.

No score is published until a trader clears every bar for the window: enough completed rounds (100 in a month), at least ten rounds on each day they trade, an average stake of $20 or more, at least $1,000 of net profit, an average ROI of 5% per round, and wins that cover the losses with room to spare. Those profiles show “Not enough data” and say which bar was missed, instead of a number invented from a lucky week.

Copy-Friendly identifies profitable traders whose results have been consistent enough, with a large enough average edge per round, that the pattern is realistically observable. It is an analytics classification. PredCheck does not offer copy trading, and this is not a recommendation to copy anyone.

Maker / Rebate identifies traders whose economics lean on liquidity provision — dense order flow, thin average edge, heavy volume. Because no payout the venue publishes says which market earned it, the call is inferred from the shape of a trader’s order flow rather than read from settled rebates, and it is marked as estimated. A thin average return alone is never enough to qualify.

Fade Candidates are traders with repeated, structural underperformance across a substantial sample. One catastrophic round can never produce this label. It is informational analytics, not advice to trade against anyone.

Accuracy matters more than filling a cell. Where a value cannot be determined from a settled source, PredCheck shows Unavailable; where it is derived rather than settled, it shows Estimated. Neither is ever silently replaced with zero.

Raw fills are retained so that every metric can be recalculated from source if the accounting improves. PredCheck is not affiliated with Polymarket, and makes no claim about any figure published elsewhere — it simply calculates its own.