Scores and badges
Three numbers follow wallets around Polywhaler: a trader-quality badge, an insider score on individual trades, and a smart money score on the wallet itself. They measure different things and are easy to confuse.
Trader quality badges#
The badge on a trade card answers one question: has this wallet earned a place on the leaderboard, and how high?
- Top 100: inside the first hundred ranked wallets.
- Top 1000: ranked, and comfortably inside the first thousand.
- Ranked: on the leaderboard, further down.
- Provisional: too few decisive closed positions for the score to mean much yet. Treat it as "not enough evidence" rather than as a weak result.
The same thresholds drive the trader-quality filter on the feed and the quality filters in the API, so a feed view and an API query agree.
Insider score#
The insider score is attached to a trade, not a wallet, and runs 0 to 100. Only Polymarket buys are scored: sells and Kalshi trades carry no insider score. It adds up five factors, each with its own ceiling, and a trade's page shows the points each one gave:
- Low probability, high stakes (up to 35): a large stake on an outcome the market thinks is unlikely. Always 0 for sports, where a big bet on an underdog is ordinary betting.
- High-risk market context (up to 20): how much private information could matter in this kind of market, rated by AI from the market and the event, with a keyword check when that rating isn't available.
- Market dominance (up to 20): the size of the trade next to the market's trading volume over the last 24 hours.
- Fresh wallet activity (up to 15): how new the wallet was when it placed the trade. Insiders often trade from a new wallet to hide who they are.
- Suspicious timing (up to 10): how close to the market's resolution the trade was placed, or an unusually large early bet on a distant event.
On crypto markets, market dominance, fresh wallet activity and timing count for a quarter of their usual points, because new wallets and thin volume are normal there and a coin's price is not private information. The total is capped at 100, and a sports trade at 50, so sports can reach medium but never high. The wallet's track record is not a factor: that is what the smart money score measures.
The score falls into one of three bands:
- Low: 40 and under.
- Medium: above 40, up to 70.
- High: above 70.
The bands are strict, so a score of exactly 40 is low and exactly 70 is medium. Trade badges, the feed's insider filters and alert rules all classify through the same function, which is why an alert set to "high" and a badge reading "high" always mean the same thing. The insider-flagged webhook for Quant subscribers draws its line slightly lower: it fires on a score of 70 or more, so it also includes trades that score exactly 70.
The exact rules behind each factor are in the Developer portal for Quant subscribers.
Note
Smart money score#
The smart money score is attached to a wallet, and it is the leaderboard's ranking number. It asks one question: does this wallet make money against the prices it pays, again and again? It is centred rather than cumulative:
- 50 means no demonstrated edge. Not bad, just unproven.
- Above 50 is evidence of a repeatable edge, below 50 is evidence against one.
- How far from 50 a wallet lands depends on both how strong the record is and how many independent weeks back it, so confidence moves the score in both directions.
Weeks, not trades#
Every closed position is grouped by the Monday-to-Sunday week (UTC) its market settled in. Markets that settle together usually move on the same news, so twenty related strikes that paid out in one week are one result, not twenty good calls. The score never counts positions; it counts weeks.
- The best week's profit is capped at the second-best week's, so no single week can carry the record.
- Each week counts for less as it gets older, so an idle record fades instead of holding its place.
- A small copy-trading cost comes off every week, so an edge too thin to follow doesn't count.
- Positions are dated by when their market settled, not by when the wallet claimed the winnings, so the timing of redemptions can't move the score.
What goes into it#
Five readings, each running from clear evidence against to clear evidence for:
- Repeatable edge: weekly profit measured against how much a week like it would swing by chance at the prices paid. A short run of heavy favourites doesn't read as certainty, because a 99¢ position can still lose everything.
- Recent weeks: the same edge over the last few weeks only, the newest counting most. A strong record from months ago cannot keep a wallet at the top while it is losing now.
- Beats the odds: how often positions win compared with what their entry prices implied. Buying at 20¢ and winning a fifth of the time is no edge; winning half the time is.
- Dollar edge: realized edge in dollars, each extra dollar counting a little less, and only to the extent the edge repeats. Size alone earns nothing.
- Return on stake: profit per dollar staked, recent weeks counting most, discounted when only a few weeks hold most of the money.
When a single week carries a large share of the profit, a positive score is discounted. A wallet that has lost money over its lifetime has its score capped, using the lower of our realized figure and Polymarket's own P&L. When the open positions are down on balance (open winners offset open losers), that net loss counts now, so taking profits while holding losers open doesn't help.
Confidence#
Confidence comes from the number of active weeks, older weeks counting for less. A hundred tickets in one week add no more confidence than one. With little evidence the score stays close to 50 in either direction.
Provisional wallets#
A wallet is provisional, shown without a rank and left out of every smart-money filter, when:
- fewer than 20 of its positions have settled with a profit or a loss;
- its results come from fewer than 4 active weeks;
- we have tracked fewer than 10 of its trades at whale size; or
- our realized P&L for it is well above Polymarket's own figure (P&L not verified), until the two agree.
The profile says which rule applies. Treat it as "not enough evidence" rather than as a weak result.
Tiers and styles#
- Genius: 80 and up, and at least $10k of realized edge (recent weeks counting most), so a long run of tiny tickets can't reach the top tier.
- Smart: 60 and up. Average: 40 and up. Naive: below 40.
Each wallet also carries a style, from what it usually pays per share (weighted by stake): Long shots under 30¢, Favorites over 75¢, Balanced in between. Style describes, it doesn't score.
Smoothing and clusters#
The ranked score is smoothed once a day, quickly downward and slowly upward. A collapse shows within a day or two; a rise has to hold.
Wallets that trade as part of a coordinated cluster lose points, so a group of wallets moving together can't lift each other up the board.
Why volume is not an input#
Trading volume is trivially washable: a wallet can trade against itself all day and manufacture an enormous number. Any ranking that rewards volume rewards that behaviour, so volume is deliberately left out, and so is the count of positions.
What changed from v3#
- Evidence is counted in settlement weeks rather than positions. v3 counted a ladder of correlated long shots that paid out together as many independent correct calls.
- Confidence comes from active weeks, not position count, so many small tickets no longer push a wallet up.
- "Recent" now means the last few weeks, newest counting most, instead of a long window that kept one old windfall in view for months.
- Return and profit no longer saturate, and the best week is capped at the runner-up.
- Provisional needs 20 decisive closes (was 10), plus enough weeks, whale trades and verified P&L.
- The rolling average was replaced by daily smoothing that falls fast and rises slowly.
Note
The full specification is in the Developer portal for Quant subscribers.
Which one to use#
- Judging a single trade you just saw: insider score.
- Deciding whether to follow a wallet: smart money score, and check it is not provisional.
- Skimming a feed: the badge, which is the smart money score compressed into something readable at a glance.