Fading the crowd on Polymarket sports markets: 1,150 games where 90% or more of the D and F money sat on one side

When the worst-graded wallets on Polymarket pile onto one side of a game, that side does trail its price, just not by enough to prove anything. In 1,150 sports markets where 90% or more of the D and F wallets' pre-kickoff dollars sat on one side, that side was priced at 64.5 cents and won 62.3% of the time: -2.25 points, with a 95% interval from -5.11 to +0.60.

The losing wallets are not alone in that. When S, A and B wallets agreed as completely, their side was priced at 68.0 cents and won 66.7%, -1.25 points (the chart below). Both crowds landed on favorites that won a little less often than priced. A one-sided crowd, whoever it is, is a reason to check the price before you take that side, not a bet on the other side.

On September 13 this page reported 844 one-sided markets in which the D and F crowd's side came in 3.68 points short of its price, with an interval from -6.71 to -0.69 that cleared zero, and said that when the two cohorts disagreed the graded side won. Graded as of each trade, every one of those figures moved toward zero (the chart below): the penalty is smaller and no longer clears zero, and the disagreement case shows nothing.

The study took each buy's grade from the ranking row dated on or before the trade day. 0xinsider updates a wallet's latest ranking row in place when it recomputes the grade, so the row a buy looked up could hold a grade decided after the trade, partly by the result being scored. That moved losing buys into the D and F cohort after the fact, which is what a fade study is most sensitive to.

Re-run on October 7, 2026 with each buy graded by the last ranking written before it, over the markets settled by then, the 90% crowd covers 1,150 markets and its side came in at -2.25. The crowd on a favorite, published at -4.20, came in at -1.59. When the cohorts disagreed, the graded side came in at +0.10 and the crowd side at -0.55, against +0.88 and -1.92 in September.

Everything after this section is the corrected study. The sharp money study made the same correction the same day.

The sample is every buy 0xinsider recorded as a large trade on a Polymarket sports moneyline, spread, or total from June 1 to September 12, 2026, placed before the market's kickoff time, on a yes-or-no market that settled after the trade, at 2 to 98 cents. The smallest recorded buy was about $3,000 until July 2, 2026 and $10,000 after. 74,189 buys, 4,472 wallets, 8,111 markets, $1.46 billion in stakes. The grade is the last ranking 0xinsider wrote for the wallet before the buy.

D and F wallets made 24,460 of the buys and S, A and B wallets 19,129. The rest were C wallets or wallets with no grade yet. 4,709 of the markets had at least one D or F buy before kickoff, 4,442 had at least one S, A or B buy, and 2,526 had both.

For each market the query adds up the pre-kickoff dollars by cohort and side. The crowd lean is the share of a market's D and F dollars on its heavier side, over the 2,509 markets with $5,000 or more of D and F money across two or more buys; 1,150 of them had 90% or more on one side (the chart below).

Ninety percent or more means the D and F wallets agreed, under 70% that they split. The side price is what the D and F buyers paid on the side they leaned to.

Taken one buy at a time, the D and F cohort in this window paid 57.3 cents and won 57.7% of the time, +0.40 points. C and ungraded wallets came in at +0.00. S, A and B wallets paid 60.3 cents and won 59.5%, -0.76. Before kickoff, buy by buy, the grade separates nothing, which is what the sharp money study found.

Group the buys by market and a pattern appears. In the 1,150 markets where 90% or more of the D and F money sat on one side, that side was priced at 64.5 cents and won 62.3% of the time: -2.25 points, with a 95% interval from -5.11 to +0.60. Those markets carried $125.7 million of D and F money, from 3.5 D or F wallets each on average.

In the 592 markets where the D and F money was 70 to 90% one way, the heavier side came in at -0.85 points, interval -4.72 to +3.02. In the 767 markets where it split under 70%, the heavier side came in at -0.05. The more of the poorly graded money that landed on one side, the worse that side did against its price, though only the top band is far from zero.

The same split for S, A and B money finds nothing to fade and nothing to follow. In the 1,391 markets where 90% or more of the S, A and B dollars sat on one side, that side was priced at 68.0 cents and won 66.7%: -1.25 points, interval -3.61 to +1.11. The 70 to 90% band came in at -1.74 and the split band at -2.92, both with intervals across zero.

Following them did not pay either. A flat stake on the S, A and B side, held to settlement, lost money in every band: -1.67% where they agreed and -8.12% where they split (the chart below).

This agrees with the timing study: the graded wallets separate from the rest after kickoff, and before it they buy at prices that already reflect what they know. A well-graded consensus before the game says the price is probably right.

In 2,250 markets both cohorts had $5,000 or more before kickoff. In 1,193 of them they were on the same side, and that side, a 67 cent favorite on average, came in at about -2.29 points for both cohorts. When everyone likes the favorite, the favorite is a little too expensive.

In the other 1,057 markets the cohorts took opposite sides. The S, A and B side was priced at 51.7 cents and won 51.8%, +0.10 points. The D and F side was priced at 48.7 cents and won 48.2%, -0.55. Both sides were close to their prices. The September version found the graded side winning here; graded at the trade, it does not.

Split the 1,150 one-sided markets by which side of 50 cents the crowd sat on. In 904 of them the D and F money was on the favorite, at 70.4 cents on average, and the favorite won 68.8% of the time: -1.59 points, interval -4.80 to +1.61.

In the 246 markets where the crowd was on the underdog, at 42.9 cents, the underdog won 38.2%: -4.68 points, interval -10.86 to +1.50. The underdog crowd did worse per market, on a quarter of the sample, and neither interval clears zero.

Soccer supplies 477 of the 1,150 markets, and the crowd side there was a heavy favorite, 74.1 cents on average, which won 71.3% of the time: -2.86 points. Baseball's 323 markets had a crowd side at 51.6 cents that won 49.5%, -2.08. Tennis, 142 markets, +0.48. Esports, 135 markets, -0.22.

In tennis and esports the crowd side came in at its price. The penalty, such as it is, sits in soccer and baseball, the two sports with the most one-sided markets.

By market type the crowd side came in at -1.86 points on 691 moneylines, -2.87 on 351 totals, and -3.03 on 98 spreads. The 10 child moneylines are too few to read.

A crowd that agrees on an over or an under is agreeing on a number. The sharp money study found D and F wallets losing most on totals too, buy by buy.

June, 569 markets, -4.63 points. July, 218 markets, -0.20. August, 259 markets, +0.70. September through the 12th, 104 markets, -0.93. June is half the sample and carries most of the finding; July and August came in at their prices.

The alert floor rose from about $3,000 to $10,000 on July 2, 2026, so August and September record only the largest buys, and the graded population thinned in the same months. A finding that rests on one month is a finding to re-check, not one to bet on.

The $5,000 floor is a choice, so here are the others. At $10,000 or more of D and F money, 1,064 markets, the crowd side came in at -1.79 points. At $25,000, 830 markets, -2.00. At $50,000, 504 markets, -2.17. At $100,000, 247 markets, -4.33, interval -10.01 to +1.36. The penalty is about two points at every floor up to $50,000, and no floor's interval clears zero.

The edge in points is the finding. The return is what a trader collects, so here it is with its assumption stated. In the 1,150 one-sided markets, a flat stake on the crowd's side, held to settlement, returned -3.81%. A flat stake on the other side, bought at one minus the crowd's average price, returned +15.94%. In the 904 favorite markets the fade returned +17.89%. In the 246 underdog markets, +8.76%.

The assumption is the price. The fade return prices the other side at the complement of what the crowd paid. In the 677 one-sided markets where someone bought the other side before kickoff, they paid 39.13 cents on average against a complement of 38.54, about half a cent more, and Polymarket's sports taker fee at that price is about 3% of the stake. A flat return on cheap contracts swings on a few results, so read it as a point estimate that shares the edge's interval, which reaches zero.

The crowd lean is known only after the crowd has bet. A trader fading at kickoff sees the same lean this study computes, but the price at kickoff may have moved from the crowd's average entry. This study scores the fade at the complement of that entry, not at the closing price.

It is pre-kickoff only. The sharp money study found the graded wallets separating from the rest in-play, and a crowd lean that forms during a game is not measured here.

The one-sided markets are a selected set: games where poorly graded wallets bet $5,000 or more before kickoff, which skews to popular games and heavy favorites. The tighter the agreement, the bigger the favorite: the 90% side cost 64.5 cents, against 57.3 for the average D or F buy (the chart below). The result says what happened in those games, not in a random game.

It is one window, weighted to June, with an interval that reaches zero. The September run, whose grades could be written after the trade, put the same figure at -3.68; this one puts it at -2.25. A finding that moves that much with how the grade is read is not one to stake money on alone.

And it scores every position as held to settlement, before fees, at the buyer's own price. It is a test of whether the crowd's agreement carries information, not the audited return of a strategy.

At a sportsbook, fading the public means betting against the side most tickets are on, on the theory that casual money is wrong more often than the price. The book publishes ticket percentages and the bettor guesses who is behind them. It is a guess because the book never says who its customers are.

On Polymarket every buy is a wallet, every wallet has a settled record, and 0xinsider grades that record from S to F. Graded at the trade, the sharp money study finds D and F wallets at -0.18 points per buy and S, A and B wallets at +0.50, both within noise of the price. The question here is what happens when several D and F wallets land on the same side of the same game before kickoff, which is the situation a fader is looking for.

And the public here is small. Behind a 90% D and F side stood 3.5 D or F wallets on average, against 7.8 where they split (the chart below). Part of that is arithmetic: three wallets agree more easily than eight. Fading this public means betting against a handful of traders you can look up, not a stadium of casual money.

The setup is common enough to watch for. In the 20 days from September 21 to October 10, 2026, 220 Polymarket sports markets had 90% or more of their pre-kickoff D and F money on one side, about 11 a day, and 179 had the same for S, A and B (the chart below). This count includes markets that had not settled by October 11, so it measures how often the setup appears, not how it paid.

The S, A and B side is on every game on a 0xinsider sports board, live on Pro and delayed or locked without it: the Profitable wallets strip sets their share of the money on each side against the chance the price implies. This study found that consensus priced about right. The D and F side is on the event page: open a game from the soccer or MLB board, and the Top holders tab lists the wallets on each outcome with their grades.

The large-trade feed lists each large trade with the wallet's grade, live on Pro and 24 hours behind without it, and the betting against the public guide covers the habit from the sportsbook side. Whatever you watch, the finding holds: a one-sided D and F crowd trails its price by about 2 points, which is not enough to bet the other side on that alone.

The page draws on dated, read-only runs against a copy of the production database, all public in the 0xinsider research repository under polymarket-sports-markets. Every figure from the study window comes from the run at 14:18 UTC on October 7, 2026, in the files whose names start grade-at-trade-2026-10-07 (section 4 is this study). The 20 days to the read, September 21 to October 10, 2026, come from the run at 12:37 UTC on October 11, 2026, in fade-crowd-charts-2026-10-11.sql and its output. The smallest recorded buy comes from the read at 14:12 UTC on October 11, 2026, in large-trade-floor-2026-10-11.sql and its output. The September files whose names carry the words fade-crowd are the run this page first published, kept as the record.

The universe is the sharp money study's universe restricted to buys placed before the market's kickoff time on a moneyline, child moneyline, spread or total. The grade is the latest daily ranking dated on or before the trade day and last written at or before the trade. The window starts on June 1 because no earlier ranking row carries a grade, and it cannot be extended backward for the reasons the sharp money study gives.

Each market is one observation, so the interval is the wider of a market-level standard error and one that groups markets of the same game. That second one matters: the 1,150 one-sided markets came from 840 games (the chart below). Expect your counts to grow as open markets settle.