Sharp money in Polymarket sports markets: 161,128 graded bets, checked against the result

Copying a graded wallet's sports bets buys you nothing you can measure, unless the game has started. Across 161,128 large sports buys, each scored against the grade the wallet held when it traded, S, A and B wallets came in +0.50 points against the price, interval -0.58 to +1.59, and D and F wallets -0.18. The 0.69-point gap between them has an interval from -1.11 to +2.48.

In play it is a different market. After kickoff, S, A and B wallets came in at +1.22 points and D and F wallets at -1.23: 2.45 points apart, with an interval from +0.50 to +4.40 that clears zero (the chart below). Before kickoff the order reverses and neither side is distinguishable from the price. Weigh the grade behind a live bet, and ignore it before the game.

On September 12 this page said wallets graded S, A or B beat the price by 1.25 points (95% interval +0.20 to +2.31) and wallets graded D or F lost 1.21, that the gap held in every price bucket, every market type and every month, and that B wallets beat the price by 1.63. Graded as of each trade, none of that holds.

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, and stamps the row with the time of that update. So the row a buy looked up could hold a grade decided after the trade, partly by the trade being scored. Re-run on October 7, 2026, 99.4% of the buys it counted as S, A or B used a row last written after the trade.

The rewrites moved winners up (the chart below). 1,462 buys the old lookup counted as S, A or B came from wallets that were D or F when they traded, and those buys beat the price by 8.80 points. 1,037 buys it counted as D or F came from wallets graded S, A or B at the time, and those lost 12.35.

The old lookup re-run on October 7, 2026 still gives S, A and B +1.27 on 54,499 buys, so the markets that settled since September do not explain the change; the lookup does. Graded at the trade, the same window gives S, A and B +0.50, B -0.80, and D and F -0.18. The table below puts each published figure beside its correction.

Everything after this section is the corrected study. The grade study made the same correction the same day. The grading method is unchanged; only this study's measurement of it changed.

The sample is every buy 0xinsider recorded as a large trade on a Polymarket sports market from June 1 to September 11, 2026, on a market with a yes-or-no result that settled after the trade, at a price between 2 and 98 cents. The smallest recorded buy was about $3,000 until July 2, 2026 and $10,000 after. 161,128 buys, 7,202 wallets, 15,517 settled markets, $2.56 billion in stakes. That is more than the 155,832 the September version counted, because markets in the window kept settling.

Each buy carries the last ranking 0xinsider wrote for the wallet before the trade: the latest daily ranking dated on or before the trade day and last written at or before the trade. A ranking written after the trade is never used. The window starts on June 1 because no earlier ranking row carries a grade. May had 238 rows a day and none graded; June had 18,461 a day, 17,144 of them graded (the chart below).

Wallets are pooled into four cohorts: S, A and B; C; D and F; and no grade at the trade. The edge is the share of buys that won minus the average price paid, in percentage points. The dollar return is stake-weighted, held to settlement, before fees. Each 95% interval is computed three ways, grouping buys by market, by wallet and by game, and the widest of the three is shown.

Wallets graded S, A or B made 51,797 of the buys, across 1,495 wallets and 9,694 settled markets, worth $907.2 million. They won 65.1% of the time at an average price of 64.6 cents: +0.50 points, and -0.37% on the dollar. The 95% interval runs from -0.58 to +1.59 and includes zero.

Wallets graded D or F made 46,857 buys across 3,004 wallets and 8,597 markets, worth $725.6 million. They won 59.3% at an average price of 59.5 cents: -0.18 points, and +2.66% on the dollar. The interval runs from -1.73 to +1.36.

C wallets came in at +0.03 points and wallets with no grade yet at +0.68, both with intervals across zero. Graded at the trade, no cohort is distinguishable from the price, and the 0.69-point gap between S, A and B and D and F has an interval from -1.11 to +2.48.

A longer window does not change the reading. Run from June 1 to October 6, 2026, with buys from September 20, 2026 graded at the moment of the fill, S, A and B came in at +0.64 (interval -0.37 to +1.65) and D and F at -0.23. Restricted to buys of $10,000 or more, S, A and B came in at +0.24 points over 27,448 buys and D and F at -0.38 over 20,278.

S wallets made 27,730 buys from 136 wallets, worth $410.3 million, and came in at +1.20 points, interval -0.41 to +2.80. A wallets, 9,613 buys from 502 wallets, +0.45. B wallets, 14,454 buys from 1,219 wallets, -0.80, interval -2.30 to +0.70. C, 38,841 buys from 2,946 wallets, +0.03. D, 11,849 buys from 1,639 wallets, -1.30. F, 35,008 buys from 1,954 wallets, +0.19.

S is the only letter more than a point above the price, and its interval still reaches below zero. From June 1 to October 6, 2026, it came in at +1.31, interval -0.20 to +2.81. B, published in September at +1.63, lost to the price. B certifies a profitable record, and in sports that record did not carry into the next buy. Below C the letters do not line up: D came in behind the price and F slightly ahead of it.

Dollar returns tell the size story. S wallets returned +1.12% on $410.3 million, because S wallets concentrate stakes in fewer, larger positions where one settled game moves the dollar figure more than the per-buy edge.

The usual objection to a grade result is that well-graded wallets buy favorites and favorites win. Graded at the trade there is no overall gap to explain, but the split by price paid shows where the cohorts differ.

Under 20 cents, S, A and B came in at +5.59 points against +3.80 for D and F, on samples too small to read. At 20 to 40 cents, -0.08 against +0.25. At 40 to 60, +0.32 against +0.05. At 60 to 80, +0.46 against +0.92. At 80 cents and up, +0.63 against -2.94.

The one gap is at the top. D and F wallets paying 80 cents or more for a favorite came in 2.94 points short of the price across 8,987 buys, while S, A and B wallets at the same prices came in at +0.63. Expensive contracts are where a bad read costs the most.

Soccer is 35,522 of the 51,797 S, A and B buys, so the cohort figure is mostly a soccer number. In soccer, S, A and B wallets won 66.3% at 65.3 cents, +0.98 points, while D and F wallets won 60.7% at 60.5 cents, +0.24.

Esports, +0.26 for S, A and B against +0.17 for D and F. Tennis, -0.16 against -1.29. Baseball, where every cohort lost to the price: S, A and B at -1.95, D and F at -1.36, no grade at -2.73.

No cohort's interval clears zero in any of the four sports. The September version found soccer, esports and tennis separating the cohorts; graded at the trade, none does. A grade earned on soccer is not, on this evidence, a sports signal on its own.

Wallets with no grade yet came in at +2.40 in soccer and +2.29 in esports. No grade means no record yet, not a bad record.

By Polymarket's market type, the cohorts sit together on moneylines, the bulk of the buys: -0.05 for S, A and B against +0.62 for D and F. Child moneylines, which cover one map, set, or period, +1.67 against +1.08.

Spreads and totals are where they part. On spreads, S, A and B wallets came in at +1.87 points and +8.63% on the dollar across 5,177 buys, against -1.28 for D and F, on intervals wide enough to include zero for both. On totals, D and F wallets came in at -3.02 points across 7,782 buys, the one cohort figure in this study whose interval clears zero on its own, against -0.08 for S, A and B.

A moneyline is a pick. A spread or a total is a number, and a wallet with a poor record on numbers kept being wrong about the next one. That is a narrower finding than the September version's, and a more useful one: the warning sign is a poorly graded wallet on a total.

Split every buy by whether it landed before or after the market's kickoff time and the cohorts look different before the game and different again during it.

Before kickoff, S, A and B wallets came in at -0.48 points, D and F at +0.37, and no grade at +1.37. Pre-game, in this window, the grade did not separate anyone, and if anything the order ran backward.

In-play, S, A and B wallets came in at +1.22 points and +2.27% on the dollar across 25,973 buys. D and F wallets came in at -1.23 points and -2.29% on the dollar across 17,072. No grade, -0.39. Each cohort's own interval includes zero, but the 2.45 points between them has an interval from +0.50 to +4.40. That gap is the finding of this study that survives the correction, and it is an in-play finding.

The timing study goes further into this: S, A and B wallets place 51% of their sports buys in-play against 38% for D and F, and the separation is widest in the first hour of play, +2.23 points against -1.70, a gap of 3.93 with an interval from +1.06 to +6.80. Whatever the graded wallets know, it shows while the game is being played.

The S, A and B cohort is 1,495 wallets, and one of them, swisstony, made 17,017 of its 51,873 sports buys: 32.8% of the buys and 17.4% of the dollars (the chart below). The 10 wallets with the most buys made 51.0% of them and staked 44.5% of the money. These counts were re-read on October 11, after more markets settled, so the cohort comes to 51,873 buys here against 51,797 in the study.

That cuts two ways. The cohort figure is, in large part, the record of one wallet that bets sports at size and all day, so a study of 1,495 wallets is partly a study of one. And one bad month for that wallet moves the cohort more than the hundreds of small ones can. Its own record is in our swisstony analysis.

Most S, A and B wallets made a handful of sports buys in the window. Judge a wallet by its own profile, not by the cohort it sits in.

S, A and B came in ahead of D and F in three of the four months, by less than a point each time, and behind in August. In June, S, A and B came in at +0.47 and D and F at -0.50. July, +0.89 against +0.34. August, -0.66 against +1.03. September, through the 11th, +0.81 against +0.39.

The later months are small. The alert floor rose from about $3,000 to $10,000 on July 2, 2026, so August and September record only the largest bets, and the wallets carrying a grade thinned from 17,144 a day in June to 3,300 in August. September is 1,846 S, A and B buys and 1,164 D and F buys.

The edge is measured at each buy's own price against the settled result. It ignores fees and slippage, treats every position as held to settlement, and scores a wallet that exited early on a result it never waited for. It is a test of whether the grade carries information, not of whether copying a graded wallet's trade is profitable after you have paid to get in.

Grades are computed from a wallet's own earlier settled trades, and each buy here uses a grade written before it. The same wallets whose history built the grade are the ones being tested, so this is not a held-out universe, and a grade written weeks before a buy can be stale by the time of it.

Two intervals on this page clear zero: the in-play gap between S, A and B and D and F, and the D and F loss on totals (the chart below). No cohort's overall figure does. Everything else on this page is indistinguishable from the market at the 95% level.

And the in-play gap may be speed. A wallet that is right about a live price two seconds before everyone else is right for a reason the grade cannot see. Whether the graded wallets are reading the game better or reading the feed faster is a question this data cannot answer.

A fair market can still have winners. The calibration study found that large buyers on Polymarket sports markets paid, on average, what the result was worth. An average like that hides who is on each side of it: if some wallets are consistently right and others consistently wrong, they cancel out in the average.

0xinsider grades Polymarket wallets from S to F on their own settled record. The grade study tests that grade against every large sports and esports buy on the platform. Its run of October 23, 2026 grades each buy as of the trade and finds S wallets 1.78 points ahead of the price and A wallets 0.64 ahead across 81,855 buys of $10,000 or more, and B wallets 0.92 behind. Soccer is more than half of the S and A buys there.

This study uses the same grade-at-the-trade lookup on sports markets. It keeps every large buy 0xinsider recorded, not only those of $10,000 or more, so that tennis, esports, and baseball have samples worth reading, and adds two splits the grade study does not make: by market type, and by whether the buy landed before or after kickoff.

In sports alone the order holds, S +1.20, A +0.45 and B -0.80, but every one of those intervals reaches zero (the chart below).

A large trade on the feed is not sharp money by default. From September 21 to October 10, 2026, the large-trade feed recorded 7,472 large sports trades, and 3,078 of them came from wallets graded F when the order filled. S, A and B together made 3,141 (the chart below). Read the grade before you read the size.

Every grade has its record behind it. The leaderboard ranks wallets by that record, and each profile shows the calibration edge, the same win-rate-minus-price measure this study uses. Each sports board (soccer, tennis, NBA, and the rest) shows which side the profitable wallets, this study's S, A and B cohort, hold on each game, live on Pro and delayed or locked without it.

Pick of the Day publishes up to 25 picks a day, each the side profitable wallets hold, and every settled pick stays on the public record. The feed runs live on Pro and 24 hours behind without it, in-play trades included, and in play is the one place this study found the grade to matter.

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 grade-at-the-trade figure comes from the run at 14:18 UTC on October 7, 2026: the query text, raw output and per-cut statistics are the files whose names start grade-at-trade-2026-10-07. The daily ranking coverage and the large-trade feed by grade come from the run at 12:37 to 12:55 UTC on October 11, 2026, in sharp-money-charts-2026-10-11.sql and its output, with the per-wallet concentration file and the script that adds it up. 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 timing split is the timing study's run at 12:26 UTC on October 10, 2026, and the grade study's figures are its October 23, 2026 run, in the repository's grade-vs-price folder. The September files, whose names carry the words sharp-money, are the run this page first published, kept as the record.

The universe is every row of the large-trade alert table for a Polymarket buy on a sports-category market from June 1, 2026, joined to the settled outcome of its market, with a yes-or-no result, the resolution timestamp after the trade and the price between 2 and 98 cents.

The grade is the latest daily ranking dated on or before the trade day and last written at or before the trade; from September 20, 2026, the grade the wallet showed when the order filled. The grading method is documented on the transparency page and explained in the trader grades guide.

The window cannot be extended backward. Before April 2, 2026 the alert table's outcome index was a defaulted zero for a large share of buys, a defect the calibration study describes, and before June 1 no ranking row carries a grade.

Expect your counts to grow. Every buy sits on a settled market, and markets keep settling. Run from June 1 to October 6, 2026, the same query moved no grade's edge by more than 0.34 points (the chart below).