Do wallet grades predict Polymarket outcomes? 81,855 large sports and esports buys, graded as of each trade

On September 12 this page said wallets graded S, A or B beat the price they paid by 1.57 points (95% interval +0.32 to +2.82) and D and F wallets lost to it by 2.06. Both numbers were wrong, and so was the claim that B wallets beat the price.

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 trades being scored. Re-run on October 7, 2026, 99.3% of the S, A and B buys it scored used a row last written after the trade.

The rewrites moved winners up. 1,190 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 8.89 points. 852 buys it counted as D or F came from wallets that were S, A or B at the time, and those lost 15.06. The old query also scored 318 buys on void or unresolved markets as losses.

On the same window, with each buy graded by the last ranking written before it, S, A and B together came in at +0.34 ± 1.07 points: no measurable edge. S came in at +1.54 ± 1.49, A at +0.54 ± 1.79 and B at -1.09 ± 1.62. D and F, published at -2.06, came in at -0.37 ± 1.34. The table below puts each published figure beside its correction.

The September study and this correction covered every Polymarket market. Everything after this section is the corrected study, re-run on October 23, 2026 over a longer window and on sports and esports buys only. Its queries take the grade at the trade, so a re-run cannot repeat the error.

A wallet grade is easy to compute and easy to fool yourself with. 0xinsider grades every tracked Polymarket wallet from S down to F on its own settled history, with guardrails so one lucky market cannot produce an S. That only means something if the grade says something about the trades a wallet has not made yet.

The test: take every buy of $10,000 or more on a Polymarket sports and esports market from June 1 to October 22, 2026, on a market that settled after the trade. Give each buy the grade its wallet held at that moment, before anyone knew the result. Then check the settled outcome. That is 81,855 buys from 3,834 wallets across 15,779 settled markets, $2.51 billion in stakes.

Since October 22, 2026, a grade ranks a wallet's settled sports and esports record only, which is why this study covers those markets only. The grade each buy carries is the one its wallet showed at the time, so nearly every buy here was graded on the wallet's record across every market. Only 197 buys in the sample came after the change, 77 of them from S, A or B wallets: too few to read on their own.

S wallets beat the price they paid. A wallets did too, by less and with less certainty. B wallets did not. B certifies a profitable record: positive realized P&L across at least 10 resolved markets. This study finds no price edge behind it.

Size is no guide either. Wallets graded C, D or F, or not graded at all, put up most of the money in these buys, and F wallets alone staked more than S and A together. A big buy tells you someone was confident. The grade at the moment of the trade is what tells you whether that kind of confidence has paid before.

A wallet that only buys contracts at 85¢ wins about 85% of the time and has demonstrated no skill whatsoever. Win rate measures which prices someone likes. Report it alone and every favorite-buyer on the platform looks like a genius.

The measure that survives is the gap between how often the chosen side won and the price paid for it. S wallets paid 66.5¢ on average and won 68.2% of the time: 1.8 points better than the market that sold to them. B wallets paid 60.0¢ and won 59.1%: 0.9 points worse. That gap is what a grade has to predict, and it is the number this study reports.

Calibration edge on trader profiles uses the same logic. Paying less than the eventual frequency is the skill. Backing the obvious side is not.

S wallets made 15,195 of the buys, from 166 wallets. They won 68.24% of the time at an average price of 66.46¢, an edge of +1.78 points with a 95% interval of +0.54 to +3.02. It clears zero.

A wallets made 7,353 buys from 502 wallets: +0.64 points, interval -0.88 to +2.16. Pooled with S, the two come to +1.41 points across 22,548 buys, interval +0.37 to +2.45.

B wallets made 12,295 buys from 1,070 wallets: -0.92 points, interval -2.39 to +0.55. Adding B to S and A pulls the group to +0.59, interval -0.32 to +1.49, which no longer clears zero.

Below A the letters do not line up. D lost 3.87 points on 3,424 buys, interval -6.91 to -0.82. F came in at -0.10 and C at +0.12, each with an interval across zero. Wallets with no grade yet came in at +2.70, interval +0.22 to +5.18: a wallet too new to grade is not a wallet without skill. The grade picks out S and, less firmly, A. It does not rank the rest.

Each interval is computed twice, once grouping buys by market and once by wallet, and the wider of the two is shown. Buys on the same market, or from the same wallet, are not independent observations, and treating them as independent is the standard way to manufacture a result that is not there.

The headline joins two sources. From September 20, 2026, 0xinsider records every grade change as it happens, so a buy takes the letter its wallet showed when the order filled. Before that, a buy takes the last ranking row written before the trade. Each source alone, and a third design, point the same way.

Graded at the fill, September 20 to October 22: S +3.02 (+1.15 to +4.88) on 2,902 buys, A +1.74 (-0.99 to +4.48) on 1,692, B -0.09 (-4.48 to +4.29) on 1,326. This is the cleanest attribution and the smallest sample.

Graded at the start of July, August and September and scored on the next 30 days: S +2.59 (+0.98 to +4.19), A +2.00 (-0.12 to +4.11), B -0.48 (-2.33 to +1.37), across 45,222 buys. The ranking rows alone, June 1 to September 19, give S +1.50 (+0.03 to +2.98), A +0.24 and B -1.02.

S is above zero in every design and its interval clears zero in each. A is above zero in every design, but its interval reaches zero in each. B is below zero in every design.

The obvious objection is that good wallets buy favorites, favorites win, and the grade measures nothing beyond a taste for high prices. Splitting the same buys by the price paid answers it.

S and A wallets came in above the price in all five bands, from long shots under 20¢ to favorites at 80¢ and up, and B wallets below it in every band from 20¢ up. If the grade only rewarded a taste for favorites, the cheaper bands would sit at or below zero. They do not, though they are too noisy to prove much the other way.

No band clears zero for S and A on its own. The narrowest, 80¢ and up, comes to +1.28 ± 1.40 over 6,572 buys. The two bands under 40¢ are too noisy to read, at ± 10.77 and ± 3.99.

Soccer is 53.1% of the S and A buys, so the headline is mostly a soccer number, and soccer's own interval, +1.50 ± 1.66, crosses zero. Esports clears zero at +3.44 ± 2.04. Every other category with 500 or more S and A buys sits above zero with an interval that crosses it. Read the categories as a map of where to look next, not as 6 findings.

The edge is measured at each trade's own price against the settled outcome. It ignores fees and slippage, and it treats every position as held to resolution. A wallet that bought at 60¢ and sold at 75¢ before the market resolved is scored on the settlement it never waited for, which is not what it made.

Grades are computed from a wallet's own earlier settled trades, so the wallets whose history built a grade are the ones being tested. This is point-in-time, not a held-out universe. The ranking-row half has a gap too: a wallet whose only ranking row was rewritten after its trades has no row written before them, so its buys count as ungraded.

Only settled markets are in the sample, which tilts it toward short-dated games and leaves out season-long futures. The design that reads the grade at the fill is the cleanest and covers only September 20 to October 22, so its interval on S is wider than the headline's: ± 1.87 points against ± 1.24.

The grade has changed since most of these buys. From October 22, 2026 it ranks the sports and esports record only, and the study cannot yet test that version: a grade read today was computed partly from the trades being scored, which is the look-ahead the correction above removed. Re-runs will add buys graded on the new basis as they settle.

And 1.78 points is a small edge, the size to expect from a market that mostly works. None of this says an S-graded trade is worth copying after fees and your own entry price. It says the S grade, and more weakly the A grade, carries information about who tends to be right. B carries a profitable record and, on this evidence, no measured price edge.

The sample is every Polymarket buy of $10,000 or more from June 1 to October 22, 2026, at a price between 2¢ and 98¢, on a sports and esports market with a yes-or-no result recorded after the trade. A market counts as sports and esports by the rule the grade uses: Polymarket ties it to a game, gives it a sports market type, or files it under a sports or esports category. Void and unresolved markets are out. Buys only, because a buy at a price on a named outcome is a clean directional bet and a sell is not.

Grade assignment: from September 20, 2026, the letter 0xinsider's recorded grade history proves the wallet showed when the order filled. Before that, the latest ranking row dated on or before the trade day and last written at or before the trade. The window starts on June 1 because grade coverage widened that month, from about 238 wallets scored a day in May to 18,464 in June.

Every number on this page comes from read-only queries against a production replica on October 23, 2026, the headline export at 13:43 UTC. The query text, the output and the sums the intervals are built from are public at github.com/0xinsider/research, under grade-vs-price, beside the all-market run of October 7, 2026 and the September 12 files it corrected. The grading method is documented on the transparency page, and the per-wallet inputs are visible on every trader profile.

Expect your counts to come out higher than ours. Every buy sits on a market that has already settled, so the sample grows as open markets resolve, and the half graded at the fill grows every day.