I’m building 0xinsider because I love data, and Polymarket has some of the best public data I’ve worked with. Every trade is tied to a wallet. Every fill settles on chain. Every market ends in a hard yes or no.

That’s millions of decisions with money behind them and a known answer at the end. I don’t know another dataset like it.

0xinsider started in February as a live feed of large trades on Polymarket and Kalshi. It’s now analytics for Polymarket sports and esports markets, because sports markets settle every day, and a market that settles is a data point with an answer. Kalshi is gone. Its trades are anonymous, so there’s no wallet to follow.

Here’s the dataset today:

None of it arrives clean. Raw fills aren’t positions. You have to match entries to exits, handle splits, merges and redemptions, and mark every open position to market every day.

I got that wrong for a while. My first P&L rollup added up per-market rows, and on some wallets it overstated the total by 7 to 10 times, sometimes with the wrong sign. Realized P&L now comes from Polymarket’s own position accounting, and I wrote up why two tools can disagree on the same wallet.

The first thing I wanted to know was whether the prices are even wrong. Mostly, they aren’t.

Across 411,770 large sports buys worth $8.2 billion, people paid 60.6¢ on average and their side won 60.7% of the time. That’s a miss of two hundredths of a point. The one range that clearly loses is under 10¢, where long shots paid 6.5¢ and won 3% of the time.

Sports turned out to be the well-priced part of Polymarket. Large buys outside sports miss their price by 6 to 10 points in most ranges.

So if the prices are that good, who’s making money? Not most people.

Of the 36,149 wallets with 20 or more settled sports markets, 48.8% are in profit and the median wallet is down $8. The top 1%, 362 wallets, hold 68% of all the profit. The bottom 10% lost $401 million between them.

That gap is what 0xinsider’s grades try to measure. Every tracked wallet gets a grade from S to F, rebuilt every day from its settled P&L, and realized profit is about 95% of it.

A grade like that is easy to compute and easy to fool yourself with. Win rate is the classic trap. A wallet that only buys at 85¢ wins about 85% of the time and has shown no skill at all. The number that holds up is the gap between how often your side won and the price you paid for it.

So I tested it on 67,531 buys of $10,000 or more, using the grade each wallet already held on the day it traded, before the market settled.

S, A and B wallets beat their price by 1.57 points. D and F wallets missed it by 2.06. C wallets and ungraded ones were indistinguishable from the market. The intervals are bootstrapped over markets rather than trades, because four wallets buying the same side of the same game are one observation.

1.57 points is small, and it should be. On a market this well priced, a study that found 10 points would be describing a bug in its own method.

My favorite detail is that the sample keeps growing. Every trade in it sits on a market that already settled, so the universe gets bigger whenever an open market resolves. Three runs on the day I published returned 67,516, then 67,517, then 67,531 trades. The edge came out at 1.57 points all three times.

Every query, raw output and bootstrap is public. This one rebuilds the intervals above from a committed CSV, with no database and nothing to install past Python:

1
git clone https://github.com/0xinsider/research && cd research/grade-vs-price && python3 bootstrap.py

About 20 seconds later:

1
cohort      trades  markets  edge_pts   95% CI
2
S/A/B        24610     5946      1.57   [+0.32, +2.82]
3
C             9223     1986      0.53   [-1.93, +2.81]
4
D/F          17106     4476     -2.06   [-3.61, -0.45]
5
no grade     16592     5250     -0.13   [-1.81, +1.57]

On 0xinsider, all of this becomes a board. Every sports and esports game shows the moneyline on each side next to how much money from profitable wallets is sitting on it, live, across the NFL, NBA, MLB, NHL, soccer, tennis, UFC, CS2, LoL, Dota 2 and Valorant. Every wallet has a profile with its grade, P&L and the rest of its metrics. The boards, grades and profiles are free.

If you’d rather pull the data yourself, there’s a read-only REST API, an MCP server, and SDKs for Python and Go. The research library has the rest of the studies, and the methodology page says what every number is made from.

None of this is financial advice. Every number here describes what already happened.