How to find people who actually beat the market

Judge a forecaster on four numbers and ignore everything else. How often they beat the market price at the moment they spoke. By how much. How early, meaning whether the price later moved toward them. And over how many resolved events. The fourth number decides whether the first three mean anything, and almost every public ranking leaves it out.

What are the four numbers that matter?

A forecaster is only measurable against a benchmark, and on real events the best available benchmark is the live market price, because it already contains what everyone else believes. Four numbers follow from that.

Beats market is the share of a person's resolved calls that scored better than the market price frozen at the moment they made the call. Edge is how much better, averaged across every resolved call. Early is whether the price moved toward the call afterwards. Events is how many of their calls have resolved.

The first three are performance. The fourth is evidence. A person with 90% Beats market over 6 events has told you nothing. A person with 62% over 40 events has told you something specific.

Why measure against the price instead of right and wrong?

Because a raw hit rate rewards easy calls. Predicting that a heavy favorite wins is correct most of the time and worth nothing, since the price already said so. Scoring against the price removes that free credit automatically: your call is compared to what the market said at the same instant, so you get paid for disagreement that turns out to be right and charged for disagreement that does not.

This is the same design that formal forecasting tournaments use. The IARPA tournaments run from 2011 to 2014 scored every entrant with the Brier score, the squared distance between the stated probability and the outcome, and the winning program selected 60 superforecasters on measured accuracy alone and saw them stay ahead of comparison groups for two more years rather than regressing to the mean. Skill in forecasting is real and it is measurable, but only against a benchmark and only over a run of questions.

Without a benchmark, the numbers collapse. CXO Advisory Group graded 6,582 public US stock market forecasts from 68 named experts collected between 2005 and 2012 and reported 46.9% terminal accuracy. Those experts were not stupid. They were being scored on everything they said, which almost nobody is.

What does Beats market tell you?

Beats market is the honest hit rate. It answers: out of everything this person called that has now resolved, how often did the call land closer to the truth than the market price at that moment.

Two readings are useful. A very high Beats market with a small Edge usually means someone who is right often on calls where the price was already close to correct. That is a careful person, not necessarily an informed one. A moderate Beats market with a large Edge means someone who is wrong plenty but right when the price was badly wrong, which is the profile that actually moves the needle on hard events.

What Beats market cannot be raised by is money, because no money is involved. On insiderz a call is a statement, not a stake, so the column is unaffected by how much anyone is willing to risk.

What does Edge tell you?

Edge is the average size of the advantage over the market price, across all resolved calls. It is an average over the whole record, not a best of, so a single dramatic call cannot carry it.

That matters because a single dramatic call is exactly what dominates money based rankings. Polymarket's public leaderboard endpoint, as of September 2026, accepts only PNL and VOL as ordering criteria and returns no accuracy field. One large winning position can put a trader at the top of it permanently. An average across resolved calls cannot be bought that way, which is the whole reason to use one. The longer version of this argument is in P&L is not skill.

What does Early tell you?

Early records whether the market price moved toward a call after the call was locked and before it went public. It is a claim about sequence: you said it, then the price came to you.

This is the hardest of the four numbers to manufacture, because producing it requires other people to change their minds later, for their own reasons. A person can be lucky once. A person cannot repeatedly be followed by the market unless they are getting to information before it is priced.

Early only exists because calls go public after a delay. During that window the call is locked, timestamped and unchangeable, but not yet visible to the market, so any price movement in that window cannot have been caused by the call itself. Followers with live access see the call immediately; everyone else sees it when the delay expires.

How many resolved events are enough?

About a dozen is where a good run stops being explainable by chance, and thirty is where the shape of a real edge shows. That is why people under 30 resolved events are not ranked on the leaderboard at all.

The arithmetic is simple. Treat a well priced market as a coin flip for someone with no information: they should beat the price about half the time. The table below shows how often a pure guesser would produce each result, computed with the binomial distribution on 4 September 2026.

Resolved calls	Calls that beat the market	Chance a guesser does at least this well	Ranked on insiderz
5	5	1 in 32	No
10	10	1 in 1,024	No
12	9	1 in 14	No
12	10	1 in 52	No
12	12	1 in 4,096	No
30	18	1 in 6	Yes
30	20	1 in 20	Yes

Read the third column against the size of the crowd. If ten thousand people are making calls, roughly three hundred of them will hit 5 out of 5 with no ability at all, and every one of those runs looks like a discovery. At a dozen resolved calls the noise has not disappeared, but the numbers become interpretable, and by 30, the point where the ranking starts, the shape of a real edge is visible.

The same logic explains why a 60% Beats market over 50 events beats an 80% Beats market over 10. Rate without sample is not a measurement.

How do you read a profile?

Open a person's profile and read it in this order.

1. Events first. If the number is small, stop. Nothing below it is interpretable yet.
2. Beats market next. This is the headline, but only in the context of the first number.
3. Edge after that. It tells you whether they are right on easy calls or on mispriced ones.
4. Early last. Consistent Early is the strongest single indicator, and the rarest.
5. Then the calls themselves. Look at what the market price was when each call was made. A record full of calls made at 90% is a record of agreeing quickly.
6. Check the misses. They are all there, because calls cannot be edited or deleted. A profile with no losing calls at all, at any sample size, means the sample is small.

Here are the top insiders on the leaderboard right now.

What can you do once you have found someone?

Follow them, and ask for live access. Calls go public after a delay by default. A follower with live access sees an insider's calls the moment they are locked, which is the difference between reading a record and watching one being made.

There is nothing to copy in the trading sense, because there is nothing staked anywhere on insiderz. What you get is the statement, the time it was locked, and the market price it was locked against. What you do with it is your business, and none of it is advice.

One honest limit. insiderz launched in September 2026, so every record on the site is young, and a leaderboard built on a few months of resolved events should be read as a beginning. The measurement design is the point: the same four numbers become more informative every month, because the denominator only grows. Public, scored, complete records are also what makes independent evaluation possible at all, which is why a 2025 study could compare frontier language models against top human forecasters across 464 resolved questions rather than across anyone's selected highlights. For the underlying idea of being right the right amount of the time, see calibration explained, and for what a record has to contain before any of this works, see how to build a prediction track record people can verify.