Is smart money real on prediction markets? What a track record can and cannot tell you

Every prediction-market price is a probability. A Yes share at 40 cents says the market puts the outcome at roughly 40 percent. A trader who buys at 40 cents and is right 40 percent of the time has no edge; they are the market. A trader who buys at 40 cents and is right 50 percent of the time is beating the market by ten points.

So the smart-money question has a clean form. Take a group of traders defined before you look at results. Record every entry and its price. Wait for resolution. Compare the group's hit rate to the average price it paid. The difference is the edge. If it is positive over a large sample, smart money is real for that group. If it is not, it is not.

A raw hit rate misleads in both directions. A group that only buys heavy favorites at 90 cents can post a high hit rate and still lose money, because one miss erases nine wins. A group that buys underdogs at 30 cents can post a low hit rate and be sharply profitable.

Edge against the entry price handles both. It also exposes the cheap-longshot trap: at very low prices, a small edge implies a huge multiple on paper, and a handful of lucky resolutions can make a bad method look brilliant for a while. Any honest record shows results by price bucket for exactly this reason.

A healthy overall edge can contain a segment that bleeds. The most common shape: one confidence tier or one price band carries the losses while the rest of the record covers for it. Averages are where those segments go to hide.

The remedy is to cut the record along the dimensions the method actually uses, and to cut them together rather than one at a time. A tier that looks fine on its own can be losing inside one price band. When a method changes, the record should say when, and results before and after should be readable separately.

• Sample size first. Small samples produce large edges by accident in both directions.
• Price buckets. Longshots and favorites should be judged apart.
• Segments the method relies on, combined, not just one at a time.
• A dated methodology line, so a change is visible instead of blended in.

Survivorship: pick the traders after you know who won and the record will look superb. The group has to be chosen by a rule, in advance, and the rule has to be applied to everyone it selects, including the ones who go cold.

Selective memory: a screenshot of a winning position is not a record. A record includes the losers, the markets that expired without resolution, and the entries that were never filled at the price shown.

Hindsight timing: a position observed today at a good price may have been opened at a worse one. Entry price has to be captured when the position is first seen, not reconstructed later.

Aligned defines its tracked traders by a rule, records every alignment it surfaces with the market price at the time, and scores each one when the market resolves. The track record page shows the settled sample, the hit rate against the average entry price, the results by price bucket and confidence tier, and the date the methodology last changed.

It does not claim that any individual market will resolve a particular way, and it does not present the record as a forecast of the next one. It presents the record as a record. Read it the way you would read anyone's: sample size first, then the edge, then the segments.