We are a small shop that evaluates players across five sports, and this is where we write down how we do it.

The starting observation is simple. A published expectation for a player — how much of some particular thing they are likely to do on a given night — is the most granular public statement anybody makes about that player. Collect enough of those statements, compare them to what actually happened, and you learn something about players, something about the people setting expectations, and a great deal about your own judgment.

The survivors of our process are called gems. A gem is a rating. It is not a recommendation, it is not addressed to anybody, and it is not for sale.

There are no picks here, and there never will be

This is the part worth being unambiguous about. We do not publish picks. We do not attach a player to a direction. There is no slate, no shortlist, no board, and no results table anybody could follow.

We also do not wager on anything. That is not a legal disclaimer, it is a description of the shop — the interest here is in whether a method can be made to work and whether we can tell honestly when it does not. Those are questions you can answer without ever acting on a single rating, and answering them is the whole activity.

Nothing on this site is betting, investment, financial or legal advice. The players, teams and competitions we describe are invented, and every example exists to illustrate a mechanism rather than to describe a real market.

What we actually write about

The method, in the order we use it. Build a corpus of prior performance, because nothing can be evaluated without one. Put every candidate through gates that disqualify it — most die here, and the gates are where the real work is. Read the published expectation as evidence about what somebody believes, never as an instruction. Cap and shrink, so no single night dominates and confident estimates get pulled back toward the ordinary. Then grade every rating inside a fixed window and check our calibration — whether the confidence we stated matches the accuracy we observed.

That last step is the one that makes this worth reading. Accuracy is the number people ask for. Calibration is the number that tells you whether a shop is any good, and we spent two years reporting the first one honestly while it functioned as a kind of self-deception.

Why we publish the misses

We publish how often we are wrong, and there is a whole desk of write-ups of specific failures — a gate that leaked for eight days, a filter that silently rejected everything for a month, a bug we believed was a discovery for six weeks, five months of grading ourselves generously without anybody cheating.

Nothing gets deleted. Ratings produced under broken conditions stay on the record and get graded however they turned out, because a record with deletions in it is not a record of the method — it is a record of the method plus our own later judgement about which parts to count.

About the writing

PlayerGem is fiction. The shop, its colleagues, the seasons and the failures are invented, and the narrator is deliberately not named — the method is the subject and a publication that grades its own errors does not need a personality attached to it.