A published expectation for a player is the most granular public statement anybody makes about what that player is likely to do. It is not advice, it is not a prediction we are meant to act on, and it is not addressed to us at all.

It is somebody's belief, written down, with a timestamp. That is the whole of what it is, and treating it as exactly that — no more, no less — is the single most useful habit we have.

We Spent a Year Asking the Wrong Question About Every Number

Our early framing was comparative: is this expectation higher or lower than what our corpus suggests. That question is easy to compute and almost useless, because it treats the published number as a competing estimate to be beaten.

It is not a competing estimate. It is a summary of what a well-informed party believes, produced by people who see things we do not — availability, condition, internal information about roles — and who are wrong in different ways than we are. Framing it as an opponent throws away the most interesting thing about it.

The symptom of the wrong framing was that our largest disagreements clustered in exactly the places where we had the least information. We were most confident precisely where the published expectation had incorporated something our corpus could not see. A disagreement was being read as an opportunity when it was usually a warning.

Worse, it made us reluctant to update. If the number is a rival, then agreeing with it feels like conceding. We had months of quiet resistance to the obvious, dressed up as independence of thought.

Reading the Number as Somebody Else's Posterior

The reframe is simple to state. The published expectation is somebody's conclusion after seeing their evidence. Our estimate is our conclusion after seeing ours. When they differ, exactly one interesting question follows: what would have to be true for their conclusion to be the right one?

That question is answerable surprisingly often. Sometimes the answer is that they know about a role change we had not detected, and we can go and check. Sometimes the answer is that they are pricing in a condition we deliberately exclude. And sometimes there is genuinely no story — nothing we can construct would justify their number — and that is the only case where the disagreement is informative about them rather than about us.

That last category is small. Much smaller than our early framing implied. Most disagreements dissolve the moment you seriously attempt to construct the other side's reasoning.

“Every time you can't explain their number, write down what you'd need to know,” Marcus said. “Half the time the thing you'd need to know is already in the corpus and you didn't look.”

The Habit This Cost Us, and It Was a Comfortable One

What we gave up was the pleasure of finding disagreements. There is a real satisfaction in a number that looks wrong to you, and the reframe removes most of it, because most of them turn out to be explicable and therefore uninteresting.

It also made the work slower. Constructing somebody else's reasoning in good faith takes time, and there is no version of it that can be automated — we tried, and what we built was a set of heuristics that reliably produced a plausible story for every disagreement, which is worse than useless. A machine that can always explain the other side is not doing the thing we wanted.

And I should admit the reframe was not a discovery so much as a concession. We adopted it after a stretch where our largest disagreements graded worst — not badly, worst, consistently, over a full season. The reasoning came afterwards to explain a result we could not argue with.

What Survived Into the Doctrine

Two things. First: a disagreement is a question, never a finding. It goes into a queue to be understood rather than a list to be used.

Second, and this is the one that took discipline: the size of a disagreement is not evidence of its quality. Our instinct was that a large gap between our estimate and a published expectation was a stronger signal than a small one. The opposite has been closer to true — the largest gaps are where somebody knows something, and the modest ones are where two parties looking at similar evidence have arrived at slightly different places, which is the only situation in which our own estimate has much claim to being useful.

We do not rate anything on the strength of a disagreement alone. That rule is old, it has never been relaxed, and it is the reason a whole category of exciting-looking candidates never reaches the output.

The queue of disagreements we could not explain is currently four items long. It has never been longer than about nine. I find that number reassuring and slightly humbling in roughly equal measure.

Note: PlayerGem is a fictional analytics shop and these accounts are invented. Nothing here is a pick, a recommendation, or betting or investment advice, and the players, teams and competitions described do not exist.