Method

Every night somebody says something remarkable about the Mets. A graphic during the broadcast, a line in the game notes, a sentence in the recap. Most of those claims are true. That is not the interesting part.

The interesting part is how much work the sentence had to do to become remarkable. Turn enough dials and anything looks rare. First Met since 1987. In September. On the road. With runners on. Each condition is defensible by itself, and stacked together they carve a slice of history so thin that almost anything inside it is a first. The claim ends up describing the search that produced it rather than the thing that happened.

So I check them, and I publish what the counting says, including the nights when the claim turns out to be exactly as good as it sounded.

Every check has the same three blocks

They said. The claim word for word, with the outlet and the date. Where the wording reached me through syndicated copy rather than the original graphic, the piece says so.

The arithmetic. What I counted, the population I counted it over, and what came back.

What’s actually true. The plain version, whatever it turns out to be.

The five verdicts

Holds up. Accurate, and as rare as it sounds.

Undersold. Accurate, and smaller than what actually happened.

True, but hollow. Accurate, with the rarity coming from the conditions stacked into the sentence.

Overstated. The parts are right and the headline claims more than they support.

Wrong. It does not survive the count.

A verdict is about the sentence. It is never about the player.

Where the numbers come from

Retrosheet, a volunteer project that has spent forty years reconstructing the play-by-play of professional baseball from scorecards and box scores. Game records reach back to 1897 and play-by-play to 1908. For the Mets I use every game since 1962.

Regular season only, unless a piece says otherwise.

Anything that did not come out of those files is named where it appears, with its source and its date, so you can go and look.

What the record cannot see

No method page is worth much without this part.

There is no play-by-play at all before 1908. Between 5 and 40 percent of games per season from 1908 to 1956 contain plays that could not be decoded, so counts of rare events inside that window are floors rather than exact figures.

Pitch sequences exist for about 44 percent of half-innings, nearly all of it 1988 and later. Anything counted in pitches comes from that window, and the piece says which.

The current season is the biggest gap. Retrosheet publishes a season the following year, so nothing that happened this year has been verified here. When a piece deals with a current claim, it establishes what the record held before it and says plainly that this season is unverified.

Every piece carries its own version of this, sized to what it actually leaned on.

Every query starts by proving itself

Before I trust a number, the query that produced it has to reproduce a figure somebody else has already published. A season total, a career line, a record everybody agrees on. If it cannot reproduce the known number, it does not get to report the unknown one.

That practice has caught three errors in my own work so far, which is the point of having it.

About the numbers on the page

The design has a place for a single score from zero to a hundred beside each finding. There is no honest way to compute one yet, so pieces carry plain frequencies instead, like 1 in 12, or 0 of 39, with the count that produced them sitting next to the claim. When that changes, this page will say so.

I do not publish the query catalogue, the thresholds or the code. What each piece does publish is its source, its scope, the population counted, and the limits of the record it was counted against.

Corrections

When something here is wrong, the correction is published and dated on the piece itself. Nothing gets quietly edited.