What "graded in public" means: how we score every call and publish the misses

The promise, in five rules

Every fantasy site says it grades itself. Most of them mean that at the end of the year someone writes a column about the calls that worked. That is not grading. Grading is a set of rules fixed before the call is made, applied by a machine that does not know which answer would look better. Ours are these.

A call counts the moment it is published. Nothing is added to the record after the fact, and nothing is removed.

A call locks at the earlier of the two players' kickoffs, exactly as it stood. It is never edited afterwards, not for an injury, not for a late scratch, not for a better argument.

A call is right when the player we said to start outscored the other one in your scoring format. A tie counts as neither.

Every call is graded, and the misses stay up on the same page as the hits, in the same type.

The hit rate reads "nothing graded yet" rather than zero percent until a real call has been graded, because a percentage of nothing is not a number and printing one would be the first lie in the record.

Where the fantasy record stands today

On September 3 the record page shows 577 start or sit calls published and none graded. That is what it should show. Grading runs each morning off final box scores, and the first game is Wednesday, September 9. The first results land the day after Week 1's last game.

Publishing 577 calls before a single one can be scored is the point. If we only started counting after the games, we could pick which calls to count. Because every one of the 577 is already in the ledger with a timestamp, the Week 1 hit rate will be whatever the box scores say it is, and it will include the calls made when no player market existed and the basis was ADP and news alone. Those are labelled on the page now and they will be labelled in the record.

What a miss looks like

A miss is not deleted, softened or explained away. The graded call shows what we said, which player scored what, and the word wrong. The reason line we published beforehand stays beside it, so a reader can see whether we missed because the market was wrong, because the player was inactive, or because the call was a coin flip we labelled as one.

That last case matters. The scoring model says out loud when two players are inside 1.5 projected points of each other, that the call is "inside the noise of a single catch." A miss on a call we described that way is a different thing from a miss on a call we said was not close, and the record keeps both.

The record this comes from

The fantasy record is new. The grading discipline is not. The same rules have graded the earlier market record on the site since June 21, 2026, and that record is the best evidence of what public grading looks like after a few thousand entries.

As of September 3 the track record page says that 93.7 percent of finished games have been graded and the rest are waiting on an official score; nothing is quietly dropped. It breaks the record out by category, and the breakdown is not flattering everywhere: the largest category is right barely more than half the time, and the day by day table keeps September 1, a day on which both entries missed. Any category under 30 decided entries is marked small sample and shown with no hit rate at all. Those rows are on the page because the rules put them there.

What the record cannot prove

A hit rate is slow to mean anything. Over a hundred start or sit calls, a genuinely good process and a coin can look the same. That is why the earlier market record steers by whether the market moved toward its call after publishing, and why the fantasy record will carry the basis label on every call: a call priced against three complete player markets is a different quality of information from one built on ADP, and lumping them together into one hit rate would hide which half of the process works.

The record also cannot make any single call safe. A 60 percent hit rate means four in ten were wrong, and the one you acted on can be any of them. The record is there so you can decide how much weight to give the next call. It is not a guarantee about it.

What to do with this

1. Bookmark the start or sit record page now, while it reads 577 published and none graded, so that when the first grades land you can see the whole population, not the summary.
2. When you read a call, read its basis label and its closeness line before you act. The record will grade every call the same way, but you should not weigh every call the same way.
3. Hold every other source you use to the first rule. Ask whether their record was counted at publish or assembled afterwards. If they cannot answer, treat the record as a column, not a ledger.