Stat corrections — why a prop can change after the game

You watched the player finish with six catches. Your over cashed. An hour later the sportsbook changed the ticket.

That feels impossible until you remember what a player prop actually settles against: not your television broadcast, not a fantasy app, and not necessarily the first box score you saw. It settles against the official statistical source named in the sportsbook's rules.

Live stats are provisional

During games, statistics change constantly.

A rushing attempt becomes a sack under a league's scoring convention. An assist disappears. A rebound is reassigned. A hit becomes an error. A tackle moves from solo to assisted.

Most of those edits happen before final grading. Some arrive after a ticket was initially marked settled.

The official source controls

Sportsbooks usually define an official league or statistics source for settlement. That matters when two apps disagree.

If one data feed shows 74 receiving yards and the official source closes at 73, the book does not average the feeds. It follows its settlement rule.

This is another reason Sharp's data layer should preserve source attribution, not just the number. A projection input and a grading source are different jobs.

Regrades are operational, not handicapping information

If a ticket changes from win to loss because of an official correction, that does not mean your original edge calculation was wrong.

Your process should distinguish:

- forecast quality — was the price +EV when you bet? - event outcome — what happened on the field? - settlement outcome — how the official statistic was ultimately recorded?

Mixing those three creates bad postmortems.

What to do with disputed props

First, check the official stat source.

Second, check the book's market rule and correction window.

Third, save the ticket and settlement record if something still looks inconsistent.

Do not use a screenshot from an unofficial scoreboard as the final argument. The book is grading a defined statistical contract.

Model builders: this matters even more

Historical prop data needs the final official stat, not a live snapshot captured five minutes after the game.

If your training set contains provisional statistics while your target market settles on corrected official statistics, you have introduced label noise. On high-volume prop models, that can be enough to distort small edges.

The rule to carry

A player prop is only as precise as its settlement definition.

Know the official source. Store the final stat. Keep forecast, result, and grading as separate fields.