The Radar Was Looking At One Week To Predict Five
We shipped 1.31. It is one big fix and a handful of smaller ones, and the big fix is the largest accuracy change the Radar has had.
The Radar was looking at one week to predict five
MLB The Show ships roster updates weekly. Only four to eight a season are the majors that actually move ratings — the rest barely touch anything.
The Radar was resetting its stats window on every update, major or minor. So when it tried to call the next major, it was looking at whatever happened since the last minor — usually about a week — and predicting a five-week stretch from it.
Measured across this season, that is 16.9% of the games actually played.
Some windows were far worse. The one containing the All-Star break gave the model a maximum of one game per player. SDS still moved 194 cards in that update. We were calling it on almost nothing.
It now measures everything since the last major.
Does it actually help?
We re-ran it against every season we hold — 2021 through 2026, 34,150 card-window observations, 48 roster updates. It came out ahead in every season and every rating band.
The gain is largest exactly where the Radar was weakest:
| before | after | |
|---|---|---|
| Cards 85 OVR and above | barely better than a coin flip | genuinely informative |
And the size of the improvement tracks how much baseball was being thrown away — seasons where the majors fell close together barely changed, seasons where they were five weeks apart changed the most. That is the fingerprint of a real fix rather than a lucky one.
Predictions from before and after this release are not comparable. The scale moved.
The best cards were the ones we could say least about
Card attributes reach 125. The curves we score against stopped at 99.
A card already above the curve had no room left to score, so its strongest attributes were invisible to the upgrade signal. At 90-99 OVR, 29% of scored attributes were dead — against under 1% at 74 and below.
The Radar was quietest about the cards you care most about, and it was quiet for a mechanical reason rather than a real one.
Each curve now extends to the ceiling its attribute actually reaches.
The probability now admits what it is
Until enough roster updates have been graded under the current model, the percentage next to a card is a ranking translated onto a 0-100 scale — not a measured frequency. It was showing 95%+ on cards where nothing had verified that number.
It is now capped at 80%, and it lifts on its own once we have real measurements to replace it with. A lower number that means something beats a higher one that does not.
Smaller things you will see
- Pitchers were showing "0 games" next to real innings pitched. The game count was never read from the pitching feed at all. It is now the real number.
- Every stat block now says what period it covers — start date, end date, games played. Previously you got the numbers with no way to tell a week from a month.
- Strikeout and walk rates for hitters; walks, hits and home runs per nine for pitchers. These hold up at small sample sizes where batting average is still noise.
One thing to know about the dates
The recent-form window changes length through the cycle — a few days right after a major update, around five weeks just before the next one. So it can be shorter or longer than the 30-day column depending on when you look. That is the window doing its job, not a bug.