Diamond Radar's Hit Rate Doubled. Its Edge Shrank.

I figured the November 6 grade would come out skewed before I looked.

Not miscounted. Measuring the wrong thing. Only playoff players were going to move in that update, and Diamond Radar gets graded on the whole league. So I re-graded it on the 12 teams that were still playing. The share of its calls that came true more than doubled, from 15% to 33%.

Its edge over a random pick shrank. And against the same teams a month earlier, it was worse.

Two grades for the same update

Part 1 of this series showed what SDS did on November 6. Nearly every upgrade went to a playoff team, and the 18 teams that went home were mostly frozen. It ended with a promise: Diamond Radar mostly missed this one, and the miss would get its own post, with the denominator.

Here it is. Radar scored 1,454 cards before the update and called 183 of them to go up. These are the same calls the track record grades, on the same line.

Chance is the share of cards in that group that went up anyway, whatever Radar thought of them. It's what you'd get pointing at cards blind.

Grade only the playoff teams and Radar's hit rate doubles. A random pick's nearly triples, so Radar's edge over it shrinks. Source: Diamond Radar track record, re-graded by team.

Same update. Same calls. Two honest-looking numbers that point in opposite directions.

Why the bigger number is the weaker one

The 2.0× is propped up by cards that were never going to move.

The 18 non-playoff teams had 681 cards between them, and 12 went up. The 318 free agents had none. That's about a thousand cards SDS had already decided to leave alone. Radar put 102 of its 183 calls on them and hit one, so they pulled its hit rate down.

They pulled chance down further. Whatever Radar scored them, a thousand cards that couldn't move drag chance to 7.5% on the whole league. Against a bar that low, 15% looks like 2× chance. Some of that 2× was Radar leaning toward players on good teams: 81 of its 183 calls were playoff players, who are under a third of the cards. Good teams made the playoffs. That's a lucky overlap, not a read on October.

Take the frozen cards out and the bar jumps to 21.3%. Now Radar has to beat a random playoff player, not a random free agent. It hit 33.3%. That's 1.6× chance, not 2×.

Same calls, graded on a different group. Nothing about the model changed between those two rows. The hit rate doubled because the dead cards went away. The edge shrank because the easy bar went away with them.

It got worse on the same teams

You could argue the playoff population is just a harder test. So I graded the same 12 teams at the update before, October 2.

Same players. One update apart. The hit rate fell from 48.6% to 33.3%, and Radar's edge fell from 1.9× to 1.6×. That one is a real decline.

There's one more measure I keep that doesn't depend on any line at all: how well the score puts the cards that went up above the ones that didn't. A coin flip scores 0.5. On these 12 teams it went from 0.79 on October 2 to 0.63 on November 6. Worse whichever way you slice it.

Where it held up, and where it didn't

Radar's top scores still meant something on the teams that mattered.

On playoff teams, scores of 60 and up beat a random pick. The 50s didn't. Source: Diamond Radar track record, re-graded by team.

Five playoff cards scored 70 or higher and three went up. Eighteen scored in the 60s and seven went up, 39% against 21% for a random playoff card. Five cards is a hint, not a rate. But the order held.

The 50s are the problem. Forty-two playoff cards scored there and 19% went up, a shade under picking blind.

And the misses at the top have a pattern. These are the highest-scored playoff cards whose overall didn't move:

Four of the nine played for teams that went out in the first round. Part 1 found that the earlier a team went out, the fewer of its cards SDS moved. The four Wild Card teams got 16 upgrades between them. The champion Brewers got 17 by themselves.

Radar had no idea how far anyone went.

The actual miss

Diamond Radar reads regular-season stats. They stopped changing at the end of September. SDS spent October watching the playoffs, and Radar spent October looking at September.

So it scored a Red Sox reliever the same whether Boston went out in the first round or won it all. It couldn't have known. That's on me, not on the data.

Why I publish the comparison, not the hit rate

This is exactly why I retired the bare hit rate. A hit rate without the chance beside it is misleading even when it's true. November 6 is the cleanest proof I have. Same calls, a 15% and a 33%, and the honest grade went the other way from both of them.

If you only remember one thing from this post, remember that a hit rate can double while the edge shrinks.

What you can do with it next November: skip cards on teams that are already out. The board shows every card's team. Radar's top 10 on the whole board had 4 cards go up. Its top 10 among playoff players had 7.

What I'm changing so you don't have to do that by hand is the third post in this series.

Sources: Diamond Radar's graded calls for the November 6, 2026 roster update (update 27) and October 2, 2026 (update 26), re-graded by team on November 9 using the same calls and the same line as the public track record. The whole-league figures match it exactly. "Chance" is the share of every scored card in the group that got an overall upgrade. Team is each card's team today, which matters more for October 2. Playoff results: 2026 MLB postseason. One update is one data point.

I'm Shaun. If something is wrong, tell me.