How Much Can You Trust a Diamond Radar Score? Six Roster Updates of Proof

At every major roster update this season, a higher Diamond Radar score meant a better chance of an upgrade, and the cards at the top of the board ended the season about where they started. What changed was the game. As the season went on, a smaller share of cards got upgraded at each update.

When I thought we were grading it wrong

The August 14 roster update graded at 41%. In May the same number had been 77%.

My first thought wasn't that the model had broken. It was that we were grading it wrong. That grade only counted the cards Radar flagged, and whether a card got flagged depended on a line I'd drawn. Move the line and the number moves, whether the model changed or not.

So in September we changed how it's graded. I wrote up the full switch in how we grade Diamond Radar. The short version: every card Radar had the stats to score now counts, flagged or not. Each one goes into a range by its score, and each range is read against chance: the share of all those cards that got upgraded, whatever Radar thought of them.

That answers a better question than "what was the hit rate". It answers how much you can trust a score. This post runs that grade on all six 2026 major updates, one at a time.

At every major update, the order held

Every card Diamond Radar scored before each major update, graded after it. The dashed line is chance: what you'd get picking a card blind.

Six updates, four score ranges each. In every one of them, 70+ beat 50–69, 50–69 beat 30–49, and 30–49 beat everything under 30. Not once did the order flip.

The top range beat chance by about 3 times in May and again in October, with a dip in June and July that I'll come back to. At its lowest, in June, it was still about 2.6 times.

The very top of the board tells the same story. Radar's 10 highest-scored cards went 10 for 10 on May 8. On October 2 they went 10 for 10 again.

Pool all six updates together and you get the season table on the track record:

The Diamond Radar track record on October 20, 2026. Every range steps up from the one below it.

That's the answer to the title. Measured against chance on the same update, a high score in October was worth about what it was in May.

What changed: the game got quieter

Here's what didn't hold: the share of cards that got upgraded at all. At the May 8 update it was 30%. At August 14 it was 19%, and it finished at 23% on October 2.

It isn't that SDS stopped making changes. By our count it changed somewhere between about 720 and 775 players at every major update, and it stayed in that range all season. October 2 even had more upgrades than May 8. There were just a lot more cards to spread them across, and SDS moved each player less.

The same 1,019 cards, graded at all six major updates. By October a one-point bump was the most common upgrade by far.

Follow the same 1,019 cards through the season and it's plain. In May, 59% of them changed overall at the update. In October, 37% did. The average upgrade shrank from +2.7 to +1.8, and downgrades shrank the same way, from −2.2 to −1.5. A smaller nudge is less likely to push a card over a whole overall point, so fewer cards show an upgrade at all.

The cheap cards felt it most. Among those same cards, Commons went from a 30% upgrade rate in May to 14% in October. Golds went the other way, to 36% in October, though that's only 78 cards.

August and September also tipped toward downgrades: among those 1,019 cards, slightly more went down than up. October swung back.

Where the model really did slip

This isn't a "the model was fine all along" post. In June and July it got worse.

Chance barely moved in June, and only dipped in July: 30% in May, 29% in June, 26% in July. So most of it wasn't the game. The top 10 went 8 for 10 in June and 6 for 10 in July, and the 70+ range's edge over chance shrank from about 3 times to about 2.6. Most of that slide was the model.

One honest caveat: August was the first update we can be sure the model hadn't partly learned from, so May's numbers may flatter it.

It recovered once the newer versions were doing the scoring, at the August and September updates, and by October the top of the board was back to May's level.

What the grade can't tell you

Radar's scoring changed several times over the summer. Three versions did the grading: one for May to July, one for August, one for September and October. The newest one scores a lot more cards high: about 12% of cards scored 70+ in September and October, against 2–5% earlier. So a 70 in October isn't exactly the same as a 70 in May. That's why the chart reads each range against chance on the same update, and why the top-10 check, which doesn't care about the scale, carries the comparison.

One thing did lose meaning. Under the newest version, a mid score says less. Cards scored 30–49 got upgraded 18% and 19% of the time in September and October, a little under chance.

And this is six major updates in one season. October 2 was SDS's final regular-season roster update, but SDS has put out a post-season update in November before, so the season may not be done.

What to do with this

1. Read a score against chance for that update, not on its own. The track record shows both, by range.
2. Treat 70+ as the range that means something. It beat chance by roughly 2.6 to 3.5 times at every major update this season.
3. Late in the season, read anything under 50 as "probably not". Under the newest version, the 30–49 range did no better than a blind pick.
4. Expect smaller upgrades late in the year. By October, more than half the upgrades on the cards we followed were a single point. Before you spend stubs on a high score, read what a miss costs.

Every card Diamond Radar had the stats to score before each 2026 major roster update, graded after it: 7,767 card grades across May 8, June 12, July 16, August 14, September 11 and October 2. Chance is the share of those cards that got an overall upgrade. Three versions of the model were graded: one for May to July, one for August, one for September and October. "Players changed" counts any rating change SDS made at that update, by our count. The score-range and chance numbers match the public track record as of October 20, 2026. The top-10 results for each update come from the same track-record data; the page itself shows each update's top 20. The same-cards figures come from our per-card grades, and every rating change in them matches SDS's published roster updates.

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