What a 90 Diamond Radar Score Is Worth, Hit or Miss

TL;DR. Pooled over all five roster updates we graded in 2026, a Diamond Radar score of 90 or better upgraded 78% of the time and averaged the best result on the board. When one of those calls missed, it also cost the most — a typical −232 per 1,000 stubs. But the typical miss is a cheap card, and most of what it costs you is just the price of selling it — that part is near-identical in every band. Misses on cards over 100 stubs cost more on top of it. What the score changes is whether the price then moves for you or against you on top of it.

---

This one started as a theory. It came to me one day in September, right after the September 11 report card, and I wrote it down before I lost it so I could dig into it later.

Here's the theory. When a top-band card upgrades, the price jumps. Fine. But when it doesn't — does the price just hold? A wash, minus the tax?

Because if misses are mostly break-even, a top-band card is a small downside against a large upside. That's an easy thing to tell you.

Half of that was right. The half that wasn't is this post.

What I actually measured

Five graded roster updates this season: May 8, June 12, July 16, August 14, September 11. For every call the Radar made, I took the last price the day before and the price two days after. Then I ran it through an actual sale. Everything is per 1,000 stubs spent, so a 7-stub Common and a 12,000-stub Diamond sit in the same column.

"Through an actual sale" is doing work there. You don't get the sticker price. You undercut by a stub to move it, and the marketplace takes 10% on the way out. Quick sell dodges the tax, but pays a flat rate that's usually well under market. So your exit is whichever of those two is better. Every number below is what you'd actually be holding afterwards.

The Radar made 1,485 calls across those five. Ten had no usable price, which leaves 1,475 calls across 809 different cards — a card counts once per update it was called in. A "miss" is any call where the card didn't upgrade.

Six bands, side by side

All per 1,000 stubs, two days after the update. Another 48 calls scored between 25 and 39. I've left them out of every table here because they're a different kind of call: mostly Diamonds and Golds, at a typical 3,000-plus stubs. 41 of the 48 missed, and the typical one fell about 11%. So the rows add up to 1,427.

Two columns there do different jobs. "Average, all cards" is the mean across the band, with the most extreme 5% at each end pulled back to the 5th and 95th percentile. That stops one Common that went up 1,000% carrying the whole row. "Typical card" is the median.

They come apart in the bottom two bands. The average is positive; the typical card isn't. Those rows are carried by a few large winners, not by most of the cards in them.

The top band pays the most and costs the most when it misses. Both bars are medians, and selling costs are included.

The 40–49 row is thin in a way worth naming. 88% of those cards come from two updates, July 16 and August 14 — 311 of the 354. The other three contribute 5, 12 and 26.

The part I had backwards

I assumed a 90-plus miss would be a wash. A card that good doesn't just fall over, and it hadn't earned a downgrade — it simply didn't move.

It was the most expensive miss on the board. A typical −232 per 1,000 stubs. Worse than the 50s. Worse than the 40s. Worse than everything. The order isn't even tidy — a miss in the 70s cost −56, roughly nothing. A miss in the 90s cost four times that.

Fourteen misses is a thin sample, and nine come from September 11 alone. I'm not calling this settled. But when I went looking for why, the answer turned out to be two separate things — and only one of them is about the score.

The typical miss costs you the sale

Here's the part I didn't expect. The typical missed call, in every one of those six bands, is a cheap card:

That last column is the whole trade on a card whose price never budged. Undercut it by a stub, pay the 10%, and you're down about 120 per 1,000 stubs. In every one of the six. The score has nothing to do with it.

So being wrong has a fixed cost. It's the same cost whether the Radar gave the card a 95 or a 45. What the band decides is what the price does on top of that floor — and there the bands genuinely differ. A missed 70-something rose 7.4% and clawed most of the friction back. A missed 90-something fell 5.4% and doubled the damage.

That reframes the top band. A 90-plus card doesn't collapse when it misses — the typical one slipped five percent. Of the six bands, it's the one where the price move works hardest against you instead of softening the blow. Only the 40s also went backwards, and by half as much. And that has a cause.

You pay for the update before it happens

Here's how much a card gained in the seven days before the update, counting only the cards that went on to miss:

Cards the Radar flagged rose hard in the week before the update. The cards it didn't flag, whose rating then didn't change, moved 3.1% — 6,106 card-by-update readings across 1,821 cards.

Cards in the top band ran up 43% in the week before the update, even when the call was about to be wrong. The ones about to be right ran up 38.5%. Same story.

That top-band figure leans almost entirely on September 11, where those nine misses ran up a typical 73%. The four earlier updates contribute five top-band misses between them. May 8 contributes none at all — all ten of its top-band calls upgraded. The broader run-up, across every card the Radar flagged rather than the top band alone, does show up in all five.

By the day before the update, the market has already priced in the thing you're betting on. It priced it into the winners and the losers alike.

That's the difference between the top band and the rest. Everybody watches the highest scores. The price gets bid up before the update lands, and when the upgrade doesn't come, that premium goes back where it came from.

Which means the same fourteen misses look completely different depending on one decision:

Same cards. Same misses. Buy them a week out instead of the night before, and the worst band on the board breaks even. The run-up that opens that gap belongs to the cards the Radar flagged. The ones it didn't — whose rating then didn't change — moved 3.1%. So this isn't the market drifting up underneath everything.

That's the practical finding: buy about a week out, not the night before. Whether further out is better is a different question. I don't have that answer yet.

What the card costs changes the picture

I promised this in the last report card, so here it is. The same misses, split by what you paid. Diamonds behave differently enough to get their own row rather than being folded in.

Card counts in brackets. A dash means fewer than three calls landed in that cell, not that nothing did. The top band has two misses above 100 stubs — Cal Raleigh at −376 and Nick Kurtz at −112. They're the top-band blanks on the second and third rows. Several of the printed cells are small enough that I wouldn't lean on them alone either.

Under 100 stubs, a miss is close to a wash from 50 to 89 — the price rise covers most of the selling cost. The exception is the top band, and it isn't a small one. Twelve of the fourteen top-band misses were cheap cards. So the −232 headline is mostly a story about cheap cards giving back a run-up, not about expensive ones collapsing.

At 100+ stubs and not a Diamond, every band loses more than the friction. The worst cells with enough calls to read sit at both ends — −246 in the 80s and −311 in the 40s. It isn't a clean slope, and the 80s cell is six cards. Treat the shape as "bad across the board", not as a ranking.

Diamonds rise less when they hit — +12% against +26% for everything else — and fall further when they miss: −7.5% against flat. Which brings me to the part I'd rather say plainly than bury.

There's no edge on Diamonds

Across 96 Diamond calls — 44 different Diamonds, counted by what they were when the Radar called them, and including 26 from that 25–39 group — the average was −107 per 1,000 stubs. Negative in every band with more than a dozen calls. Only 24% upgraded.

Diamonds are already priced like good cards. A rarity crossing is what re-prices a card, and a Diamond has one rung left above it. Not one of the 96 Diamond calls here crossed into Red Diamond. If you're spending 12,000 stubs on a Diamond because it scored well, this data doesn't support you. And every Diamond cell here is thin enough that I'd want another season before saying anything stronger than "no edge."

What a free account sees

Worth saying plainly, because this whole post is about calls you can actually act on: about a third of the calls above — 519 of 1,475 — involve a rarity a free account doesn't get. Free and guest accounts see Common and Bronze rows. Pro sees every rarity.

That cuts both ways, and you should have both halves. The free slice upgraded more often — 58% against 44% — because low-rarity cards cross a line more easily. But its typical winner paid +64 per 1,000 stubs against +209, on a median card of 44 stubs against 254.

Which is the same lesson as the rest of this post, pointed at our own paywall: on the cheapest cards, the cost of selling eats most of what you win. Pro isn't a better hit rate — the free slice hits more often. It's the third of the board where a hit is worth about three times as much.

Whether that's worth paying for depends on how many stubs you're actually moving. I'm not going to pretend five roster updates can tell you that. The tiers are here.

One trade that explains all of it

Chris Sale, September 11. The Radar had him at 55 — middling. He upgraded, 87 to 89. The call was right.

Buy him the day before at 11,900 stubs and he's worth about 8,396 two days later. After the undercut and the 10%, that's 7,555 back on 11,900 spent — about 4,345 stubs gone. On a card the Radar called correctly.

He's a Diamond — the group where an upgrade moves the price least. He isn't in that table, which counts only the calls that missed. His upgrade arrived and the price didn't care. It had already assumed it.

A hit isn't profit if you paid the hype. That's the post.

Chris Sale's last thirty days. The upgrade landed on September 11. The price had already had it.

The same season, scored again

Re-score this season on one scale and five times as many cards land in the top band. They still hit. The 50s through the 80s all get worse.

Everything above uses the score the Radar actually showed you. That was the only number anyone could act on. But the five updates I graded were scored under three different versions of the model, so those scores sit on three different scales.

Core now publishes a second version in the track record's public data. It isn't on the track record page. It re-scores four of the five updates on one common scale — and that scale isn't the one the Radar serves today either. May 8 isn't in it — the data the re-score needs doesn't exist for that update, so it can't be rebuilt. Same four updates, both ways:

Same four updates, scored two ways. The re-score fills the top band and still hits there. Its hit rates are a best case.

The re-score puts about five times as many cards in the top band, and they still land 79% of the time. Every band from 50 to 89 does worse under it, by five to twelve points. The 70s take the biggest hit.

So the new scale is more generous at the top and less sure of itself in the middle. That's the gap.

I wouldn't lean on it much yet, and neither does Core. Its own labels on that series say three things. It's a narrower re-grade than the score you saw, and a different scale. It's graded on the same updates the current weights were fitted to, so its rates are an upper bound. And it couldn't rebuild 2,204 of the 5,289 cards those updates graded — the ones it dropped upgraded 9% of the time, against 34% for the ones it kept.

It also only publishes band totals, not a score per card. So there's no way to run the stub math above on it. Every "per 1,000 stubs" figure in this post stays on the score you were shown.

What would make this wrong

The top two bands are thin — 14 and 20 misses pooled — and September 11 dominates the 90s. The five updates were scored under three different versions of the model, so pooling them pools three models, not one. The section above is the closest thing to a single-model check, and it comes with its own caveats. Prices are daily closes in UTC, so "two days after" is roughly two days.

And the horizon matters more than I'd like. On September 11 the top-band misses looked recovered by day four, then gave it back by day five and day seven. Two days is a choice. A different choice tells a softer story.

What survives all of that is the run-up, which is the part I'd actually change my behaviour over.

Every figure here comes from the same public endpoints that serve this site — the prices on any card page, and the called rows on the track record. Nothing is drawn from anything you can't pull yourself. I rebuilt the whole thing from scratch for this post. The previous version kept its working files somewhere that no longer exists. A number you can't re-derive outlives the moment anyone should have checked it.

Scores are as the Radar served them at the time, except in the re-scored table, which is labelled. Averages over five updates are not a promise about the next one.

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