Buy a Week Out. Don't Sell the Night Before.

TL;DR. Pick cards by the Diamond Radar score showing on the day you buy, and a week before the roster update is still the best day to do it: a typical +333 per 1,000 stubs for cards scored 70 or better, against +62 the night before. Going further out doesn't help. And once you own them, hold through the update. Selling the night before broke even, a typical 0.0%, where holding kept +16.5%.

---

Buy about a week out, and hold through the update. That's the whole post. The rest is why I believe it, and the prediction I made that only half came true.

The miss-cost post left a question open. Buying a week out beat buying the night before, and in my words: "Whether further out is better is a different question. I don't have that answer yet." This is the answer.

It also answers a question that finding raised for me. If the cards the Radar flags run up hard in the week before an update, why hold through it at all? Buy a week out, sell the night before, bank the hype and never find out whether the call was right. That sounded clever. It isn't.

The number I couldn't use before

Every buy-timing figure I've published picked cards by their Diamond Radar score at the update. Nobody buying a week earlier saw that score. They saw whatever the Radar showed that day, which might have been lower, or below the line entirely.

So this time I picked every card by the score the Radar was actually showing on the day you'd have bought it.

That needs the score as it stood on each of the fourteen days before every update. The site doesn't serve that yet. Our own database does, so that's where these scores come from. Prices and outcomes are the same public data as always. The scores are not something you can pull yourself yet, and I'd rather say that than have you find out.

This covers all six graded updates of 2026: May 8, June 12, July 16, August 14, September 11 and October 2. Coverage is complete, all 84 of the days before them. One of those days, August 2, has no scores: that night's run finished without saving anything. When that happens the Radar keeps showing the last board, so August 2 uses August 1's scores. That's what anyone looking that day saw.

It made a difference. Only 71% of the cards scored 50 or better a week out were still scored that high the night before. And more than a quarter of the cards that ended up flagged weren't flagged yet.

A week out still wins

Same method as the miss-cost post. Buy at that day's closing price, sell two days after the update, undercut by a stub and pay the 10% on the way out, or quick sell if that's better. Everything is per 1,000 stubs spent.

Seven days out was the best day for both groups. The cards nobody flagged lost money every day, mostly to the cost of selling.

Typical result (the median), per 1,000 stubs, pooled over six updates. 1,487 cards at a score of 50 or better a week out, 489 at 70 or better.

The flagged cards' lead over the ones nobody flagged peaks at seven days out: +323, against +155 the night before. Drop any one of the six updates and re-run it, and seven days out is still the best day every time. Update by update, cards scored 70 or better did better bought a week out than the night before in five of the six. July 16 was the exception, and it was close to a tie.

Two days out is the odd one. It comes second, at +262, and it's second in five of the six leave-one-out runs too. I don't have a story for it, so I'm not going to make one up. It sits between three days out and the night before, and both of those are weak.

Going further out doesn't help. Fourteen days out returned a typical +18 for cards scored 50 or better, and nothing at all for 70 or better. Ten days out lost money for both.

Picking by the score on the day costs something. Re-run the old way, picking by the score at the update, and buying a week out returns +238 for cards scored 50 or better. Picked by the score showing that day, it's +185. Fourteen days out falls furthest, from +110 to +18. The shape held. The level was flattered.

It also helps on the cards that missed. Scored 50 or better and bought the night before, a typical miss cost −96. Bought a week out, it cost −32. This is every miss scored 50 or better, and it doesn't depend on knowing the final score.

The thing I can't separate

Five of the six updates landed on a Friday. So "a week out" was a Friday too, every time but once.

And ten and nine days out sit low for everything, including the cards nobody flagged: −179 and −172, against a usual −143. Some of what looks like timing is probably just the calendar. Six updates can't pull those apart.

The updates don't agree with each other either. On July 16, cards scored 50 or better and bought eight to fourteen days out lost 229 to 387 per 1,000 stubs. On June 12, it was the last four days that lost. October 2 was flat almost everywhere: a typical 0 a week out for cards scored 50 or better, and only about half of them upgraded. And September 11 and October 2 supply about two-thirds of the cards scored 70 or better, because the model in use by then flagged far more of them. The curve above is an average of six rather different weeks.

So I'd put it like this. A week out is the best measured day, it held in every leave-one-out, and it's the answer I'd act on. It is not a law.

Don't sell the night before

Now the clever idea. Buy a week out, sell the night before, keep the run-up.

Typical result, net of the undercut and the 10%. 1,829 cards at or above the Radar's flag line a week out (50 for most updates, 40 for July 16 and August 14), 97 in the top band.

Same cards, same buy day. The only difference is when you sold.

Holding wins, and not narrowly. Whatever run-up there is, the night before mostly hasn't got it, and the selling costs eat the rest. The bigger move is the upgrade itself, and you only collect that by still owning the card. On cards that upgraded, the flip kept a typical +5.7%. Holding kept +36.4%.

The flip doesn't even buy you safety. The idea was that you'd give up some upside to dodge the misses. On the misses, it's slightly worse: −5.6% flipping, −4.1% holding. And in the top band the flip had the worse worst case: −929 per 1,000 stubs, against −564 for holding.

That worst case is A.J. Ewing on July 16. A Diamond Radar score of 91 a week out, bought at 70 stubs, listed at 5 at the night-before close, by which point his score had slid to 47. Then he upgraded. The flip sold the bottom and missed the jump.

Buying fourteen days out and flipping is worse again: a typical −10.0%, with only 35% ending in profit.

I owe you a correction here, even though it never went out. My first pass at this question said the flip was half right: smaller gains, but steadier, and it rescued the misses. That pass valued a sale the night before at the card's quick-sell value after the update, a floor the card didn't have yet. It was crediting the flip with money nobody could have collected. Fixed, and with cards picked by the score on the day, the steadier-and-safer story is gone. If you saw those numbers anywhere, they were wrong.

The prediction I got half wrong

In the market timing post I put a guess on the record. In the days before a roster update there's a real reason not to sell a card that might be about to get better. So, if that held at any scale, the usual morning slump should flatten, or invert.

The morning slump did flatten, a little. The rest of the day got choppier.

The prediction was a calmer day. Mornings calmed a little. The day as a whole didn't, most of all on the cards everyone was watching.

I can only test it on four updates. The hourly sales data goes back 90 days, so July 16, August 14, September 11 and October 2 are in and the first two aren't. That's twelve days before an update, against 60 normal ones.

On a normal day, the average price curve for the cards the Radar flagged (scored 50 or better that day) swings about 18.4 points from its cheapest hour to its priciest. In the three days before an update, 22.2. Across all cards, it's 14.4 on a normal day and 16.8 before an update. (The timing post measured 18.3 across all cards, over a partly overlapping stretch of the season.)

The flagged cards' swing beat the unflagged cards' swing by more than on a typical normal day on ten of the twelve days. The two that didn't were the last two before October 2, the quietest update of the four.

Here's what came true. Between 6 and 9 AM Eastern, the flagged cards' prices sat less far below their daily average before an update: −5.7% against −8.5% on a normal day. But the low point didn't go away. It moved later, to around 10 AM, and got deeper. And daily sales did thin out before all three of the updates I can measure, by 7% to 17%. Sales held around 1.2 to 1.5 million a day until mid-August, then fell by about two-thirds by late September, so I can't tell how much of that is the update.

So, half wrong. The early morning looks like what I predicted, a few people holding back. But nobody stops selling, and for the cards everyone's watching, the last three days before an update are usually more nervous than a normal day, not calmer.

What this means for the next update

• Buy about a week out, picked by the score the Radar is showing that day. Not two weeks. Not the night before.
• Hold through the update. Selling the night before gives back almost all of the gain and doesn't protect you on the misses.
• None of this is about one card. These are typical results over hundreds of cards: 489 scored 70 or better a week out, 97 in the top band. Any single card can do what A.J. Ewing did.

What would make this wrong

Six updates, graded on three versions of the model. Counting the buy days, at least six versions served scores, because new ones landed inside two of those fourteen-day stretches. A score of 70 a week before July 16 and a score of 70 a week before October 2 don't mean quite the same thing. Picking by the score on the day is the honest choice, because it's the only number anyone could act on. It just isn't one model.

The top band is thin. Ninety-seven cards a week out, and only thirteen of them missed, which is too few to say anything about top-band misses at all.

"The score that day" means the board as stored at the end of the day. On some of the days in range, a same-day re-run replaced the morning board, so anyone who bought before it saw something slightly different.

The cards nobody flagged are the Live cards that were listed on September 21, when I pulled the listings.

And the scores come from our database, not a page you can check. That would change if the score history went public.

Scores are as the Radar showed them on each buy day. Prices are daily closes in UTC. Averages over six updates are not a promise about the next one.

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