Thirty-Two Days After We Called the Mania: the Slab Lost 15.6%, the Card Lost 3.5%, and One Slab Now Has Five Prices

A month ago we did something the hobby almost never does, which is publish a mania while it was still running and print our own numbers so that somebody could come back and mark them. Anatomy of a Mania went out on August 28 with a registry growth rate, a sales rate, a gem rate and a grading return, all of them checkable. This is the marking. Two of those numbers were fine, one of them was 15% too low, and the biggest surprise is not in the case study at all. It is that the same card, on the same day, carries five different prices in our own files, and knowing which one to read is worth more to you than anything the Charizard did.

Every figure in the left column below is a number that appeared in the August piece. Every figure in the right column is recomputed from the same recorded series 32 days later. The registry line reproduces exactly: the +730 a week we printed is the run rate across the two population readings we held on August 28, 48,561 on August 12 and 49,604 on August 22, which is 1,043 slabs in 10 days. It is what the tape said at the time and it turned out to be the slower half of the story.

Two of those need an apology rather than a verdict. The $2,001 PSA 10 figure we printed in August came from the older of the two key spellings our graded file uses, and that spelling had stopped being refreshed on August 10, so the number was 18 days old on the day we published it. The live spelling read $1,917.50 that morning. And the +$796 grading return traces to exactly that. Run August's own model on the stale $2,000.50 and the $673 raw ask the article quoted and it returns $796.21, which is the number we printed to the dollar. Run it on what the live series actually held that morning, $1,917.50 against a $703.07 card, and it returns +$712.99. Neither error changes the direction of anything here, and both are the kind of thing that only surfaces when you go back and mark your own work, which is the argument for doing it.

The one that matters is the bull case we named ourselves. August's argument was the Umbreon Standard: a registry can grow forever as long as sales velocity keeps pace, and 43.8 verified PSA 10 sales a week is genuinely that kind of absorption. Velocity did keep pace. Recorded PSA 10 sales went from 1,822 to 2,035 over the 32 days, about 46.6 a week, slightly ahead of what we printed. The price fell 15.6% anyway. Absorption at that rate was not enough, and that is a real strike against our own stated test rather than against somebody else's.

Here is what our files hold for this card's PSA 10 on September 29, 2026. Not five guesses. Five different questions, each answered correctly, each answering something else.

Three of those five sit inside $33 of each other and they are the three shortest windows. The two that stand 23.6% and 29.8% above them are the two longest. That is not a fault in either one. A 90 day median on September 29 still contains most of July and August, and July and August on this card were the top of the run, so a window that long has to read high on a card that fell inside it. It reads high by construction, in a direction you can predict before you look. Our card page publishes the $1,999.48 and labels it as a 90 day median, which is the honest thing to do with it, and it is still the wrong number to price a sale off today.

This is not one card being odd. Across all 340 PSA 10 lines in our files that carry both a provider figure and a 90 day median of 20 or more verified sales today, the 90 day median reads a median 12.21% higher. It reads higher on 82.35% of them, more than 10% higher on 57.35%, and more than 20% higher on 31.76%. It reads more than 10% LOWER on 5.59%, which is the same mechanism running the other way on cards that rose. And the size of the error moves with the size of the fall.

So the trailing window is not randomly wrong. It is wrong by roughly the size of the move, in the direction the move came from, and a 90 day window needs a full quarter to work a fall out of the number. On a card that dropped, it is the last figure in the building to notice.

There is a worse version of the same trap, and we walked into it in a draft of this article. Our graded file stores a recorded price and a recorded sale count side by side, and it is natural to read them as one claim: $1,618.00 on 2,035 sales. They are two separate fields from the provider and they do not describe the same pool. The tape settles it arithmetically. Between September 28 and September 29 we hold 182 PSA 10 lines with 200 or more recorded sales that gained fewer than 2% new ones. Of those, 16.48% moved their recorded price by more than 2% in that single day and 8.24% moved it by more than 5%. One line moved 17.92% while its sale count did not move at all. A median over a pool of 235 sales cannot change when the pool stays at 235 sales, so whatever that price is, it is not a median of the number printed next to it.

August ran the registry-average model from our ROI study, and it has three outcomes rather than two: the gem rate times the PSA 10 figure, plus the recorded PSA 9 rate times the PSA 9 figure, plus whatever is left over valued at 85% of the ungraded card, less a $20 fee, against the price of the ungraded card. Nothing in the model changed. Its price inputs did.

That is a 25.5% compression in a month, and it is honest to say out loud that +78% is still a strong number. Nothing here says stop grading this card. What it says is that the whole return now leans on one outcome. The 59.74% that come back a ten contribute $966.63 of expected value. The 34.14% that come back a nine contribute $228.06, the 6.11% that come back lower contribute $35.27 at the model's 85% of raw, and the reason the nine leg is so thin is the striking number in this section: a PSA 9 of this card has a recorded figure of $667.93 while the ungraded card is quoted at $678.59. The nine is worth less than the card you sent, before the fee. Across the 726 cards in our files priced at $100 or more ungraded, the median PSA 9 is 1.80 times the card and 26.72% sit below it, so this is not rare, but at 0.98 times the card this one is near the bottom of that distribution. If you grade, you are buying the ten and eating the nine.

The tempting headline is that manias live in the slab and not in the card, and our cohort will not support it as a law. We built 130 comparable lines: a PSA 10 reading on both August 28 and September 29 with new recorded sales in between, an ungraded price on both days, a clean population pair, and no set released inside the window. Across those 130, the middle line's seven day slab figure fell 6.61% and its ungraded price fell 4.04%, and the slab fell harder than the card on 52.31% of them. That is a coin flip. The middle gap between those two is 0.41 points. This card's seven day gap of 15.6 points ranks 12th of 130, so it is unusual, and eleven lines were wider.

The clearest evidence against a law is one shelf over. Mega Charizard X ex 130, the Mega Hyper Rare printing of the same character in the same set over the same 32 days, did the reverse: its recorded PSA 10 figure moved -0.6% and its seven day figure moved +10.2%, while its ungraded price fell 12.7%. And Phantasmal Flames as a whole rose over the window, a median 7.69% across the 139 cards priced on both days, with only 35.3% of them down. The set went up. The flagship's card went slightly down. Its slab went down hard. Three different things happened inside one set, which is why the finding here is about one card and one measurement problem rather than about slabs in general.

We expected registry growth to be the variable. It is the mechanism behind the Population Growth Tax, it is what makes a fast registry different from a closed one, and in this cohort it looks like it works. Sort the 130 lines by how fast their registries grew and two columns fall in step across all four quarters: the median move in the seven day figure and the share of those figures that fell at all. The card column does not, and that mismatch is itself part of the story, because it is the slab that separates cleanly and not the card.

Mega Charizard X ex is the 8th fastest registry of the 130 at 1.70% a week, and it sits in that top quarter, where its seven day figure fell 19.1% against a 3.5% fall in the card. Umbreon VMAX, the card we held it against in August, grows 0.24% a week and lives in the bottom quarter, and there its two prices moved together: a seven day figure down 5.3% against a card down 6.9%. The story fits.

And we cannot separate it from how recently the set released, so we are not going to hand you a screen built on it. Registry growth and release date rank-correlate at 0.603 in this cohort, strong enough that the two are largely one variable wearing two hats. Growth against the slab move ranks at -0.335 and release date against the slab move ranks at -0.29, so the table above is about as consistent with "young sets fell" as it is with "fast registries fell". Cutting to the 50 lines from sets released in 2024 or later keeps the growth effect alive: the faster half of them lost a median 14.42% on the slab against 7.83% for the slower half. Cutting again to the 35 lines from 2025 or later removes it, because there the two halves land on top of each other at -11.01% and -10.44%. At 130 lines and 32 days this does not resolve. A fast registry and a young set are the same card most of the time, and telling you to screen on one of them while we cannot tell them apart would be selling you a rule we have not earned. We will run it again on a longer window.

It does not say whether the run is over. Thirty-two days is one month of one card, the price fell into a rising set, and nothing here separates a mania cooling from a normal post-peak drift. It does not say the slab is the worse instrument: the cohort says that is a coin flip and the sibling printing in the same set did the opposite. It does not say registry growth predicts prices, for the reason in the section above. It does not price the card for you, because the honest answer to that is a range and the range is in the table. It says nothing at all about grades below 9 or about other graders, and nothing about the Japanese printings, which carry their own registries. And it is not a statement about what any single copy will sell for: the five figures are population statistics, and the copy in your hand has a centring and an edge that none of them know about.

Sources, all moonstone recorded data read on September 29, 2026. The ungraded prices are price_history, the TCGplayer market price recorded daily, which is derived from recent sales rather than from live listings, so it is neither an asking price nor a comp you can point at. Its window here is August 30 to September 29, 2026, its first stored day through its last; the daily file stamps a value two days later than the Near Mint series does, and the Near Mint series holds $703.07 for this card on August 28, which is the same figure the daily file stamps on August 30, so the publication day quote is the same number in both. The day after this window the file already reads $672.28, which is the figure the card page shows. The graded figures are graded_snapshots. Two of its columns carry a reading on both August 28 and September 29: the provider's recorded figure for the grade, which marks -15.62%, and the provider's seven day market figure, which marks -19.08%. Our own verified medians begin on September 25, when the dated price model shipped, so they cannot measure this window and are not used to. We publish the recorded figure as the headline because it is the primary column and the more conservative of the two. The recorded figure is NOT a median over the recorded sale count beside it: those are two separate provider fields, which the one day test in the body demonstrates on 182 lines. That card's graded readings also skip August 30 to September 13, because the daily graded refresh rotates across the catalogue, so the series holds 18 readings across the 32 days rather than 33. Population and gem rate are the daily PSA registry snapshots, which for this card run August 22 to September 28, 37 days, 49,604 slabs to 54,054 and 29,648 tens to 32,293. Two readings inside that stretch, September 9 and September 11, record a gem count of zero against a null PSA 9 count and are excluded as unknown rather than read as zero. The +730 a week we printed in August reproduces from the two readings we then held, August 12 and August 22, 48,561 to 49,604. The grading model is the registry-average model in our ROI study, read from that study's own methodology rather than recalled: the registry's PSA 10 share times the PSA 10 figure, plus its PSA 9 share times the PSA 9 figure, plus the remaining share valued at 85% of the ungraded price, less a $20 fee, against the ungraded price. The shares come from the same population readings, 29,648 tens and 16,909 nines of 49,604 on August 22 and 32,293 tens and 18,456 nines of 54,054 on September 28. It carries no shipping, no insurance and no eBay fee, exactly as August ran it, and on the stale $2,000.50 figure with the $673 ask the August article quoted it returns $796.21, which is the +$796 that article printed. The 130 line cohort requires a PSA 10 reading on both endpoints with five or more recorded sales AND a seven day figure recorded on both endpoints, which is the largest single exclusion: 195 lines carry the recorded figure on both days and only 147 of them carry the seven day figure on both, so 48 are dropped there before any other rule runs. It then requires MORE recorded sales at the end than at the start so that a line which never re-quoted cannot be counted as flat, an ungraded price on both days, a population pair at least 20 days apart whose readings never fall and never rise by more than half in a single step, a starting population of 500 or more, and a set released before the window opened. Those rules exclude 7 lines for recording no new sales, 8 for a short or absent population pair, and 2 for population readings that shrank or jumped, one of which is Jungle Pikachu going 11,792 to 18,919 in one reading, which is a registry remap and not grading. One catalogue row per TCGplayer product id throughout, because 1,215 product ids in the catalogue are claimed by two rows each; this card is not one of them and resolves one to one to product 662184. The 340 line trailing window comparison is every PSA 10 line holding both a provider figure of five or more sales and a 90 day verified median of 20 or more sales on September 29; lines whose window had to widen past 90 days are excluded so the window is one claim. Correlations are Spearman rank correlations. Phantasmal Flames released on November 14, 2025 per the price file's own set metadata, well before the window, which is what rules out a release day collapse being read as a decline. Not financial advice.