How Many Pokémon Cards Have a Real Market?

A price is supposed to summarise transactions. We checked how many of ours do. We evaluated 63,600 catalog rows, which collapse to 62,388 entries once the 1,212 marketplace products claimed by two catalog rows each are deduped. Of those entries, 44,395 carried a quoted price on September 22, 2026, and 11,747 carried a recorded sale of any kind behind that price. The other 32,648, which is 73.5% of every priced entry on the book, carry a number and nothing else. That is not a coverage gap on our side. Every single one of those 32,648 entries had been checked against the graded sold record, and the record came back empty.

Sort every catalog entry by the strongest evidence we hold for its price and the hobby separates into tiers that have very little to do with how confident the price looks on a page. Note the median quote in the right-hand column: evidence and value rise together, so the tier with nothing behind it is overwhelmingly cheap cards. That is a mercy, and it is not the whole story, because 197 entries between $100 and $500 and nine above $500 also sit in it.

Weight the same ladder by dollars rather than entries and it looks healthier, which is the honest way to read it. Summing every quoted price on the tape gives $739,142 for one copy of everything priced. Entries with no recorded sale hold $157,135 of that, 21.3%. Entries with a tape of at least ten verified sales hold $431,617, which is 58.4%. So most of the catalog's entries are unevidenced and most of the catalog's quoted value is evidenced, and which of those two sentences matters depends entirely on whether you are pricing a collection or pricing a single card you are about to buy.

The last column is the one people misread. The share of entries where the quoted price is exactly the cheapest copy for sale RISES with price, from 69.3% under a dollar to 92.9% above $500. Expensive cards do not attract a crowd of competing sellers; they attract one. Combine the two failure modes and 25,534 priced entries, 57.5% of the book, carry a number that is simultaneously the only listing visible and unsupported by any recorded sale. In the $25 to $100 band that is 886 entries, one in four.

A study like this is worthless if the missing evidence is just missing work, so we separated the two. Our cache keeps a negative marker for every card we asked the graded record about and got nothing back. All 32,648 quote-only entries carry that marker. Not one of them is unchecked. That turns the claim from a gap in our coverage into a measurement: for these cards, at the moment we looked, the graded sold record held nothing.

It does not turn into a claim that these cards never sell. A bulk common with no graded sold record is a card nobody grades, not a card nobody buys, and the correct label on its row is "no recorded sale", never a zero and never a dash that a reader will round to zero. We learned that the hard way when an absent PSA 10 row was read as a 0% gem rate and a ranking engine promptly scored it as the best candidate on the board.

Japanese entries outnumber English entries in our catalog by three to two and carry less than half the graded-evidence rate: 12.6% against 28.5% for any tape at all, and 6.0% against 17.1% for a tape of ten sales or more. Any cross-language price comparison inherits that asymmetry: the English number is likelier to be standing on something. That is a data fact about our book, not a claim about the Japanese market, which trades heavily on venues our sold record does not read.

The finding is that the modal catalog entry is a quote with no recorded transaction behind it. It falls if a second independent sold record, plugged into the same 32,648 entries, returns transactions for a large share of them; that would show the emptiness was our source's blind spot rather than the market's silence. It also falls if the negative markers turn out to be stale, so the honest test is to re-ask the record for a random thousand of those entries and count how many come back non-empty. Anything above a few percent and the headline share is too high.

The sold record we check is a graded one, so a card that trades briskly in raw form and never gets slabbed reads as having no recorded sale here and should not be read as illiquid. The 63,600 figure is the number of rows we evaluated as they sit under our own id convention, and the 62,388 entries are those rows deduped on the marketplace product id; neither is a canonical count of distinct Pokémon cards, which is a separate census. The 17,993 unpriced entries are mostly entries carrying no marketplace product id, which is a catalog-linkage fact rather than a market fact. And the whole thing is one day: September 22, 2026. We have not tracked whether the unevidenced share is rising or falling and make no claim that it is.

Sources: `cards` (63,600 rows, the catalog as our id convention holds it), `price_history` on day 20718 (September 22, 2026 UTC) for the quote and the cheapest listing, `graded_snapshots` for the graded sold tape, read as the LATEST RECORDED ROW PER CARD KEY AND GRADE with day 20719 excluded because it was still being written, an entry tape being the largest such sample size across every grade and across both key forms behind it, `ebay_divergence` on day 20717 for the raw sold median, and the `gradednone:` negative-cache markers for the checked-and-empty test. A row is matched to price and tape by its catalog id and by its marketplace product id, both, because the graded tape is keyed on either. EVERY COUNT IS DEDUPED ON THE MARKETPLACE PRODUCT ID: 1,212 product ids are claimed by two catalog rows each, 2,424 rows of which 1,212 collapse, all 1,212 pairs sitting inside a single language, so an undeduped count would overstate the catalog by 1.9% and could place one product in a band twice. The 63,600 rows become 62,388 entries; a row with no product id is kept as its own entry, and an entry inherits the deepest tape of any row behind it. "Any recorded sale" means a graded sold median on at least three verified comps, or a raw sold median on at least five. Those floors are exclusions: lowering the graded floor cannot raise the count because the tape itself refuses to publish below three, while lowering the raw floor to one comp adds a negligible number of rows. The dollar-weighted totals sum one copy of each priced row at its quoted price, which deliberately ignores how many copies exist and is a measure of catalog breadth, not market capitalisation. Quoted prices are ASKS. Graded medians are SOLD medians. An absent row is not recorded, never zero. Not financial advice.