How we price a Pokémon card

Every number on a PokeInvest card page is a median of real completed sales at that exact grade and grading company, and you can click it to see the sales. This page is the whole method — including the parts we are still not happy with.

The problem with one price

A card does not have a price. A raw Gengar EX, a PSA 9, a PSA 10, a CGC Pristine 10 and a BGS Black Label are five different things that sell for five different amounts — on one card we track, $949, $8,700, $26,525, $2,454 and $37,000. Most price guides show you one number, or a formula: PSA 10 times 1.3 for a BGS 10, times 0.6 for a CGC 10. Those are guesses wearing decimals.

So we don't do that. We collect the actual sales, file each one to a card and a grade, and publish a separate median for every grade at every grading company that has enough of them.

Where the sales come from

• eBay completed listings — the bulk of the market — read every night.
• TCGplayer latest sales, matched by product id, for ungraded cards. That is where most raw cards actually change hands.
• Goldin auction results for the high end: Black Labels, vintage 10s.
• Fanatics Collect, Heritage, MySlabs and Pristine results, through 130point.

The same eBay listing seen twice is stored once. And where eBay shows a best-offer sale at its asking price, we store what the buyer actually paid — on the first batch we corrected, the accepted price averaged 17% under the ask.

How a sale is filed

The grading company has to be named in the listing. "PSA 10 potential" and "gem mint candidate" are raw cards, and they count as raw. Tiers above 10 are their own thing: CGC Pristine 10 is not a CGC 10, and a BGS Black Label is not a BGS 10 — on one card the Black Label sold for $8,000 against $3,500 for the gold label. Auction-house wording ("PSA NM-MT 8", "PSA PR 1") is read too.

Sales are set aside, with the reason recorded, when they are not the card: another language's printing filed under the English card, a card number the title contradicts, a lot or playset sold as one listing, a proxy, or a raw card described as damaged. They are never deleted; they just don't count.

How the number is made

• Median, never average. One mispriced sale cannot move a number.
• Windows that fit the grade. Ungraded: the last 90 days. Graded: the last 90 days when they hold five or more sales, the last year otherwise. A year's median lagged a moving card by half; the shorter window fixed it.
• An evidence bar. A price is published only with three or more sales inside its window. Fewer than that, the grade shows a dash — not a guess.
• Direction. The arrow beside a price compares the last 90 days with the period before, and only appears when the move is 5% or more and both sides have three sales.
• Inversions are held back. A lower grade priced above the next grade up is hidden until it has five sales of its own. With five, it's the market, and it shows.

Checked before it goes live

Every night the whole table is recomputed and compared with the last good run. A price that moved more than 60% on fewer than five sales, a price with too few sales, a price whose newest sale has fallen outside its window — any one of those stops the run, and nothing is published until a person has looked. This has happened three times; each time it was a correction working (a Japanese copy leaving an English bucket) and was accepted by hand.

Two readers read every listing title. Rules do the filing. A second, independent reader — a language model running on our own hardware — reads each title again and flags any it thinks was filed to the wrong card. Where the two disagree is where we look; its first pass found three rules we then fixed.

And we audit by hand. On 27 September 2026 we read 200 random sales behind published prices, one by one: the grade was right on all 200, the card on 197. The three misses were one pattern, now caught.

What we don't do yet

• Ungraded medians include lightly played copies alongside near mint, the way the raw market does. Damaged copies are set aside; played ones are not.
• Coverage is growing, not complete. 6,799 cards carry measured prices today, several thousand more each night; the rest still show the vendor estimate, marked est.
• The "should I grade it" line uses the population's gem rate, not your copy's. Your card's odds depend on its condition, which we haven't seen.
• Receipts are text for most sales. Photos are next.

See it on any card

Open a card page and click any price in the table. You'll get every sale it was made from — date, amount, listing — and a link to buy the same grade now. If a number looks wrong to you, the sales are right there to argue with. That is the whole point.

Method as of 30 September 2026. When it changes, this page changes.