How these prices are collected

Where the data comes from

Prices come from completed Yahoo! JAPAN Auctions listings — auctions that ended in a sale. Each record carries the closing price, the end date, the seller-stated condition, how many days the item was listed, and how many bids it drew.

This means the figures are sold prices, not asking prices. That distinction matters more than it sounds: most cross-market price tools compare a Japanese asking price against a Western sold price, which systematically overstates the gap.

How a SKU is identified

Product lines that publish an official release number — figma and Nendoroid — are keyed on that number. figma 214 is one product, permanently. We do not guess identity from listing titles.

Release numbers are matched on their numeric value, so a listing written figma 063 and one written figma 63 land on the same page rather than splitting one product into two.

Product names, series names, manufacturers, release dates and sculptors come from the manufacturer's own catalogue wherever we have it — taken from their English pages, so the names here are the ones the manufacturer itself uses in English. The Japanese name is shown alongside, because that is what you need to search Yahoo! Auctions or Mercari for the actual item. Where we have no catalogue entry, the name shown is derived from what sellers consistently wrote across that SKU's listings — useful, but clearly a fallback, and it stays in Japanese.

Lines without a numbering system are not given fabricated numbers. Two real cases we hit and corrected: フィギュアーツ 555, where 555 is the title of Kamen Rider 555, and figma 1/12, where the 1 is a scale, not a release number. Both had merged unrelated products into one page before we caught them.

What gets excluded, and why

• Lots and bundles — "3 点セット", "まとめ売り". A per-unit price derived from a bundle is not comparable to a single-item sale.
• Third-party accessories — doll clothes, pouches, stickers, bags. These are not the figure.
• Replicas and out-of-category items — listings marked レプリカ, and items that merely share a keyword (a wah pedal called "Cry Baby", a mall brand called モーリーファンタジー).

Exclusions are applied per row, not per keyword, because search relevance decays down the result pages: the first page can look clean while page 20 has drifted entirely off-topic.

Where retail prices come from

Where a page shows a retail price, it is the manufacturer's own published price for that release — taken from Good Smile Company's and Max Factory's product catalogues, matched to the sold data by official release number. Every such page links back to the manufacturer's page for that product so the figure can be checked at the source.

Retail prices are shown tax-inclusive, the same basis as the auction closing prices they are compared against. Premium over retail is simply the median sold price measured against that number — no adjustment for inflation, exchange rates, or re-releases is applied.

Not every page has one. Nendoroids released after February 2024 are not in a catalogue we are permitted to crawl, so those pages show no retail price at all rather than an estimated one.

Where the US price and stock status come from

Some pages also show an official US price and whether the item can still be bought new. Both come from Good Smile Company's own US store — the dollar figure is what the manufacturer itself charges there, not a currency conversion of the Japanese price. A store listing is attached to a page only when its title aligns exactly with the manufacturer's catalogue entry for that release number; anything short of an exact match is dropped rather than guessed.

Stock status is dated: each page says as of the day the store was last checked (refreshed daily), because a "sold out" without a date is a claim that goes stale silently. The US store carries its current catalogue only, so most older releases have no listing there — those pages simply omit the block.

How the numbers are computed

• Median, not mean — a single ¥1 opening bid or one collector's overpay should not move the headline.
• Trend compares the median of the earlier half of recorded sales against the later half. Moves under 8% are reported as flat. Every page also states how the split fell — "3 earlier sales against 4 later" — because a percentage drawn from one sale against two means something very different from one drawn from eight against nine. We give you the number and what it rests on; the weight you put on it is your call.
• Sales per month is the recorded count divided by the observed window, and the window is always printed next to it. Three sales across 29 days and seventeen across 169 days can both read as "about 3 a month" — the span tells you which one you are looking at.
• Sealed premium is the gap between the median new-condition sale and the median used sale, shown only when both sides have at least two sales.
• Percentile ranks place a figure against others in the same product line. The reference group is limited to releases with five or more recorded sales — not to filter what you see, but so the percentile has a stable thing to be a percentile of. Every ranked page states the size of that group and its own sale count.
• Charts are scatter plots, not lines. Sales are discrete events; a connecting line would draw prices that never existed.

Known limits

• One venue. Yahoo! JAPAN Auctions is a large slice of the Japanese secondary market, not all of it. Mercari, Suruga-ya and dealer sales are not included.
• A rolling window. Yahoo publishes roughly 180 days of completed auctions; the archive deepens going forward but cannot be back-filled.
• Sample depth varies a lot. Some figures have twenty recorded sales, many have three. The count is printed on every figure and next to every derived number, so you can weigh them yourself. Pages under five sales are published in full but marked noindex, which keeps them out of search results without hiding anything from anyone who is here.
• Partial retail coverage. Premium over retail appears only where the manufacturer published a price we can cite; other pages omit it rather than estimate it.

Corrections

Data bugs get fixed at the source and the whole derived layer is rebuilt, so a correction applies retroactively to every page rather than leaving old pages stale. If a page looks wrong, it probably is — and the raw records behind it are kept precisely so it can be rebuilt.