Methodology

Every price on this site is a completed sale — a transaction that actually happened — never a seller's asking price. Here is exactly how we turn raw sold listings into a number, and where we refuse to show one.

What a price is

A card's headline is the median of completed sales in the freshest window that has enough of them. Asking prices (active listings) are not prices — a card listed at $200 that never sells at $200 is worth whatever it last sold for. Competitors that publish "prices" from active listings are showing you hope, not history.

Where the sales come from

Completed-sale sources in this snapshot: ebay-us-sold, ebay-us-sold-single, mercari-jp-sold, yahoo-jp-closed. JP marketplace figures are one aggregated sold feed (sources named for provenance, not counted as independent confirmations). US figures are eBay completed sales. Currency is converted at the ECB reference rate (160.24 JPY/USD, as of 2026-07-31); every USD figure is derived from the native amount and a dated rate, so it is re-priceable and never rests on a hidden constant.

How we filter

• Bootleg & non-card guard. Counterfeit code families and accessories/lots are dropped before anything is priced.
• Outlier fence. Sales outside 4× of the sample median are discarded — a fat-finger or a shill bid cannot drag the median.
• Spike quarantine. A single sale far above the trailing median is held back for days so the market can dispute a fake print before we ever publish it.
• Dedup. The same sale re-observed across scout runs is collapsed to one — we never manufacture sample depth.
• Retraction voiding. A "sold" listing that reappears active (non-payment / relist) is voided; it never settled, so it is not a price.

Confidence — how much to trust a number

Confidence measures sample depth, nothing else. It is separate from recency.

We show it as a 4-bar signal-meter — the bar count is sample depth, on a single colour, so confidence is never confused with the separate freshness axis or read as a good/bad "traffic light".

• HIGH — 10+ completed sales. Shown as a plain price.
• MEDIUM — 4–9 completed sales. Shown as a plain price.
• LOW — 1–3 sales. Never shown as a bare number — only as a qualified, thin-sample estimate.
• NONE — no verified sales. We say "no verified solds — tracking live" rather than invent a figure.

Recency is a separate FRESH / STALE / RECENCY UNKNOWN badge. A card can be deeply sampled yet stale (no recent sales) — we show both axes, never fold them into one.

Dates, honestly

When a marketplace gives a real sale time we say "sold on". When it only tells us when our scout saw the sale, we say "last verified" — never a fabricated sale date. We never claim price movement ("up 20% this week") from data whose dates are inferred.

Sold vs ask, and what we don't show

We would rather show less than imply a number we can't back. Grading value, pack expected-value, and a JP↔US arbitrage spread only appear once we have the underlying sold data — until then, they are simply absent, not guessed.

This page describes the same rules the price engine enforces in code; the public scorecard tracks whether those rules are actually producing accurate calls.