How accurate are we?

How much of the game do we actually price? the denominator first, not the flattering number

254 of 280 cards (91%) carry a published price; 218 (78%) rest on 4+ completed sales and show as a plain number; 36 sit on 1–3 sales and show only with that qualifier; 0 are temporarily withheld while source provenance updates; 26 say exactly what is true: “No verified solds — tracking live.”

A site that quotes a number for every card this early is using asks or a model. We do neither — the cost is this denominator. Browse every card we track.

How fresh is the number? recency is a separate axis from sample depth — we never fold them into one score

Snapshot built 2026-09-17 from completed sales on eBay US. The sweep runs daily, and a build over 26 hours old puts a stale banner on every page. Each card wears its own recency badge: FRESH newest real sale within 14 days, STALE older than that, RECENCY UNKNOWN the marketplace never dated the sale, so we don't claim it's recent. A deeply sampled card can still be stale; you see both facts.

What we refuse to print, and why the part no asking-price site can copy

• A thin price, bare. 36 cards rest on 1–3 sales. They never appear as a plain number anywhere — not in tables, search, share images or structured data; only as “~$X — rough number, only a few sales”, or not at all.
• A number for a card nobody has sold. 26 cards have no verified completed sale. We could interpolate one in a minute; they say “tracking live” instead.
• A median we don't trust. 13 price series across 13 cards are flagged suspect — sales too dispersed for a median to be honest. Marked on the page, withheld from the weekly report.
• A price move we can't back. 0 series held enough dated, deduplicated sales in both windows to be eligible for a price move this week, and none survived the honesty gate. So the movers page says so, in those words. We would rather print nothing than a move we cannot back.
• Asking prices, ever, as a headline. A card listed at $200 that never sells at $200 is worth what it last sold for.
• Numbers we don't have the data for. Graded (PSA) values and pack expected-value are simply absent — not estimated, not "coming soon" with a placeholder figure — until the underlying sold data exists.

Every refusal is enforced in code — the build fails rather than render a bare thin price. Card charts list source-attributed sale evidence, with a public listing link on a dot when available.

How often are we wrong? head-to-head against the sites that show a price for everything

The test: freeze a basket of cards, record the number every site displays for each, then wait for the card's next verified completed sale and measure who was closer. Every scored round is published below — including the rounds we lose. The data format carries every round or none; there is no path that filters a loss out.

48 further snapshots are still awaiting their next verified sale; every one of them will be scored and published here.

Scoreboard

Read accuracy and coverage together: a site can look precise on the few cards it dares to price. Error columns stay "—" until rounds are scored — we do not print a number we cannot back, even about ourselves.

Every scored round newest first — losses included, plainly

The sale link is the receipt — the completed listing that settled the round. Competitor figures are cited as number, site and retrieval date only; we don't rehost or link their pages.

How the benchmark is run

• Our number: sold-based median (completed sales only) as published on the card page at snapshot time
• Competitor numbers: the single publicly displayed price on the competitor card page, captured manually (attended) and cited with URL + retrieval date; no page content copied
• Ground truth: the next verified completed sale for the card after the snapshot, from our receipt-linked sales ledger
• Scoring: absolute % error and signed bias vs that next sale; winner = closest; coverage = how often each side showed a number at all
• Honesty: every scored round is published, including rounds we lose; a missing number is a coverage miss, never hidden

Competitor prices are captured manually, on a dated attended pass — not crawled. Their numbers render in-browser and their data endpoints are closed to robots, and we honour robots directives: we don't crawl where a site says not to, the same courtesy we ask of others. A manual pass means competitor coverage can lag. A missing competitor number is recorded as an honest coverage miss for them, exactly as a missing number of ours is recorded against us.

The full record

• How we price — every rule between a raw sold listing and a published number, exclusions and filters included.
• Corrections — every change to a published number, dated, wrong-then-corrected included.
• Market overview — the shape of what we can price, and the tracked-value floor built only from confident prices.

Tell us we're wrong the fastest correction loop this site has

A price here disagrees with the market you're seeing? Say so: contact@palworldprices.com — name the card and, if you can, the sale we missed or should not have counted. Fixes are published on corrections.

The same address reaches the operator for rights-holder inquiries, including card images (shown under the publisher's fan-content policy). No form, no intermediary — it goes to a person.