How buycards.ai ranks sports-card listings

buycards.ai is a sports-card search and scouting tool built on top of public eBay listings. It is a filter + ranker + dashboard, not a marketplace. We do not buy, bid, sell, or message sellers on the user's behalf, and we do not capture marketplace credentials.

The site has three main surfaces: free-text search with quality ranking, discovery rails (ending soon, below median, low-numbered, top/small sellers), and Buyer Scout — a personalized daily shortlist with honest action labels.

Before ranking, every candidate runs through a quality ranker that hard-drops listings that aren't real card inventory:

• Repacks, hot packs, mystery boxes, sealed product
• Lots ("5-card lot," "lot of 10," wholesale)
• Reprints, replicas, fakes, custom/proxy cards
• Digital cards (Topps Now Digital, NFT, Panini Blockchain)
• Off-brand products with thin resale liquidity (Leaf, Wild Card, Sage, Onyx, Press Pass, Pro Set)
• Listings whose claimed serial is contradicted by an audited checklist ("/25" on a verified-unnumbered parallel)
• Non-card line items (signed photos, jerseys, plaques) that slip in via the broader sports-memorabilia categories

Clean listing is not the same as good buy. A clean listing is one we believe is a real card from a real product family — not a recommendation to purchase.

Each scout candidate carries a typed evidence object with four dimensions:

1. Identity confidence — none / low / medium / high. Driven by parsed year, set, product, grade, parallel, serial, and a registry-backed player name (canonical "Marvin Harrison Jr.", not a generic "big-name" token).
2. Rarity confidence — none / title-only / checklist-backed / contradicted. Checklist-backed requires an exact product + parallel phrase match against an audited print-run table, with a disambiguator defense so "Topps Chrome Sapphire" never matches "Topps Chrome". Contradicted fires when a seller claims a serial the checklist refutes.
3. Active-comp confidence — none (sample < 3) / low (3-7) / medium (≥ 8). High is reserved for a future sold-comp adapter and is never emitted today.
4. Price confidence — none / active-ask only. Sold-comp slots are reserved.

The four dimensions compose into a single 0–100 evidenceScore that drives ranking. Taste fit is a tiebreaker, not the primary signal.

A serial number in a listing title is not proof of rarity. Sellers regularly write "/25" on cards that are actually unnumbered to ride the rarity-keyword bump. Our rarity model is deliberately conservative:

• checklist-backed requires the (year, product, parallel) triple to match an audited row exactly.
• title-only is the most we'll claim when the title says "/25" but no audited row confirms it.
• contradicted fires when the title's claim conflicts with an audited row — the card is demoted to inspect-only.
• POP/10 and similar tokens are not parsed as serial numbers (a load-bearing defense).

When Scout shows a card as "below active asks (n=15)", it means the candidate price is below the cohort of currently listed comparable cards. Active asks are seller expectations, not cleared sales. They are useful as a directional signal but they are not a market price.

Sold comps would tell us what cards actually transact for. We do not have sold-comp data wired today, and we do not pretend to. Every Scout card carries a persistent Sold comps: not connected disclosure for this reason. When that adapter ships (likely eBay Marketplace Insights or user CSV import), the system will surface sold-based confidence explicitly.

Every Scout card carries one of five action labels. They describe what the buyer should DO, never what we predict will happen:

Inspect

Strong card, strong fit, no severe risks. Worth opening immediately.

Bid candidate

Auction closing soon with strong fit + quality. Set max bid manually in eBay; we never bid for you.

Watch

Interesting but a meaningful blocker (comp confidence missing, price uncertain, auction earlier in cycle).

Uncertain

Mixed signals. A grail card with a severe risk demotes here rather than to Inspect.

Pass

Hard reject, severe risk, or fit below threshold.

Scout is an inspection workflow, not financial advice.

Raw eBay returns 1,500+ listings for "Wembanyama PSA 10" on a typical day, mixed with repacks, sealed product, fakes, off-brand, and misleading rarity claims. buycards.ai surfaces a short list of clean listings ranked by source-backed evidence, with explicit confidence labels on the rarity and the comp cohort. When the evidence is too weak to support a number, we withhold the number instead of guessing.

The win is not "deals." The win is less time inspecting junk.