How we score

Live listings come from eBay's official APIs. A listing counts as live only after an adapter has seen the individual item — nothing is ranked off a search-page glimpse alone, and clicking through re-verifies the exact item.

Sold prices come from confirmed transactions: provider-reported sales and sales you record yourself. Each comp carries its provenance, and comps of unknown provenance never price anything.

Player data comes from a curated prospect roster with stats, level, age and position. A seller's listing title can never establish who a card shows — when the title doesn't match a rostered player, the card says so instead of guessing.

A sale is only evidence about a card if it was the same card: same player, year, product, parallel, serial run, and grade. We parse each title into those fields and score how sure we are (the "Identified %" on every card). When identity is not settled, the card is marked and excluded from exact-comp pricing — a wrong match would silently price one card off another's market, which is worse than admitting uncertainty.

Confirmed fair value is built from actual sales of the exact card: a range and midpoint after outlier filtering, with a confidence level (high / medium / low) that reflects how many recent sales back it.

Modeled value is labeled arithmetic from related evidence — adjacent grades, the parallel ladder, comparable players. It is always named as modeled and never dressed up as market value.

Absent is an honest state. When nobody has sold the exact card recently and no defensible model applies, the value cell shows a dash and the reason. We never zero-fill, never average unrelated cards, and never let an asking price stand in for a sale — asks are shown as live-market context, clearly apart from sold-based numbers.

Player (0–10): prospect quality from stats, age relative to level, and trajectory. Unmatched player = no score, not a neutral one.

Card (0–10): how strong this exact collectible vehicle is — 1st Bowman status, certified autograph, parallel scarcity, serial run. Price-independent.

Investment (0–100): how desirable owning this card is at a reasonable market price. Price-blind — a discount cannot inflate it.

Deal (0–100): how attractive this listing is against its own market. Player-blind — a star cannot inflate it. Unpriced cards have no deal score at all.

Deal Score (0–100): the composite that ranks the board. The words beside scores (STRONG PROSPECT, CONVICTION, and so on) are re-expressions of the same numbers at the same thresholds the gates use — a word can never disagree with the figure beside it.

BUY appears only when every gate passes at once: verified live, a core vehicle, player and card above their bars, a confirmed-sales discount past the bar (steeper for ordinary cards than elite ones), acceptable condition, product age, and seller quality. When a card misses, the app shows which gates failed and what would clear them — that is the "what would make this a buy" list, and the buy zone is the same arithmetic run backwards.

Research completeness (0–100) measures how much we know about an exact opportunity — identity certainty, sold evidence, live-market visibility, player data, verification recency. A great card can score low (nobody has researched it) and a mediocre card can score high (we know everything and it is still mediocre). Do not read it as a recommendation.

No invented numbers: unknown values print as dashes with reasons. No future-return promises: scenario figures are labeled scenarios. No affiliate influence: we may earn a commission when you buy through our links, and that commission appears in no scoring path — how we make money. No cherry-picking: the track record freezes every call, winners and losers alike.

The limitations are real: we see the market we scanned, not the whole market; thin evidence stays thin however good the card; and prospect outcomes are genuinely uncertain — no score changes that.