Research & methodology

Two gates, not a weighted average

A conviction buy needs two independent things to be true at once: the card is worth owning, and this listing is underpriced against credible evidence. They are gates, not ingredients — a 60% discount cannot rescue a card nobody wants, and an important card at full retail is not a deal. When a listing misses, we name which gate it missed rather than showing a number that hides it. PASS is the resting state, and zero buys is a legitimate answer to a day's market.

The card gate

Collector significance is curated, never inferred from price. Each card carries separate dimensions — canonical rookie importance, historical significance, hobby recognition, scarcity, liquidity, population risk — each with its basis, its source and the date it was reviewed. A card with no review is unrated, and unrated is not zero: it can be researched, browsed and watched. What it cannot be is promoted. An unrated card is capped below the top quality band, so no discount can turn one into a conviction buy. The gate needs a curated read, a card score ≥ 80, and one of the markets this site is about (core or collector) — or a curator's explicit iconic/high rating, which outranks our own taxonomy in both directions.

The value gate, and the evidence ladder

Fair values are recency-weighted medians over confirmed transactions only — imports and verified sales. A listing that merely disappeared from eBay is never counted as a sale; it powers a clearly-labeled modeled estimate at most. Values are shown as ranges because a point estimate would claim precision the evidence does not carry. The ladder runs fresh exact sales → older exact sales → base × a measured parallel premium → cross-rung triangulation → tracked disappearances → a display-only shadow estimate, and every value says which rung it came from. Confidence: HIGH is 6+ confirmed comps within 90 days, MEDIUM 3–5, LOW 1–2, and below that we show no value at all. An exact comp must match the year, product, card number, print run, the grade (grader included), the parallel and the physical form — a raw card is never priced off a PSA 10, and a Topps Now #5 sale never prices a #23.

The gate itself: a confirmed valuation at medium-or-better confidence, and an ask at least 20% under it — 15% when a curator has rated the card iconic or high. LOW confidence and modeled values can never mint a buy, at any discount. That is a rule with teeth, not a preference: the whole point of a modeled number is that nobody witnessed it.

The Index Score

45% card significance + 30% verified value + 15% market quality + 10% market timing, on a 0–100 scale. It ranks; the gates above decide. Unknown timing or market data redistributes its weight over what is known. An unknown price does not: a card whose price cannot be checked is capped at 55 and shown under “Needs pricing” — an important card at an unmeasurable price is homework, not an opportunity. Conviction also needs the composite itself at 82+.

Tiers

CONVICTION BUY requires a live-verified listing, a curated significance read, card ≥ 85, value ≥ 75, and confirmed comps at medium-or-better confidence — with high-confidence evidence required at the premium shelf. STRONG VALUE is 84+, GOOD VALUE 77+, WATCH 70+, SPECULATIVE 60+. Aftermarket signatures, reprints and lots can never be recommended at any price, and neither can a card a curator has rated low importance. Every card also states the confidence-adjusted edge: the raw discount haircut by the strength of the evidence behind it.

What the model refuses to do

No invented probabilities — the $1K read is a ranking against today's board, not a forecast. No dollar gaps quoted off a LOW valuation. No score without an explanation: every number carries its component breakdown. No recommendation on unverified listings: clicking through to eBay re-verifies the exact item first. Freshness and urgency reorder the board; they never create a deal. And no commission input anywhere in scoring — see how we make money.