Deal Sniper
Deal Sniper ResearchUpdated 2026-08-28

How to read a card seller before you buy

Feedback is the anchor, not the answer — and the four things that matter most are the ones eBay's summary does not carry.

Key finding

Seller risk is a discount to apply, never a veto: a great card from a mediocre seller is still a great card, and the things that actually go wrong are mostly invisible in a feedback score.

Two listings, same card, same price. One seller has 99.8% positive feedback across 14,000 sales; the other has 97.2% across 31. Most buying advice stops there and tells you to take the first one. That is roughly right and almost entirely useless, because it answers a question that was never really in doubt and skips the ones that are: what condition is this card actually in, are those photographs of this card, and what happens if it arrives bent.

This site scores seller risk on every listing it ranks, so it is worth being explicit about what that number is built from — and, more importantly, what it cannot see.

Feedback percentage is a cliff, not a slope

The instinct is to read feedback as a continuous grade where 99% is a little better than 98% and much better than 96%. It is not. eBay feedback is overwhelmingly positive by construction — buyers who are mildly disappointed usually leave nothing at all — so the distribution is squashed against the ceiling and the interesting information is in a very narrow band near the top.

That is why the quality score behind this site's rankings is a step function rather than a curve:

Positive feedbackQuality scoreWhat it means
99.5% and up9.5ordinary for an established card seller
98% – 99.5%8.0normal; a few bad months or a rough patch
95% – 98%6.0a real pattern of unhappy buyers
below 95%2.5something is wrong, and it is not variance

The gap between 99.6% and 99.9% is noise. The gap between 97% and 99% is a different seller. Reading the third decimal place is a way of feeling rigorous while learning nothing.

A small sale count is unproven, not bad

A seller with nine sales and 100% feedback has not demonstrated anything — nine buyers is not a sample, and 100% of nine is the most fragile statistic in commerce. So a feedback count under ten caps the quality score at 5 no matter how perfect the percentage looks, and a count under 25 adds a full point of risk on top.

Note the framing, because it matters: unproven is not the same as bad. Plenty of good cards are sold by people clearing a collection who have never sold anything before. The right response is a smaller first purchase and a closer look at the photographs, not avoidance.

An unstated condition costs more than a mediocre rating

This is the part that surprises people. In the risk model a listing with no stated condition adds 1.2 points, and a listing with no photograph of the actual card adds 1.5 — while the whole drop from 99.5% feedback to 98% costs 1.5, and the drop from 98% to 96% costs 2.

The reasoning is that feedback describes the seller and the missing fields describe this card. A conscientious seller with a spotless record can still ship you a card with a soft corner they genuinely did not notice, and if the listing never stated a condition and never showed the card, you have no claim and no evidence. You did not buy a card. You bought a description that was never written.

  • No condition stated — +1.2. Common on bulk listings and on sellers who list fast. Not dishonest; just undefined.
  • Condition stated but unreadable — +0.6. “Mint” and “NM-MT” are opinions, not grades, and they cannot be placed on any scale that means the same thing twice.
  • No photograph of the actual card — +1.5, the single largest factor here. A stock image on a raw card is not a listing, it is a lottery ticket.

The four things the score cannot see

This is the honest limit, and the reason the number is presented as a discount rather than a verdict. The eBay data this site ingests carries a feedback percentage, a feedback count, a condition string and one image URL. That is all it carries. Four things that matter at least as much are simply not in it:

  • the return policy;
  • the quality of the description;
  • whether the photographs are the actual card or stock images;
  • the seller's history with cards like this one.

Inventing proxies for those would have been easy and would have made the score look more complete than it is. Instead they are listed on screen as unavailable, so a 2.1 reads as “low risk on four real signals” rather than “audited and cleared”. The last one is worth dwelling on: a seller with 40,000 sales of phone cases has excellent feedback and no idea how to pack a card. Volume is not expertise.

What to actually do with it

The bands are deliberately coarse — low up to 2.5, moderate to 4.5, elevated to 6.5, high above that — because the number does not deserve finer resolution than that, and because the decision it feeds is coarse too.

Seller risk is a small input to the investment score by design. A great card from a mediocre seller is still a great card; the risk is a discount you apply to what you are willing to pay, not a veto. What it must never become is a tiebreaker dressed up as a judgement — “I passed because the seller was 98.2%” is a story you tell yourself after deciding for other reasons.

Concretely: on an elevated-risk listing, ask for a photograph of the corners and the surface at an angle before you offer. A seller who sends one has just converted your largest unknown into evidence, and a seller who does not has told you something the feedback percentage never could. If you are working out what to offer in the first place, the best offer calculator handles the arithmetic.

The rule underneath this one

Feedback tells you about the seller. Photographs and a stated condition tell you about the card. When they disagree, believe the card — you are not buying the seller's record, you are buying one specific object that has to survive the post.

Basis & limits

What this is built on. The rules and figures this project's own identity and valuation engines enforce, plus the domain research behind them.

Where it stops. This is an explainer, not a study: it carries no sample size and makes no forecast. Figures that move in the real world — grading fees, print runs, marketplace behaviour — can date it; the updated line above marks the last material revision.

Methods are documented on the methodology page; sources and their limits on trust & data sources.

More on doing the deal

All doing the deal guides →

Keep reading

See this applied to real prospects: every player page shows live listings, sold evidence and the parallel ladder for one player, and the market board ranks what is mispriced right now.