Reading a Comp: What the Median Hides

A comp is a comparison. That is all it ever was. Somewhere along the way the hobby started treating it as a verdict — a single dollar figure that tells you what a card "is worth." That is a category error, and it costs collectors money every day.

A price is not a fact about a card. It is a record of one transaction between two people under one set of conditions. String enough of those records together and you can estimate value. Read any single one as gospel and you are guessing with extra steps.

Dealers know this. They do not look at a number; they look at the shape of the data behind it. This piece is about that shape — what it tells you, and what the tidy summary on top is quietly leaving out.

The "last sold" trap

The most dangerous number in the hobby is the last sale.

It is the easiest to find, the easiest to cite, and the easiest to anchor on. It is also a sample size of one. The last sale tells you what one specific copy did, at one specific moment, in front of whatever pool of bidders happened to be awake that night.

Auctions are not vending machines. The same card sold twice in the same week can swing wildly because:

Two determined bidders showed up for one listing and not the other.
The first listing ended on a Sunday night; the second ended on a Tuesday afternoon.
One had a clean scan and an honest title. The other buried the card on page nine of search.
One seller had feedback in the thousands. The other had eleven.

None of that is signal about the card. It is signal about the sale. The last-sold price packages all of it into a single figure and presents it with false confidence.

Treat the last sale as one data point. Never as the answer.

Median is a summary, not a truth

The median is better than the last sale. It is the middle value, so it shrugs off the single insane outlier in a way an average never could. That is exactly why it is the default summary statistic in any serious comp tool, including ours.

But a median is only as honest as the pile underneath it.

A median computed over three sales is a coin flip wearing a lab coat. A median over forty recent sales of the same grade is something you can lean on. The number looks identical on the page — $X, bold, confident — but the two are not the same kind of object, and a tool that shows you one without the other is hiding the part that matters.

Ask two questions of every median before you trust it:

How many sales is it built on?
How recent are they?

If a comp tool cannot answer both at a glance, it is not a comp tool. It is a number generator.

Sample size and recency are the whole game

These two variables decide whether a comp is worth anything.

Sample size. A handful of sales describes a handful of sales. The fewer the comps, the more each one bends the result, and the more a single motivated buyer distorts the picture. Thin data is not wrong — it is just uncertain, and that uncertainty has to travel with the number, not get rounded away.

Recency. Card markets move. A six-month-old median can describe a world that no longer exists — a player got hurt, a set got re-popular, a grading population doubled. Old comps are not data; they are history. Useful for telling you where the market has been, useless for telling you where it is.

The trap is that recency and sample size fight each other. Narrow the window to last week and you get current prices on three sales. Widen it to two years and you get four hundred sales describing a market that has since moved. There is no clean answer. The honest move is to show the collector both the window and the count, and let them weigh the trade-off themselves.

A median with no window and no count is a magic trick.

Raw and graded do not belong in the same pile

This is the single most common way comps get poisoned, and it is almost always silent.

A raw card and a graded card are different products. A PSA 9 and a raw copy of the same card are not two prices for one thing — they are two things. Blend their sales into one median and you get a number that describes neither: dragged up by the slabs, dragged down by the raw, accurate for nothing.

It gets worse, because raw is not even one category:

A raw card a thoughtful seller calls "NM" might grade a 7, an 8, or come back with a crease nobody photographed.
"Raw" spans pack-fresh to played-with-by-a-child, and the price gap between those ends is enormous.
Two raw copies at wildly different conditions sell for wildly different prices — and both are correct.

So a "raw median" is doing real work to be meaningful, and a median that quietly mixes raw with graded is doing real damage. The grades have to be separated. Within raw, condition has to be acknowledged as a range, not a point. Anything else is blending milk into the gas tank and calling it fuel.

What actually pollutes a median

Even within a single grade, raw comps are full of sales that should never count. A good comp tool spends most of its effort throwing these out:

Autographed copies. An on-card auto is a different card with a different market. It does not belong in the base comp.
Altered or "custom" examples. Trimmed, recolored, cleaned, or doctored cards trade in their own murky lane. Including them poisons the honest copies.
Lot sales. "47 vintage commons" closing at one price tells you nothing about any single card in the box. Lots are the number-one quiet contaminant of cheaper-card comps.
Damaged copies sold as project cards. A creased, paper-loss example is a real sale, but it is not your card's comp unless your card is also creased.
Wrong-card matches. Same player, wrong year, wrong parallel, wrong set. Title-matching is sloppy by nature, and one wrong-variant sale can swing a thin comp by a fortune.
Shill and relisted noise. Retracted bids, cancelled sales, and the same item cycling through three failed listings.

Every one of these is a real transaction. None of them describes the card in front of you. The work of a comp is not collecting sales — it is deciding which sales count. That filtering is the product. The number is just the receipt.

Price guide ≠ realized sale

Hold two ideas apart, because the hobby constantly mashes them together.

A price-guide value is an opinion. Someone — a guide, an algorithm, a "book" — looked at the market and published an estimate. It is smoothed, it is often stale, and it is frequently aspirational. It is what someone thinks the card should bring.

A realized sale is what someone actually paid. Money changed hands. A real buyer met a real seller at a real price.

Guides have their place. They are fine for ballparking a card you have never traded, or for sanity-checking the rough tier. But a guide value is not a comp, and treating it as one is how people end up overpaying with total confidence — "but the book says." The book is not a counterparty. The book never bought anything.

When the guide and the realized sales disagree, the realized sales win. Always. Reality outranks opinion.

How to triangulate — and what an honest tool owes you

No single source is complete. Each one sees part of the market and misses the rest. The way through is triangulation: read several sources, understand each one's bias, and let them check each other.

Auction sales show real demand under competition, but they swing on who showed up.
Fixed-price and best-offer sales show what sellers will accept, often above true clearing price.
Grade-by-grade value services show structure across conditions, but lag the live market.
Population reports are not prices, but they tell you how scarce a grade actually is — which is half of why a price exists.

When several independent sources cluster, you have a real number. When they scatter, you have a thin or moving market, and the correct read is "I do not know yet" — which is itself worth knowing.

So here is the standard. An honest comp tool does not hand you one bold figure and a clean chart. It shows you:

The count and the time window behind every number.
Raw and graded kept separate, never blended.
The individual sales — so you can see the spread and the outliers, not just the summary.
The junk it excluded — autos, lots, alterations, wrong matches — and the fact that it excluded them.
An honest "not enough data" when the data is not there, instead of inventing confidence.

A median is a starting point. The dealer's edge has never been knowing the number — it is knowing what the number is made of. The whole job of a comp is to show you that, instead of hiding it. Everything else is decoration.