Population Screener Methodology

What the Screener Measures

For each card the screener reports:

Annualized population growth. How fast the number of professionally graded copies is increasing, expressed as a yearly rate. Low is the interesting outcome here: a low rate means the slab supply is barely moving, which is what genuine, durable scarcity looks like. This is the reverse of most metrics on the site, and the screener is deliberately styled so a low number does not read as a bad one.

Gem rate. The share of the graded population that came back at the top grade. A low gem rate means a card is genuinely hard to grade well, which is what makes its top-grade copies scarce relative to the total.

Population, split two ways. The total across all grades, and the top-grade count on its own. A card can have a fast-growing total and a nearly static top-grade population, and for most purposes the second is the more interesting number.

The evidence behind each rate. How many population reports the fit used and how long a period they span, summarized as a confidence tier. More reports over a longer window read as more confidence, and the tier rises on its own as the weekly reports accumulate.

Why a Fitted Curve, Not Two Dates

The obvious way to measure growth is to compare the latest population against one from a year ago. We do not do that, because it throws away almost all the evidence and swings wildly on the luck of two readings.

Population reports arrive weekly in theory and irregularly in practice: a crawl can be skipped, a set is onboarded partway through a quarter, and an occasional capture is incomplete. Under those conditions a two-point comparison is hostage to whichever two points it happened to pick, and one unlucky endpoint distorts the entire result.

Instead we fit a growth curve through every usable snapshot we hold for the card and annualize its slope. Fitting handles irregular spacing naturally, because the curve is fitted against actual elapsed time rather than against snapshot number. It uses all the evidence rather than two readings of it. And where a card genuinely only has two readings, the fit reduces to exactly the simple comparison anyway - which is why two readings no longer buy a rate at all. A line through two points passes exactly through both by construction, so it scores as a flawless fit while resting on the thinnest evidence in the stream. That is not a fit; it is the two-point comparison in better clothes, and we would rather say so than sell it.

The cost of that flexibility is that a fit can be well supported or shaky, which is why every rate is published with the number of reports behind it and the period they span.

Six Guards That Keep a Rate Honest

Population data is messier than it looks, and a naive fit produces confident nonsense. Six guards address the failure modes we actually observed in live data.

Attribution flips are separated from real growth. A crawl can occasionally capture a card under a different attribution than the week before, or capture only some of its grade buckets. That shows up not as drift but as the recorded population flipping between two stable levels: a real card's history reading 82, 104, 82, 1201, 1201, 1202, 84, 84, 1222 is two regimes interleaved in time, not a card that grew fifteenfold and shrank back. So the observations are grouped by level and the fit uses only the group the most recent reading confirms, which is also the population the screener displays. This replaced an earlier rule that simply discarded any decrease, which sounds right (populations never shrink) but was wrong about the data: it walked a path across both regimes, could not see an undercount sitting at the start of a series, and biased every rate upward by only ever deleting downward moves.

Zeroes are excluded from the fit. A grade going from zero to some number is a grade first appearing, not exponential growth of an existing stock. Those observations are left out of the curve rather than fudged into it, which matters most on exactly the low-population cards this screener exists to find. They still count toward the raw change in copies.

Too short a history publishes nothing. Below a minimum span the fit is not published at all. The card shows how long it has been tracked instead, so a recently onboarded set reads as accruing history rather than as having a rate we invented.

Implausible rates are withheld, not clamped. A fit that comes out extreme, whether far too fast to be real or implying the population is shrinking, is an artifact the earlier guards did not catch. A graded population cannot shrink: cards get slabbed, never un-slabbed. Rather than clamping such a fit to a boundary and presenting it as a measurement, we publish no rate. Everything published is therefore a non-negative number we are prepared to stand behind, which also keeps the projections and the years-to-double figures finite and meaningful.

A rate needs at least three population reports. Two readings produce a line that passes exactly through both of them, which is the simple latest-versus-earliest comparison wearing the appearance of a perfect fit. That is the worst of both: the thinnest evidence in the stream scored as the tidiest result, and rated for confidence on that basis. Below three reports we publish no rate.

A rate needs a population big enough to carry a percentage. Below about twenty graded copies the arithmetic still works and the answer still means nothing: on a card with three copies, one new grade is a move of several hundred percent a year, so a board ranked on the rate is really ranked on one divided by the population. The same reasoning withholds the gem rate below about fifty graded copies, where a single lucky submission moves it by whole percentage points.

Those two guards are also why a card that has not moved at all is now reported as no new slabs rather than as a growth rate of zero. A flat series has no variation for a curve to explain, so the usual measure of fit quality is undefined for it rather than perfect; treating it as perfect is what let cards that had gained nothing carry the highest confidence on the board. Where the population is real and the history is long enough, that finding is stated plainly, along with how long it has held.

How to Read a Result

A few things worth holding in mind when you scan the table.

Low growth is the signal. The default ranking puts the slowest-growing populations first, because those are the cards whose scarcity is holding. A high rate is not a bad card, it is a card whose supply is still being discovered.

Growth and value are separate questions. A card can have flat population and no market interest at all. The screener answers the supply half; the value filters exist so you can combine it with the demand half rather than mistake one for the other.

A confidence tier is about the evidence, not the card. It reflects how many snapshots, over how long, and how cleanly they sit on a curve. A low tier means our measurement is thin, not that the card is bad.

Cards with no estimable rate are hidden by default. Rows where the guards withheld a rate are excluded unless you ask for them, with the count shown so you know how many there are. A screener padded with rows that have no number is the data dump this tool exists not to be.

"No new slabs" is an answer, not a gap. It means the card has a real graded population and gained no copies across every report we hold, and the row says how long that has been true. It is usually the most interesting line on the page. "Not enough data yet" is the opposite: we cannot measure the card at all, and the row names which of the three reasons applies. Those rows are hidden by default and counted underneath the table.

Limitations

Known limits:
  - It measures PSA supply. The rates describe copies graded by PSA, which is where our population history is deepest. Other companies' populations are tracked but do not feed these rates.
  - It is backward-looking. A fitted rate describes what has happened. Grading submissions respond to price, hype, and promotional rates, so a quiet card can start moving quickly with no warning in this data.
  - Population is grading events, not surviving copies. Resubmissions of the same physical card are counted again, so growth includes some churn that is not new supply.
  - Population reports update on the graders' cadence, not ours, so recent weeks can be sparse.
  - Projections are arithmetic, not forecasts. Where a projected population or a years-to-double figure is shown, it is the fitted rate carried forward mechanically. It assumes the current rate continues, which is exactly the assumption the point above says to distrust.