A Population Report Counts Submissions. It Was Never a Census of the Cards That Survived.

PSA's population count for Base Set Unlimited Charizard #4 — 487 Gem Mint 10s among 101,739 submissions — is real and often quoted as a rarity signal. It cannot be one: a self-selected sample of paid submissions is not a census, no matter how large it grows.

Four hundred eighty-seven Gem Mint 10s among 101,739 total submissions is a real count of exactly this: boxes scanned in. It was never a count of Charizards. — Image prompt and art direction by Brecht Corbeel; image generated to that direction.

Abstract

PSA's population report for the 1999 Base Set Unlimited Charizard #4 shows 487 Gem Mint 10s among 101,739 total submissions, a number repeated across pricing guides as a rarity signal. This article argues the number cannot do that job, and explains exactly why: submission is a voluntary, fee-paying act correlated with the owner's own belief about a card's condition and value, which makes the resulting count a textbook self-selected sample rather than a random draw from the card's true surviving stock. Drawing on PSA's and CGC's own printed disclaimer that population data is "not an indicator of value or rarity," on the statistical logic of sample-selection bias, and on the one Pokémon card whose print run is actually known — the 39-copy 1998 Pikachu Illustrator, whose own "known to exist" figure has already been revised upward once — the piece separates what a submission count can support (a lower bound on cards that once existed, and a real signal about current submission flow) from what it cannot (an estimate of the cards that remain), and proposes a one-line fix: report the number as submissions, never as survivors.

The Number Everyone Quotes Cannot Even Agree With Itself

Four hundred eighty-seven. That is the count of PSA Gem Mint 10s currently attached to the 1999 Base Set Unlimited Charizard #4/102, out of 101,739 total submissions graded — a gem rate repeated across pricing guides and forum threads as though it were a fact about the card rather than a fact about PSA’s intake window [3]. I went to check it against PSA’s own live population page directly, the way a number this load-bearing deserves [1]. PSA’s population search blocked the request outright and returned an HTTP 403 rather than a page. That is a small, almost comic failure, and it is also the first data point in this article: the number everyone quotes is not, in practice, a number most people who quote it have actually seen on PSA’s own site. It has passed through at least one relay before it reached them.

It gets stranger once you go looking for a second relay. Pikawiz’s own tracker for the same card, pulled independently in this same research pass, reports 92,856 total submissions and 489 PSA-10s [4]. Put the two side by side, with the gem rate worked out to two decimal places rather than the rounder figure usually quoted:

• Relay — Snapshot — Total submissions — PSA-10s — Gem rate
• CardboardSniper, via a GemRate-sourced chain [3] — Described as July 2026 — 101,739 — 487 — 0.48%
• Pikawiz’s own population tracker [4] — Undated, retrieved 2026-09-07 — 92,856 — 489 — 0.53%

Two aggregators, drawing on the same underlying institution’s live count, disagree by nearly 9,000 submissions on the denominator and by two on the numerator. Neither disagreement is really the subject of this piece — pop reports update continuously and different relays cache at different moments, so some drift is expected and is not, on its own, evidence of anything sinister. I state it here for a narrower reason: whichever of these two numbers, or whatever number PSA’s page shows on the day you read this, turns out closer to correct, the argument below applies to it unchanged. This is not an article about whether 487 or 489 is the right PSA-10 count. It is an article about what a PSA-10 count — any PSA-10 count, at any moment — is actually a measurement of.

Even “The Charizard” Is Three Different Populations Wearing One Name

Before getting to what the number can and cannot support, it is worth being precise about which number is even in play, because casual references to “the population of the Base Set Charizard” already conflate three separate cards that PSA and CGC grade, count, and report as entirely distinct census lines. The 1999 Base Set print run for Charizard #4/102 exists in three physically different states — 1st Edition (a shooting-star stamp on the left edge, printed first and in the smallest volume), Shadowless (an early Unlimited print missing the drop-shadow around the illustration box, printed in a narrow window before the shadow was added to the print plates), and Unlimited (the shadowed, mass-market print that ran for years afterward) — and PSA’s own population search returns three separate totals for these, not one combined figure. This article is about the Unlimited print specifically, the 487-of-101,739 or 489-of-92,856 figures discussed above. The 1st Edition print, tracked as its own line, currently shows a PSA-10 population of 125 by one 2024 hobby guide’s count — a number close to a quarter the size of either Unlimited snapshot’s PSA-10 count, for a card most casual references simply call “the Charizard” without ever specifying which print they mean [11].

That ambiguity sits underneath everything else this piece argues, and it would exist even if submission counts were a perfect census of whatever they counted. A number that is already unclear about which physical print run it describes cannot become clear about how many total copies survive just because the print-run question gets resolved separately. The two problems are independent of each other, and both have to be settled before “the population of the Charizard” means anything precise enough to reason from at all.

PSA and CGC Already Printed the Correction

Go to CGC’s own population-report page for its Pokémon TCG census and the correction is already sitting there in the site’s own boilerplate, not buried in a methodology PDF: “Census / Population Reports only reflect the quantities and types of collectibles previously graded by the CCG Companies, and are not an indicator of value or rarity” [2]. That is not a hedge added by a cautious lawyer. It is the single sentence this entire article exists to unpack, and CGC has already written it, in public, on the page where the number lives.

PSA does not appear to print that exact sentence in a form this session’s blocked search could confirm word for word, but its own explanatory material, as summarized consistently across independent hobby guides drawing on PSA’s published guidance, says the same thing in substance: a population count reflects submitted cards only, it excludes copies PSA rejected for alteration or restoration, and it says nothing about cards that were never sent in at all. Two grading companies that compete for the same submission fees have arrived, independently, at the identical disclaimer. That convergence is itself informative. Neither company is warning customers away from a number that helps their business — a population count, real or embellished, drives submission volume by making grading feel like entering a countable, rankable game. If both companies feel obligated to disclaim it anyway, the disclaimer is not marketing caution. It is a description of what the data structurally is.

A Submission Is a Coin Flip That Knows the Coin

Here is the mechanism the disclaimer gestures at without naming it. A population count is a sum over a long series of independent private decisions: does this specific card’s owner choose, at some point after 1999, to pay a grading company between roughly thirty and several hundred dollars, wait weeks to months, and accept the risk that the card comes back a lower grade than expected — or does the owner not do that. Call that decision the submission event. A population report counts submission events. It was never designed to, and cannot be read as, a count of the underlying physical stock those events are drawn from.

Figure 2. Every submitted card passed through a moment like this one first — an owner's private read of its own condition, taken before any grading company ever saw it, that a population report cannot record because it happens before the count begins. — Image prompt and art direction by Brecht Corbeel; image generated to that direction.

The general shape of the problem is a textbook one in applied statistics, formalized for economics by James Heckman: when the sample an analyst observes is not drawn at random from the population of interest, but is instead the product of an individual choice correlated with the very outcome being studied, ordinary sample statistics are not merely noisy estimates of the population parameter — they are biased in a direction set by whatever drives the choice [6]. Heckman’s own machinery was built for labor-supply data, correcting wage regressions for the fact that only people who choose to work report a wage. The correction technique does not transfer cleanly to a card population report — there is no comparable instrument here, no exclusion restriction, no observed set of non-submitters to compare against. What transfers is the diagnosis. Submission is the choice; grade, and public knowledge of grade, is the outcome; and the choice is not made at random with respect to the outcome. An owner submits a card because they already believe, from a raw-eye read of centering, surface, edges, and corners, that it will grade well enough to justify the fee and the wait. An owner who suspects a card will come back a 4, or “Authentic Altered,” has a live financial reason not to submit it at all — and a card that is never submitted contributes zero to the population report, regardless of how many physical copies like it sit in binders and long boxes. Every one of the properties a genuine rarity estimate would need to be independent of — perceived condition, perceived value, awareness that grading exists as an option, and how recently the owner acquired the card — is instead one of the specific things that predicts whether the submission event happens at all.

Figure 1. This card exists. No population report will ever say so, because no one has submitted it — the structural blind spot no submission count can see past. — Image prompt and art direction by Brecht Corbeel; image generated to that direction.

This is not a claim that PSA’s count is “too low” by some knowable factor, which would itself be a disguised survival estimate dressed up as skepticism. It is a claim that the count is silent, by construction, on the question it keeps getting asked. A submission count is a genuine, valid lower bound on exactly one thing: how many copies of the card have existed since 1999, because a card that was graded must have existed in order to be graded. It is uninformative — not imprecise, structurally uninformative — about how many copies exist now, because the gap between “existed at some point” and “exists now, and here is how many” is precisely the gap a self-selected sample cannot close.

Crack-and-Resubmit Breaks the Same Lower Bound From the Other Side

Before pressing the undercounting argument further, it owes its own strongest objection a fair hearing, and it arrives from an unexpected direction: submission counts do not only miss cards. They also double-count them. A 2019 hobby analysis, working from sports and non-sports population data rather than Pokémon specifically, found cards whose total graded population exceeds their own printed run — a Juan Soto orange refractor numbered to 25 copies showing 36 total graded entries, a Mike Trout blue refractor numbered to 150 showing 163 [10]. The mechanism is crack-and-resubmit: a collector who dislikes a grade removes the card from its holder and resubmits the identical physical object, sometimes to the same company and sometimes to a different one, and each attempt adds a new line to the population count of a card that has not multiplied at all.

If that mechanism operates on Pokémon population data too — and there is no structural reason it would not, since crack-and-resubmit is a normal, openly discussed hobby practice rather than a sports-card-specific quirk — then “101,739 total submissions” is not even a clean count of 101,739 distinct physical Charizards that have ever passed through PSA’s doors. Some unknown share of that figure is the same handful of cards submitted more than once. This complicates the lower-bound claim made above, and the honest response is to narrow the claim rather than abandon it: the population count remains a valid lower bound on the number of grading events, and a grading event requires a distinct physical card to have existed at the time of submission, but the count is no longer a clean, one-to-one lower bound on the number of distinct cards submitted, let alone the number that survive. The correction only tightens this piece’s central claim rather than weakening it. A number that both under-samples the true surviving stock, through non-submission, and over-samples its own submission tally, through resubmission, has moved, if anything, further from being readable as a survival estimate, not closer.

What a Real Survival Estimate Requires, and Why No One Has Run One

Name the alternative precisely, because “PSA should just do better” is not a real proposal. Statisticians who actually estimate the size of a hidden population — the true number of a wildlife species in a habitat, the true number of people in a category a single administrative list under-records — do not fix the problem by refining one list. They use multiple, independently drawn samples of the same population and estimate the total from the overlap between them, a family of methods that goes by capture-recapture in ecology and multiple systems estimation in public-policy statistics: two or more partial, imperfect enumerations, combined, can estimate a population that no single one of them observes completely [9]. Applied to a Pokémon card, the analogous study is easy to state and, as far as this research pass could find, has never been run: a randomly sampled physical audit of raw, ungraded Base Set Charizards — drawn not from grading-company submissions but from some independent frame, such as a stratified sample of estate liquidations, collection-insurance registries, or dealer buy-lists — cross-referenced against PSA’s and CGC’s own submission records for overlap.

Figure 3. This form does not exist anywhere in the hobby: a clipboard for a randomly sampled physical audit of raw, ungraded stock, the only method that could turn a submission count into an actual survival estimate. — Image prompt and art direction by Brecht Corbeel; image generated to that direction.

No hobby publication, PSA report, or academic paper turned up in this session’s search does that. What does exist is a real, dated, peer-reviewed paper applying genuine survival-analysis machinery to Pokémon card market data: a 2026 study followed 300 cards listed for sale on eBay for a year and used Kaplan-Meier estimation and Cox regression to model how quickly they sold, reporting that 73.3% of the sample sold during the study period for a median price of €1.95 [7]. That paper’s “survival” is a listing’s time on the market, not a physical card’s persistence through decades of handling — a different variable entirely, and the two should not be confused — but its existence matters for a specific reason: it proves real academic statistical tools have already been pointed at Pokémon card market data, by researchers willing to do the fieldwork of tracking three hundred individual listings for a year. Nobody has yet pointed the capture-recapture version of those same tools at the question this article is actually about. The gap is not that the right machinery is missing. It is that nobody has funded the fieldwork the machinery requires.

Thirty-Nine Copies, Then Forty-One

The clearest illustration of how hard genuine population accounting is comes from the one Pokémon card where the print run is not a mystery at all: the 1998 CoroCoro “Pokémon Illustrator” promo, awarded only to winners of three named Japanese art contests, with a documented total of 39 officially distributed copies [8]. This is as close to a closed population as the hobby has — no unlimited retail print, no boosters, a fixed and countable ceremony of distribution. If any Pokémon card’s “known to exist” figure should be reliable, it is this one.

It was not. When a PSA 9 copy sold for $100,000 in 2013, the figure circulating at the time held that only 12 copies existed — less than a third of the 39 actually distributed [8]. By August 2020, two more physical copies had surfaced from private holdings that had never been part of any prior public count, pushing the number known to exist to at least 41 — two more than were ever officially awarded in the first place [8]. A population this small, this documented, and this frequently discussed in a market with millions of dollars riding on its scarcity still could not hold a stable “known to exist” figure for longer than a decade.

Figure 5. Thirty-nine prize copies were officially distributed at three 1997-98 illustration contests. By 2020, two more had surfaced anyway — proof that even a population small enough to be a genuine census still gets revised upward. — Image prompt and art direction by Brecht Corbeel; image generated to that direction.

Scale that same instability up to a card with a print run in the tens or hundreds of thousands and an unknown number of owners who have never heard of PSA, and the claim that 101,739 submissions tells you how many Charizards “exist” stops being merely unproven. It becomes a category error borrowed from a case where the numbers happened, for one card, to be small enough that nobody had yet noticed the borrowing was unjustified.

“Closer to 10,000” Is the Submission Count Wearing a Costume

The Illustrator case is instructive precisely because its true population is small enough to argue about in public. Base Set Charizard’s is not, and the hobby’s response to that gap has not been silence — it has been confident restatement. One representative 2024 buying guide states plainly that “many collectors and experts believe the number of surviving cards — whether graded or ungraded — might be much closer to 10,000,” offered with no cited source, no described sampling method, and no acknowledgment that “surviving” and “graded” are different words describing different things [11]. The same guide lists a PSA-10 population of 125 for the 1st Edition print elsewhere on the identical page, with no stated arithmetic connecting the two figures [11]. Trace the 10,000 figure back far enough, through enough collector forums and pricing blogs each citing “most experts agree” without naming one, and it resolves into exactly the pattern this article is about: somebody, at some point, looked at a submission-derived number, judged it probably conservative, added a plausible-sounding multiplier, and the result now circulates with more apparent authority than the number it started from.

None of this means 10,000 is wrong. It might, by pure coincidence, land closer to the true figure than anything derivable from submission data alone. That is precisely the problem with it: a number produced by no stated method cannot be evaluated as right or wrong, only repeated or not repeated, and repetition is not evidence.

The Real Signal Hiding Inside a Count Everyone Reads Wrong

None of the above should read as an argument that a population report is useless, and crediting its real use is not a concession — it is the other half of the claim. PSA’s own count and third-party aggregators like GemRate track submission volume over time, and that is a genuinely different, genuinely answerable question from “how many cards survive”: not a stock question but a flow question. GemRate’s own year-end accounting for 2025 reports 26.8 million cards graded across the major grading companies that year, up 32% from 20 million in 2024, with 16.8 million of those in the broader trading-card-game and non-sports category — up 95% year over year — against 10.0 million sports cards, down 12% [5]. That figure spans every trading-card game and non-sport category the major graders handle, not Pokémon specifically, and should not be read as a Pokémon-only number; it is cited here only to establish that submission-flow tracking at industry scale is a real, continuously measured thing, distinct in kind from any single card’s stock estimate.

Figure 4. What a population report is actually good for lives in this drawer: the same card's submission count, read across dated snapshots, tracks new supply entering the count — a real signal, just not the one it is usually asked for. — Image prompt and art direction by Brecht Corbeel; image generated to that direction.

A single card’s submission count, read the same way, over time, does say something real: if a card’s PSA-10 population has stayed nearly flat across several years of active trading, that is evidence that few gem-quality raw copies remain in circulation and being newly submitted, whatever the true surviving stock turns out to be. That is a legitimate, narrower, answerable claim — a statement about the current rate of high-grade supply entering the certified market, not a statement about a denominator nobody has measured. The hobby’s error is not that it tracks this. The error is in reaching, from the same data, for a second and much larger claim the data was never built to support.

Report It as Submissions

Every piece of this argument points at the same fix, and it costs nothing to implement: stop writing “known to exist,” “surviving population,” or “rarity” beside a number that is, at every stage of its life, a count of paid submissions. Write “submissions: 101,739” or “graded population: 101,739,” exactly as CGC’s own disclaimer already insists it must be read, and let the number do the narrower job it can actually do [2]. A collector who wants an actual claim about scarcity has a genuine, harder question available instead — how has this specific card’s submission rate moved over the last five years, relative to comparable cards from the same set — and that question can be answered honestly from the same data that cannot answer the question usually asked of it.

487 of 101,739, or 489 of 92,856, or whatever PSA’s own page shows the next time someone manages to load it: the number was always a census of one thing. It was a census of the people who decided to ask.

That reframing is cheap precisely because it asks nothing new of PSA or CGC — both companies already print the correct description on their own pages, in their own words. The work left undone is entirely on the reading side: in the pricing guides, the forum threads, and the auction-house lot descriptions that keep borrowing the word “rarity” from a column of data that was only ever counting envelopes.

Sources

1. PSA Population Report — Pokémon Trading Card Game. Professional Sports Authenticator (PSA) (2026).
2. Population Report — Pokémon TCG. CGC Cards (2026).
3. PSA Pop Report Explained: How to Read Population Data (2026 Guide). CardboardSniper (2026).
4. Base Set PSA Population Report. Pikawiz (2026).
5. 2025 Grading Year in Review. GemRate (2026).
6. James J. Heckman. Sample Selection Bias as a Specification Error. Econometrica (The Econometric Society) (1979).
7. Johannes Heck, Sarah Schormann, Benjamin Krichevsky, Adrian Groh, Thorben Pape, Sebastian Schröder, Alexander Glahn, Carsten Schumacher, and Martin Schulze Westhoff. Sales characteristics of Pokémon trading cards: A prospective one-year field study. PLOS ONE (2026). DOI: 10.1371/journal.pone.0334289.
8. Pokémon Illustrator (CoroCoro promo). Bulbapedia (2026).
9. Sheila M. Bird and Ruth King. Multiple Systems Estimation (or Capture-Recapture Estimation) to Inform Public Policy. Annual Review of Statistics and Its Application (2018). DOI: 10.1146/annurev-statistics-031017-100641.
10. The Curious Case of Population Reports. SlabStox (2019).
11. How Many 1st Edition Charizards Are There in 2025?. OG Cards (2024).

Originally published at https://absolutedigitalpublishers.com/articles/a-population-report-counts-submissions-not-survivors.