Analysis · Methodology risk

Every pack EV figure rests on card values nobody audits

Expected value is a simple calculation: the chance of each outcome multiplied by what that outcome is worth. The odds half gets all the attention — platforms publish them, we analyse them, people argue about them. The values half is where the real uncertainty lives, and almost nobody checks it.

Card values on these platforms are sourced by the platforms themselves, typically from marketplace data. Nothing verifies them. And because a pack pool can hold hundreds of cards, overstating a handful lifts the advertised expected value without anything looking wrong from the outside.

What we found, and what it does and doesn’t prove

While reviewing a competitor’s guaranteed-value pack — a product promising you receive a card worth at least what you paid — we checked the two cards it could return. One was listed at a fair market value just above the $50 pack price. Independent sales data for that card suggested it trades closer to $30.

If that is right, the guarantee fails on its own terms: you pay $50, you receive roughly $30 of card, and the interface tells you that you received more than you paid.

We are not naming the platform yet. This is one card against one price source. It could be a stale third-party feed rather than anything intentional — that distinction matters enormously, and we have not earned the right to assert either. We are checking more cards against more sources, and we will publish the result whichever way it goes.

What one card does establish is narrower but still important: stated card values in this category can be materially wrong. Once that is true anywhere, it is a live risk everywhere, because the mechanism is identical on every platform.

The better test: hold a platform to its own stated method

We asked. The platform explained how its values are produced, and the answer was more forthcoming than we expected — more methodological disclosure than anyone else in this category offers:

“Market value is based on real sales data from major marketplaces… recent sales of the same card on places like eBay, alt.xyz, and pricecharting.com. It calculates the median price of those last-sold cards. Prices are updated daily.”

That is a reasonable method on paper — the median of recent comparable sales is close to what we would do ourselves. But we initially credited the platform for unusual transparency here, and that was wrong: near-identical language appears across this category. Describing a sensible method is standard. Demonstrating you follow it is not, and we have seen nobody do the second thing.

Which makes a stated method a double-edged thing. It gives a buyer assurance, and it is the only lever anyone has to check the assurance is earned — a described process nobody verifies is worse than no process at all, because it manufactures confidence without supplying grounds for it.

Which is where it gets interesting. Applying that stated method to the sales history visible on the certification page the platform links itself, the card in question has recent sales of $31.99, $23.82 and $15.50 — a median of $23.82. The displayed value is $51.10, roughly 2.1× what their own described process should produce.

There are ordinary explanations. Their sources may capture sales the certification page does not show. The card is thinly traded, so a “recent” median may reach back much further than it appears to. Something in the pipeline may not be doing quite what the description says. We have asked, and we will publish the answer.

The point is the test, not this one card. Comparing a platform’s values against our preferred price source is an argument about whose source is better. Comparing them against the platform’s own stated methodis not — it either reconciles or it doesn’t. That is the check we will run on every platform that publishes a methodology, and the absence of a stated method is itself worth knowing about the ones that don’t.

Why this is nearly invisible to a buyer

The pools are large. A pack listing hundreds of possible cards cannot realistically be checked by hand before you buy. Nobody is pricing three hundred slabs to evaluate a $50 purchase.
The error only needs to be small and selective. Inflating the values of a few high-probability outcomes moves the headline EV meaningfully. Nothing else on the page has to change.
Graded cards are genuinely hard to price. The same card in the same grade can trade across a wide range depending on grader, population, eye appeal and recency. A generous-but-arguable figure is not obviously wrong, which is exactly what makes it hard to challenge.
The industry is unregulated. No disclosure standard governs how a platform arrives at a card value or how often it must refresh it.

None of that requires bad intent. A platform pulling prices from a third-party feed inherits whatever that feed gets wrong, and stale prices in a falling market overstate values without anyone deciding to mislead. The effect on your wallet is the same either way, which is why it is worth measuring rather than assuming.

This applies to our own numbers

We publish that Collector Crypt’s machines return 93.7–95.1¢ per dollar at the buyback. That figure is arithmetic on two numbers Collector Crypt publishes: their stated expected value, and their buyback rate. We have been careful to call it published rather than estimated, and it is.

But it is a published figure about their valuation of the cards, not an independently verified market price. If their card values skew high — for any reason, including an upstream data feed — then the true return is lower than we have published.

So our figure is a ceiling, not a midpoint. The error in it, if there is one, runs in one direction only: reality can be worse than 93.7–95.1¢ per dollar, and cannot be better. We would rather tell you which way our own number can be wrong than present it as more precise than it is.

This changes what we do next. Spot-checking a sample of each platform’s stated card values against independent sales data is now a standing job rather than a footnote, and the results will appear in each platform review. It is slow, unglamorous work and it is the part that actually determines whether any EV number in this category means anything.

What you can do before buying

Spot-check two or three cards. Take the most likely outcomes — the common tier, not the jackpot — and look them up on an independent sales-data site. The common tier is what you will almost certainly receive, so it is where an inflated value costs you most.
Be most sceptical of guarantees. A guaranteed floor is only as good as the valuation behind it. It is also where a platform has the strongest incentive for a generous number, because the guarantee is the selling point.
Check what the buyback actually pays. A buyback quote is a real price the platform will really pay, which makes it harder to inflate than a notional market value. Where the two disagree sharply, believe the buyback.
Or skip the question. Buying a specific card from a marketplace at a stated price removes the valuation problem entirely — you can check that one card properly before paying.

Observed 2026-08-04 · platform unnamed pending verification across multiple cards and sources · we will publish the result either way