How much shipping adds to a Pokemon card listing

By marketplace

Each marketplace is reported in its own currency and never pooled with another — adding a GBP figure to an AUD figure would be arithmetic on incompatible units. Percentages are within-listing ratios, so they are comparable across rows even though the money is not.

IT had fewer than 20 usable listings. The rows are shown for completeness; they are too small to describe a marketplace and should not be read as one.

How the population was defined

Decided before any figure was calculated, and applied in this order:

1. Retained listing records marked active at the observation cutoff: 1390.
2. Fixed-price only: 1231. Auctions are excluded because their figure is a current bid, not a price.
3. Deduplicated by listing id: 1100, so one listing counts once.
4. A recorded shipping charge above zero only: 429 usable. 671 listings did not record one and were excluded, not counted as zero.

That last exclusion is the most important one, and it cuts two ways. The excluded listings are 61% of the 1100 deduplicated listings. Folding them in as free shipping would have taken the analysed sample from 429 to 1100 listings — an increase of 156.4% — and pulled every figure below toward zero.

The cost of excluding them is a real limit on what this page can say. Every listing counted here charges for shipping, so these figures describe listings that charge. They are not an average over all listings, and they say nothing about how common free or included shipping is.

What “comparable” was made to mean

Two listings sit in the same group only when every field below matches. Some of these are settled by a single identifier and some had to be added explicitly.

Every usable listing carried a card id and a recorded condition, so 0 were dropped from this stage for missing identity. Deduplication is by listing id, which on eBay carries the variation, so two variations of one listing count as the two separate offers they are.

Why a product id does not settle the printing

Earlier versions of this page said the catalogue product id settled the exact card and its printing on its own. Our own retained listings disprove that. At this observation, 5 product ids in the sample were shown by their own listings to cover more than one finish. The clearest case carried nine listings under a single id: one titled “Non Holo”, one “Reverse Holo”, and the rest simply “Holo”. The catalogue records one printing per id — the printing its market price is quoted for — and that is not a promise about what every listing under it is selling.

That matters here because a reverse holo and a plain copy of the same card are different products at different prices. Comparing them and calling the gap a shipping effect would be measuring the wrong thing.

So each listing's printing is now resolved individually, using the same rules the rest of the site uses to decide whether a reference may price a listing at all. Only the seller's own words count as evidence: the catalogue entry cannot settle it, because the catalogue entry is the thing in doubt. Where the printing cannot be established — because the listing is silent on an id already shown to carry several finishes, because it rules out the only printing we hold, or because its title claims two at once — the listing is excluded rather than assigned a likely answer. That removed 43 of 429 listings (10%) before any group was formed.

How a ranking reversal is counted

A group counts as reversed only when no listing tied for the lowest item price also achieves the lowest delivered total — that is, when choosing on item price cannot reach the cheapest delivered total however the tie is broken. This matters: 7 of the 44 groups contain a tie at the lowest item price, and in 3 of them a naive “sort and take the first” would have recorded a reversal that is only an artefact of which tied row happened to sort first.

How this snapshot can be re-checked

A figure is only reproducible if the rows behind it still exist. Freezing the totals is not the same thing — it just fixes a number nobody can re-derive. Earlier versions of this study computed from live listing records, which are overwritten as prices move, so their inputs were gone the moment the run finished and the published result could not be re-checked by anyone, us included.

This revision runs in two steps. The rows are captured and hashed before any figure is calculated, and the calculation then reads only that captured file — no database, no provider, no network. Re-running it on the same input reproduces this page's artifact byte for byte.

Observation cutoff

22 September 2026

Method revision

3

Listings captured

1390

Input digest (SHA-256)

647777d6f6f3a4f7994da0abf2eed91d3cbbae010827db5b159cf0fdca098f55

The captured rows are kept as private project evidence and are not published here. They carry only what the calculation needs — prices, shipping, marketplace, condition, grading, delivery basis and the listing title the printing rules read — and no seller identity, no listing or affiliate links and no raw listing ids. Publishing the digest rather than the rows is what lets a figure be tied to a specific input set without republishing other people's listings.

The two steps are scripts/studies/freezeShippingCostInput.mjs and scripts/studies/buildShippingCostStudy.mjs.

What this does not show

• These are listings, not sales. Nothing here says anything about what anyone paid, or whether these listings sold at all.
• It is a snapshot, not a history. One observation cutoff, one set of retained records. It does not describe a trend and cannot.
• It is not a sample of the market. It is a sample of what we had retained and eligible at that moment, which reflects what we scan.
• Delivery destination is not normalised. Each listing's shipping figure is whatever that listing recorded for its own offer. The domestic / cross-border split keeps a group on a consistent recorded basis; it does not establish that two listings ship to the same place, and it is not a like-for-like international comparison.
• These are recorded totals, not checkout prices. What you would actually pay depends on where you are, which delivery option you choose, and any tax or import charge that applies to you. Nothing here models that.
• Listings whose printing could not be established are excluded, so the comparison covers the subset we could show to be genuinely like-for-like — not every listing in the sample.
• Currencies are never combined. No conversion happens anywhere in the study.
• Only listings that charge for shipping are counted. A listing with no separate charge recorded is excluded throughout, so nothing here is an average over all listings.
• The reversal figure is about 44 groups, not about listings in general. It is not a rate, and it should not be read as one.
• The marketplace rows are descriptive. Nothing here isolates a cause, so no row explains why one marketplace's charges sit where they do.