Grading, rarity & authentication

1. Card identity and the SKU

Most card databases model a card as a name plus a set. That is not enough to trade against. "Charizard, Base Set" describes at least six economically distinct objects, and if you match buyers to sellers on that string you will produce disputes for a living.

We split identity into three layers. A printing is what came off the press. A SKU is a printing plus a grade opinion — the fungible unit, the thing that gets an order book. An item is one physical object with a cert number, an LPN, a bin and an owner.

printing = (game, set_code, collector_number, language, finish, edition)

// all six are load-bearing:
(pokemon, BS,  4/102,   EN, holofoil,     1st Edition)   → six figures
(pokemon, BS,  4/102,   EN, holofoil,     Shadowless)    → four figures
(pokemon, BS,  4/102,   EN, holofoil,     Unlimited)     → three figures
(pokemon, CEL, 4/102,   EN, holofoil,     Unlimited)     → two figures, same artwork

sku = printing + (grader, grade)
item = sku + (cert_number, lpn, facility, bin, owner)

The last line of that example is the one that catches people. The Celebrations Classic Collection Charizard is the same Mitsuhiro Arita artwork, the same layout, the same collector number. It carries a small anniversary stamp and it is worth roughly a thousandth of the 1999 1st Edition. A model that keys on artwork will happily match those two and settle a trade that ruins someone's month.

2. The grading scales

Five graders dominate cardboard, and they do not agree with each other. They use different increments, different tolerances, different philosophies about what a "10" means, and their grades do not convert cleanly. We store the grader alongside the number, always, and we never normalise them into a single house scale — that would be inventing precision we do not have.

3. Centering, measured

Centering is the one grading input that is genuinely objective, and it is the single most common reason a card that "looks mint" comes back a 9. It is measured as the ratio of the border widths on each axis. A card whose left border is 60% of the combined left+right border width is a 60/40 on the left–right axis.

Our intake scanner computes this from the raw 1200 DPI capture: detect the card edge, detect the inner art frame, measure all four border widths in pixels, and report both axes front and back. No human eyeballing, no "looks good to me."

Published tolerances vary by grader and, for PSA, by era — vintage is graded with a lighter touch than modern, because 1999 print registration was worse than 2024 print registration. These are the working thresholds our pre-grade model uses:

4. Subgrades and the Black Label problem

BGS, CGC and TAG report component scores. Four numbers carry far more information than one, and they change how a card should be priced.

The final BGS grade is not the average. It is weighted, and the lowest subgrade drags hard: a card with 10 / 10 / 10 / 9 subgrades typically lands at 9.5, not 9.75. This asymmetry is what makes the Black Label — four straight 10s — so rare and so expensive. The population difference between a BGS 9.5 and a BGS 10 on a modern chase card is routinely three orders of magnitude, and the price difference tracks it.

// Two cards, same final grade, materially different assets.
{ "grader": "BGS", "grade": 9.5,
  "sub": { "centering": 10, "corners": 9.5, "edges": 9.5, "surface": 9.5 }}
// ↑ "quad" — one step from Black Label on every axis. Crossover candidate.

{ "grader": "BGS", "grade": 9.5,
  "sub": { "centering": 8.5, "corners": 10, "edges": 10, "surface": 9.5 }}
// ↑ carried by three strong axes. Will never improve — centering is permanent.

Those two cards will be quoted at the same price by any system that stores only the final grade. The first is a genuine crossover candidate — crack it, resubmit to PSA, and there is a real chance of a 10. The second is what it is forever, because centering is decided at the printing plant and no resubmission changes it.

5. Population reports and pop pressure

Every major grader publishes a population report: how many copies of a given printing they have assigned each grade. It is the closest thing the hobby has to a supply figure, and it is the number most often misused.

The naive read is "low pop means rare means expensive." The useful read is the shape of the distribution above your grade. A card with 12 copies at PSA 9 sounds scarce until you notice there are 4,000 at PSA 10 — your 9 is not rare, it is the bottom of a pile that mostly gemmed. We surface this as pop pressure: the share of the graded pool sitting strictly above the grade you hold.

pop_pressure = pop_above_grade / (pop_at_grade + pop_above_grade)

// 1999 1st Ed Charizard, PSA 9:  748 at grade, 121 above
pop_pressure = 121 / 869 = 0.14   → genuinely near the top of the pool

// Modern bulk chase card, PSA 9:  310 at grade, 5,102 above
pop_pressure = 5102 / 5412 = 0.94  → a 9 is the consolation prize

6. Rarity taxonomy by game

"Rare" is a word three different publishers use to mean three different things, and each of them changed the meaning several times. We keep the native rarity exactly as printed, and map it to a cross-game tier used only for sorting and modelling — never shown as if it were the real rarity.

Pokémon

The base ladder is printed as a symbol in the bottom corner: ● Common, ◆ Uncommon, ★ Rare. Everything interesting sits above that, and the modern Scarlet & Violet era reorganised the top end substantially.

Two Pokémon-specific identity flags matter more than the rarity itself on vintage. 1st Edition is a small stamp to the left of the art; Shadowless is the absence of the drop shadow on the right edge of the art frame, plus a thinner HP font. Both are print-run markers, both are trivially faked in a photograph, and both are worth multiples. They live in the printing key, not in a description field.

Yu-Gi-Oh!

Konami's rarity ladder is the deepest of the three, and it is a foil-treatment ladder rather than a print-frequency ladder. The card name, the artwork border and the card body each get treated differently at each tier.

Yu-Gi-Oh! print codes carry the set and region inline: LOB-EN001 is Legend of Blue Eyes White Dragon, English, card 001. The region segment is part of identity — European English, North American English and Asian English prints of the same card are different assets with different populations. And as with Pokémon, 1st Edition versus Unlimited is a multiple, not a footnote.

Magic: The Gathering

Magic's rarity is the least informative of the three, because the interesting axis is treatment rather than rarity. Common / Uncommon / Rare / Mythic tells you the pack slot. What moves price is finish and frame: foil versus non-foil, borderless, showcase, extended art, retro frame, and serialised.

• Foil curl — modern foils are a laminate that reacts to humidity. This is the single strongest operational argument for holding Magic at a controlled 47% RH; a curled foil loses grade permanently.
• Alpha versus Beta — 1993 Alpha cards have a visibly larger corner radius than Beta. It is a physical, measurable print-run difference, and it is the difference between two very expensive cards and two extremely expensive cards.
• Serialised cards — a stamped 0348/0500 is part of the item identity, not the SKU. Serial 001 and serial 348 share a book unless the market says otherwise, but the individual serial travels with the physical object.

7. Raw condition and pre-grading

Not everything is slabbed, and not everything should be. Raw cards are assessed at intake on the standard five-step ladder, and we apply a published haircut to the Near Mint mark rather than pretending condition is a free parameter.

Because we already capture centering ratios, gloss index and defect counts at intake, we can answer the question every raw submission actually wants answered: is it worth grading? The pre-grade model returns a predicted grade, the binding constraint, and a confidence.

// request — measurements as the intake scanner reports them
{ "centerLR": 58, "centerTB": 52, "backLR": 66,
  "corners": 0.5, "edges": 0, "surface": 0, "gloss": 0.94 }

// response
{ "predicted": 9.0,
  "psaEquivalent": 9,
  "centeringCap": 9,          // 58/42 front → cannot gem
  "conditionScore": 9.5,
  "frontRatio": "58/42",
  "limiter": "centering",     // what is actually holding it back
  "confidence": 0.62 }

The honest part of that response is limiter. Knowing a card will grade 9 is less useful than knowing why, because the why tells you whether a resubmission could ever change the answer. Corners and surface are luck and grader variance. Centering is permanent.

8. Authentication at intake

Every item that arrives goes through the same bench, whether it is a $4 raw card or a six-figure slab. The sequence is fixed so that the audit trail is comparable across every item we have ever held.

9. Counterfeit and tamper detection

Custody only means something if what is in the bin is what the record says. The failure modes are well known and each has a cheap physical test.