For collectors who actually grade

How we grade.

Last updated: May 10, 2026

Every prediction CardPrepAI makes is built against three targets — the standards collectors actually chase. Not theoretical models. Not "we used computer vision." The same criteria a real grader sits down with at PSA, BGS, or TAG.

The three Pristine standards

We grade every card against all three of these simultaneously. You see all three predictions. You decide where to submit.

Standard 01

BGS Black Label

Beckett Pristine 10 — perfect 10 across all four subgrades: Centering, Corners, Edges, Surface. Front and back. Less than 1% of submissions hit it.

Standard 02

PSA 10 Gem Mint

PSA's flawless tier. Single composite score, but graders look at all four criteria internally. Most-recognized grade in the hobby. Highest secondary-market premium.

Standard 03

TAG 10 Pristine

TAG Grading's machine-learning Pristine 10 — measures centering to 0.1mm, scores subgrades algorithmically, then a credentialed human grader reviews flagged defects to prevent false positives. 1000-point scale. Growing fast among modern collectors.

Why three targets? A card that hits PSA 10 may miss BGS Black Label because BGS subgrades the back. A card that hits BGS Black Label is almost always TAG 10 too. We tell you which standard is most likely — and for highest-value submissions, which grading service to send to.

The pessimist rule. All three graders apply some version of the same hidden rule: your final grade is capped by your worst category. PSA's final grade is the LOWEST score across centering, corners, edges, and surface. BGS's composite is non-linear — three 10s and one 9 give you a 9.5, not a 9.75. TAG's algorithmic score weighs the worst category most heavily. One bad corner caps an otherwise perfect card. That's why we score every category independently and surface the worst.

Where we start: centering and whitening

Per the consensus of working graders, two criteria account for ~70% of grade-cap decisions on submitted cards:

Centering — measured front and back. Off-center caps the grade before any other criterion is even examined.
Whitening — corners and edges, front and back. Even one whitening tell visible from 24 inches caps a card at PSA 9.

Centering — what we measure

For every card we calculate:

Front centering, left-right — pixel-distance from artwork edge to card edge, expressed as a ratio (50/50 = perfect, 65/35 = the practical PSA 10 cutoff)
Front centering, top-bottom — same, vertical axis
Back centering, both axes — BGS scores this as a separate subgrade. PSA doesn't formally subgrade the back, but a back centered worse than 70/30 will cap the final grade.

Centering ratio	BGS subgrade cap	PSA grade cap	TAG grade cap
50/50 to 55/45	10	10	10
55/45 to 60/40	9.5	10	9.7
60/40 to 65/35	9	9	9.3
65/35 to 70/30	8.5	8	8.8
>70/30	8 or below	7 or below	8 or below

Whitening — what we look for

Tiny visible white tells appear at corners and edges where the print layer chips off. Most submissions die here. We score:

Each of the 4 corners, front side
Each of the 4 corners, back side
Each of the 4 edges, front side
Each of the 4 edges, back side

That's 16 individual whitening checks per card. The card is rated on the worst-scoring location — graders are pessimists.

The back kills more cards than the front. Almost every collector inspects the front, photographs the front, brags about the front. Most rejected high-grade submissions die on a single back-corner ding. We score the back with the same scrutiny as the front because BGS does, and PSA's repackaged report-card feedback consistently lists "back" defects as the cap.

The other criteria

Criterion	What we score	BGS	PSA	TAG
Corners	4 corners × 2 sides — sharp 90° vs fuzz/fray	Subgrade	Bundled with edges	Subgrade
Edges	4 edges × 2 sides — chipping, dings, whitening	Subgrade	Bundled with corners	Subgrade
Surface	Print lines, scratches, holo damage, ink spots, foil cracks, print bubbles	Subgrade	Major factor	Subgrade
Print quality	Registration, ink density, dot-matrix issues	Surface subgrade	Surface bucket	Separate score

Base cards vs full arts — different specs entirely

Treating every card the same would be the #1 way to be wrong. The artwork — and where the print layer ends — fundamentally changes what graders are looking at.

Base cards · yellow / silver / white border

Examples: 1999 Pokémon Base Set, modern Pokémon yellow-border, Topps silver-border, vintage white-border baseball.

Centering: measured by the visible border width on each side. Even ¼ mm of asymmetry shows.

Whitening: brutal. Yellow and silver borders make every corner chip and edge nick obvious.

Surface: more forgiving — lots of solid color background where minor flaws blend.

Path to PSA 10: tight 50/50–55/45 centering + zero whitening on corners and edges + clean front face.

Full arts · alt arts · hyper rares · Manga Rares

Examples: Pokémon alt arts, Topps Chrome refractors, Magic Showcase / Borderless, One Piece Manga Rares, Lorcana Enchanted.

Centering: judged by artwork-to-card-edge ratio (no defined border). Harder for graders to call, but TAG measures algorithmically.

Whitening: more forgiving on the front (artwork covers it) but the back is critical and caps the grade.

Surface: brutal. Every holo crack, scratch, print bubble, foil flake shows. Foil-pattern alignment matters.

Path to PSA 10: immaculate front surface (no holo damage) + sharp corners + clean back centering.

What we evaluate per photo

Front photo (required)

Centering — left/right, top/bottom (pixel-precise)
Each of 4 corners — sharpness, whitening, fraying
Each of 4 edges — whitening, chipping, dings
Surface — scratches, scuffs, print lines, holo damage, foil cracks, ink spots, residue
Print quality — registration shifts, dot-matrix issues
Border-aware adjustments based on card type detected

Back photo (required)

Centering — left/right, top/bottom
Each of 4 corners — same scrutiny as front
Each of 4 edges — same scrutiny as front
Surface — scratches, registration shifts, ink consistency
Authenticity tells — backstamp registration, color consistency, era-appropriate cardstock signatures

What you get back

Per card, every prediction includes:

BGS subgrade prediction — Centering / Corners / Edges / Surface (each 1-10) and final composite
PSA grade prediction — single 1-10 composite
TAG grade prediction — single 1-10 composite plus subgrades
Confidence per criterion — so you know exactly where the AI is most certain and where it's hedging
Verdict — Submit / Hold raw / Walk, with EV math
Best grader for this card — based on which standard it most likely hits

Confidence transparency

We will not pretend to be more certain than we are. Every prediction comes with a confidence score per criterion. Low confidence → flagged "submit at your discretion." If we miss by more than one full grade with high confidence, you get a free Concierge re-analysis and a credit. Full AI disclosure.

Reality check — PSA 10 hit rates on flagship cards

Most cards do not hit PSA 10. Even pack-fresh cards from 2026 fail. These are the actual current pop reports for marquee cards — the math you should run before submitting anything you care about:

Card	Total submissions	PSA 10s	Hit rate
1999 Pokémon 1st Ed Shadowless Charizard #4	3,000+	121	~4%
1986-87 Fleer Michael Jordan #57	14,000+	189	~1.4%
2018 Topps Chrome Acuña Refractor (typical modern flagship)	~12,000	~25%	~25%
One Piece OP01 Romance Dawn Japanese (mass-graded)	50,871	~40-50%	~45%

Vintage cards: 1-5% PSA 10 odds. Modern with Bandai-grade factories: 30-50%. The gap is centering and surface — old cards were factory-cut sloppy and have decades of wear; modern cards are tight at the factory but pack contact creates micro-scratches.

What this means for you: if the AI predicts "PSA 10 with 60% confidence" on a vintage card, that's actually extraordinary. The realistic ceiling is often a confident PSA 9 with a 30% PSA 10 upside. That's still a +EV submission for many cards. The math is everything.

Photo requirements for accurate predictions

Sharp focus across the entire card surface
Even, diffuse lighting — no glare on holo areas
Neutral background (white, black, or solid gray)
Both front and back, captured separately
Macro mode preferred for surface analysis (most modern phones have this)
Card flat against the surface — no tilt

Questions about methodology? Want to see how we scored your specific card? Email [email protected]. Pro and Concierge members get priority response.