Methodology · Updated November 18, 2026

How SnapGradeAI grades cards, and where it stops being reliable.

A plain explanation of what goes in, what comes out, how confidence works, what a photo cannot show, and what we do and do not claim about accuracy.

Inputs and outputsLimits statedNo accuracy claim

Updated

16,000+
Cards graded
11,000+
Accounts created
4
Subgrades per card
The short version

A prediction, not a PSA grade.

Everything below describes what the tool does today. If something is not listed, we are not claiming it.

Two photos in

You upload the front and the back of the card. The back matters: PSA grades it, and edge and corner wear often show there first.

Four axes scored

Centering, corners, edges and surface are each scored on PSA's 1–10 scale. Each axis is judged on the most severe defect visible on it, not on an average.

Confidence per axis

Every axis comes back high, mid or low confidence. If glare, blur or a sleeve hides an area, the axis is marked low instead of guessed.

A whole-number grade

The overall prediction uses only grades PSA issues: whole numbers from 1 to 10, never a 9.5. The report names the axis that limits the grade.

Era-aware

The grader reads the card's era from its design and copyright line. Vintage cards rarely reach a 9 or 10, so predictions reflect aging.

No authentication

It predicts condition from photos. It cannot tell you whether a card is genuine.

What happens when you grade a card

  1. Photos are checked. The grader looks at both images. Areas it cannot see well are not scored as if they were clean.
  2. Each axis is scored. Corners are checked for whitening, fraying and rounding; edges for specks, chipping and nicks, especially on the back; surface for scratches, print lines, dents, creases, stains and holo scratching; centering by comparing opposite borders on the front and the back.
  3. Confidence is attached. Each axis gets high, mid or low confidence depending on how clearly the photo shows it.
  4. An overall grade is produced. It is limited by the weakest axis and returned with a short plain-English summary naming the flaw that decided it.
  5. Credit handling. If the photos do not let the grader assess at least two of the four axes, the credit is returned automatically.

What a photo cannot show

Known limits of a photo-based prediction
LimitWhy it mattersWhat to do
Faint surface flawsHairline scratches and shallow dents often need raking light to seeTilt the card under a lamp before you submit anything predicted 9 or 10
Glare on foils and chromeReflections hide detail and lower confidenceRe-shoot with soft light; see how to photograph cards for grading
AuthenticityA condition prediction says nothing about whether a card is realAuthentication is a grading company's job
Borderless and full-art cardsNo clear border to measure centering fromRead the confidence on centering and measure by hand if it is low
Photo qualityBlur, tilt and low light change the resultPhotograph straight on, sharp, in even light

What we claim about accuracy: nothing numeric

SnapGradeAI does not publish an accuracy percentage. We removed earlier figures from the site because they were not backed by real PSA results, and we will not quote one until it is. Users can report the real PSA grade (cert number) of a card they submitted, and those reports are the data we are collecting. Until then the honest signal is the confidence level on each axis, and your own test: grade a card whose PSA grade you already know.

In November 2026 the grading was recalibrated toward PSA's real distribution. In an internal check on 41 test cards, the share predicted 9 or higher fell from 95% to 27%. That was a small internal check, not a validation study.

How to use a prediction well

More on what the AI can and cannot see: Pokémon card grading AI. Questions or a result that looks wrong? Contact us with the card photos and what PSA graded it.

Common questions

Everything you actually want to know.

How does SnapGradeAI predict a grade?
It analyses your front and back photos on four axes (centering, corners, edges, surface), scores each on PSA's scale using the most severe visible defect, attaches a confidence level, and returns a whole-number overall grade limited by the weakest axis.
How accurate is it?
We do not publish an accuracy percentage. Confidence levels per axis are the honest guide, and you can test it on cards with a known PSA grade.
Why is there never a 9.5?
Because PSA does not issue half grades on its standard 1–10 scale, so the prediction uses only whole numbers.
What if my photos are poor?
The affected axis is marked low confidence. If two or more axes cannot be assessed, your credit is returned automatically.
Can it detect fake cards?
No. It predicts condition only.
Can I tell you what PSA actually graded my card?
Yes. The app has a report form for the PSA grade of a card you submitted, which adds one credit and helps us measure real accuracy.
Ready when you are

Test it on a card
whose PSA grade you know.

Sign up free, 1 free credit, and compare the prediction with the real grade.

On signup
+1 credit
No credit card · No expiry