HOW GEMMR WORKS

The gap betweenlooks like a 10and is a 10.

Gemmr predicts what PSA, CGC, BGS, SGC, and TAG would each return on your card — raw or already-slabbed — across the same four pillars every grader uses. So you know before you ship, before you pay, and before a 9.5 you were sure was a 10 costs you the resubmission fee.

~30s
Average scan time
5
Grading services predicted
4
Pillars analyzed per card
100%
Refund on bad photos
THE PIPELINE

What happens when you scan.

Four stages, each doing one thing well. Each stage's output feeds the next so a confident answer earlier means a faster, cheaper answer overall.

  1. 01
    ~3s

    Preflight checkIs this photo even gradeable?

    Before we charge you for the scan, a fast vision model checks the image for the things that would ruin a grade: blur, glare, partial card, multiple cards in frame, or just — not a card. If anything fails, your Gems get refunded automatically. We never pretend to grade an unreadable photo.

    Too blurry
    Poor lighting
    Partial card
    Glare on surface
    Wrong angle
    Not a card
    Common rejection reasons we catch.
  2. 02
    ~4s

    Slab detectionIf it's encapsulated, we crop the label out.

    For graded slabs from PSA, BGS, CGC, SGC, or TAG, we detect the label region, extract its metadata (service, grade, cert, card name), and then crop those pixels out of the image before grading begins. The grader physically cannot read the label. This is what lets us honestly predict — we're not reading the grade and rounding.

    Label gets cropped out. Card stays in.
  3. 03
    ~5s

    Card identificationName, set, year, number.

    We identify what you've sent us — card name, set, year, and printed number. This becomes context for the grader (cards from different eras have different print quality baselines) and shows up in your report so you can reference it later. Skipped for slabs where the label already provided all this.

    nameCharizard ex
    setPokémon 151
    year2023
    number199/165
    confidence0.94
    Identification data feeds the next stage as context.
  4. 04
    ~15s

    Four-pillar analysisThe actual grading.

    The grader analyzes the card across centering, corners, edges, and surface — the same four dimensions PSA, CGC, BGS, SGC, and TAG use internally. For each pillar we return a sub-grade, a confidence level, and notes on what we saw. Then we synthesize those into per-service predictions, because graders weight the pillars differently.

    9.0
    Centering
    9.5
    Corners
    9.0
    Edges
    10.0
    Surface
    PSA9
    CGC9.5
    BGS9
    SGC9.5
    TAG925
    Four pillars in, five service predictions out.
WHAT THE GRADER SEES

The four pillars, up close.

The grading criteria are the same ones every major service uses. The difference is how each service weights them — which is what makes one card a PSA 9 and a CGC 9.5 at the same time.

01

Centering

How offset is the card image inside its border.

Measured as left/right and top/bottom border percentages. Heavy centering offsets are the most common reason a clean-looking card gets a 9 instead of a 10.

02

Corners

Each of the four corners, individually.

We assess each corner separately: clean, minor wear, whitening, or damaged. Whitening on a black-bordered card is the silent killer of a perfect grade.

03
CLEAN
MINOR WEAR
WEAR
DAMAGED

Edges

Top, right, bottom, left.

Edges are graded on a four-level scale: clean → minor wear → wear → damaged. Edge wear is what most graders notice first when the card comes out of the binder.

04
scratchesvisible
print linesnone
indentsnone
stainsnone

Surface

Scratches, print lines, indents, stains.

The hardest pillar to judge from a photo because lighting matters. We flag what we can see and tell you what we can't — better honest uncertainty than false confidence.

FIVE GRADERS, FIVE OPINIONS

The same card. Different grades.

Cards rarely grade identically across services. Gemmr predicts all five because the question 'should I submit?' depends on which service you'd submit to — and which gives you the best return for your fee.

PSA
1–10

Most recognized brand. Strict on centering and surface.

CGC
1–10

Fast turnaround. Slightly more generous on edges than PSA.

BGS
1–10

Sub-grades for each pillar. Toughest on centering.

SGC
1–10

Vintage-focused. Often grades half a point higher on older cards.

TAG
1–1000

Granular. AI-assisted. Returns sub-grades down to the integer.

THE REPORT

What lands in your dashboard.

Every scan returns a full report you can reference, share, and save as a PDF.

01

Five grade predictions

One predicted grade per service, each with a confidence range. So you can see 'PSA 9 (likely)' next to 'CGC 9.5 (possible)' on the same card.

02

Four-pillar breakdown

Each pillar's sub-grade, the specific notes the grader wrote, and a confidence level. You can see exactly why the verdict landed where it did.

03

Submit / borderline / don't submit verdict

The bottom-line question, answered. We tell you whether the math says this card is worth the grading fee.

04

PSA pop data

For PSA slabs, we pull the live population count from PSA's public database, so you can see how rare your grade actually is.

05

Shareable URL

Every report has its own URL with optional summary or full detail view, ready to drop into a sales listing or send to a buyer.

06

Clean PDF export

Paid plans get a watermark-free PDF you can save, print, or attach to a marketplace listing.

FAQ

The honest questions.

What collectors actually ask when they hear 'AI grading.'

How accurate is Gemmr?
Gemmr is a pre-grading tool, not a grading service — we don't put your card in a sealed case, and we never will. What we do is tell you what PSA, CGC, BGS, SGC, and TAG would most likely return so you can decide whether the submission fee is worth it. Accuracy keeps improving as more users verify their real grading outcomes back to us — every verified result sharpens the next prediction. On clean, well-lit photos today, the predicted grade lands within a half-grade of the real one on the majority of submissions, and the submit / borderline / don't-submit verdict is right more often than that. With enough verified data, our predictions get better than any single grader's first read. We track our accuracy publicly on the accuracy page — we don't hide the misses.
What if you're wrong?
It happens. We track our own accuracy publicly (the 'accuracy' page) using a verification system where users report back what their cards actually graded as. We don't hide the misses. The system is better when the photos are sharp and well-lit; the closer to a flatbed scanner you can get, the closer our prediction lands.
Do you actually grade the card, or just read the slab label?
We crop the label out before grading. The grader physically cannot see PSA's 'MINT 9' or BGS's '9.5' — it only sees the card itself. Our prediction is independent. If we predict a PSA 9 on a slab that came back as PSA 10, that's our miss, not us copying.
Should I always submit when Gemmr says 'submit'?
No. 'Submit' means the predicted return justifies the fee for THAT service at THAT card's current market value. If you're not sure your card is worth what you'd pay to grade it, the answer is don't submit regardless of what we say. Use the prediction as one input, not a directive.
What about cards we don't fully support?
We grade most major TCGs and sports cards. Cards outside our supported categories will still go through the pipeline, but identification accuracy drops because we're working with less context. We add categories as the catalog data matures — Discord is the fastest way to request one.

Know the grade before you pay for one.

Start free. 75 Gems per month — about 7 grade scans, 10 centering checks, 75 price checks, or any mix. No credit card.