How CardGradeCheck Grades Your Cards

No black box, no "proprietary secret sauce." Every rule, every tolerance, every capping decision — on this page. If you disagree with a grade we gave your card, this is where you check our work.

Last updated: 2026-11-14

The Short Version

We grade the four axes professional graders grade — centering, corners, edges, surface. We compare your card against the published (and community-reverse-engineered) standards for PSA, BGS, CGC, SGC, and TAG. We output a predicted grade per grader, plus a confidence score per axis, plus the specific reasoning behind the number.

We can't physically touch your card. We work from photos. That means we're very good at some things (centering, visible whitening, obvious damage) and limited at others (sub-pixel corner fuzzing, angle-dependent holo scratches, 3D defects). We tell you where we're confident and where we're guessing.


The Four Axes

Every grader — PSA, BGS, CGC, SGC, TAG — grades the same four things. Weighting differs; the axes don't.

1. Centering

What it measures

The ratio of left-to-right and top-to-bottom border widths on the front, and the same on the back.

How we measure it

We use deterministic pixel-measurement (OpenCV contour-detection + ratio math). A real grader eyeballs borders and reaches for calipers on edge cases. Our pixel math is, on this axis specifically, more accurate than a human eye — and more consistent run-to-run.

What a grader looks for

PSA 10 requires 55/45 or better on the front, and 75/25 or better on the back. BGS 10 Pristine requires 60/40. BGS 10 Black Label (the unicorn grade) requires 50/50 exactly.

GradePSA FrontBGS 10 PristineBGS Black LabelAce 10
1055/4560/4050/5050/50 ±2%
960/40——55/45
865/35——60/40
770/30———
6 and belowworse than 70/30———

Our capping rule

If front centering is worse than 65/35, we cap our PSA prediction at 8 regardless of other axes. Worse than 70/30 caps at 7. This mirrors the industry rule.

What we're confident about: the centering ratio. Our confidence here is typically ±1%.
What we warn about: cards shot from an angle (perspective distortion makes centering look worse than it is). We detect perspective skew and ask you to re-shoot if the front isn't flat-on.

2. Corners

What it measures

The sharpness, fraying, whitening, and rounding of all four corners, front and back.

How we measure it

We run high-resolution edge-contrast analysis on each corner region. Whitening shows up as a lightness-delta where the sub-layer is exposed. Rounding shows up as a geometric-angle deviation from 90°.

What a grader looks for

A 10x jeweler's loupe is held over each corner under 5500K light. PSA 10 requires sharp 90° corners with no visible whitening. One corner with sub-millimeter whitening typically caps at PSA 9. Two corners cap at PSA 8. Chipping caps at PSA 6. Pronounced rounding caps at PSA 5.

Our capping rule

The worst corner sets the ceiling. A single damaged corner on an otherwise perfect card will cap the overall grade at that corner's score. This is an industry-standard rule, not a CardGradeCheck invention.

What we're confident about: visible whitening (≥~1mm). Visible chipping. Rounded apex vs. sharp apex at our image resolution (~1000+ pixels per corner region).
What we warn about: sub-pixel fuzzing. A grader with a 10x loupe can see corner softness smaller than our image resolution can resolve. When our corner score is 9 or 10, we add a confidence caveat: "sub-pixel sharpness cannot be verified from a photo — assume PSA 9 ceiling until 10x inspection."

3. Edges

What it measures

Chipping, whitening, nicks, and fraying along all four edges, front and back.

How we measure it

Edge-contrast analysis along the border strip. Whitening appears as a lightness spike against the expected colored border.

What a grader looks for

10x loupe tilt-scanning each edge under angled 5500K light. Full chipping caps ≤ PSA 6. Visible whitening along ≥25% of one edge caps at PSA 7. Minor whitening caps at PSA 8–9. Clean edges under 10x = PSA 10 eligible.

Pokemon-specific note: vintage 1999–2003 Base Set holos often show "factory silvering" — tiny shiny flecks from the factory cutting process. This is not wear. Graders distinguish factory silvering from edge damage and don't always penalize it (though heavy silvering can still block Gem Mint). We try to detect factory silvering vs. wear but the distinction requires domain knowledge we don't fully encode. If your card is vintage Base Set and we score edges low, cross-check whether the issue is factory silvering before paying for a submission.

Our capping rule

Visible edge whitening on ≥25% of any edge caps at PSA 7. Chipping caps at PSA 6.


4. Surface

What it measures

Scratches, print lines, print dots, staining (wax / ink), gloss, holo scratches, fingerprints, indentations (binder dings), and creases.

How we measure it

This is the hardest axis for an AI working from a flat photo. We analyze the uploaded image for contrast-lines (scratches), repeating patterns (print lines), localized light anomalies (print dots), and shadow-lines across the surface (creases).

What a grader looks for

Raking light at 45° under a 5500K lamp, tilted and rotated slowly, with a 10x loupe. The grader watches the holo reflection change as the card tilts — print lines and holo scratches pop at specific angles that a single flat photo won't capture. Surface is the axis where real graders flex most and where two graders are most likely to disagree.

DefectTypical CapNotes
Crease / foldPSA 5 or lowerStructural, near-universal
Deep scratch through holoPSA 7Visible without magnification
Print line (uniform, across holo)PSA 9Most common PSA-10 killer on modern holos
Print dot / ink speckPSA 9 (minor), PSA 8 (prominent)Depends on location
Light holo scratchPSA 9Only visible under angled light
Wax stain back-onlyPSA 9Front staining caps harder
Binder dentPSA 6–7Dimensional defect — we usually miss these
Fingerprint smudgePSA 8–9Graders don't clean
Factory silvering (minor)PSA 10 possibleFactory artifact, not wear
Loss of glossPSA 8 or lowerFull gloss required for PSA 10

Our capping rule

Detected crease → max PSA 5. Detected deep scratch → max PSA 7. Detected print line on holo → max PSA 9.

What we're confident about: creases (shadow-line signature). Deep scratches (contrast-line signature). Major print lines on holo if the photo captures them at the right angle.
What we warn about — every single scan: fine print lines, light holo scratches, binder dents (3D), and angle-dependent reflection defects. These need a loupe or multi-angle inspection. Surface is our widest confidence interval, every time. If your card has anything borderline, we default surface to ±1 grade uncertainty unless you upload a second photo at an angled light.

The Capping Rules — All in One Place

We openly publish these because the other AI pre-graders don't. If any of these fire, our final predicted grade cannot exceed the cap, regardless of how good the other axes look.

TriggerMaximum PSA Grade
Centering worse than 65/35 (front L/R)PSA 8
Centering worse than 70/30 (front L/R)PSA 7
Any single corner shows visible whitening ≥ ~1mmPSA 9
Any two corners show visible whiteningPSA 8
Any corner shows chipping or missing layerPSA 6
Any corner shows pronounced roundingPSA 5
Visible edge whitening on ≥25% of any one edgePSA 7
Edge chipping / missing materialPSA 6
Detected crease / foldPSA 5
Detected deep scratch through holoPSA 7
Detected print line uniform across holoPSA 9
Detected loss of glossPSA 8

Final grade logic

Our predicted overall grade is the worst axis sub-score, rounded down, with no bonus for the other three axes being perfect. This mirrors the BGS "lowest sub-grade anchors final" rule and matches PSA's internal head-grader rollup.

We do not apply the "all-four-close bonus" that BGS uses to bump 9.5 → 10 Pristine. We would rather under-predict and under-promise than inflate a borderline card into a disappointment at the slab return.


What We Nail

The axes where our AI is, on average, as good as or better than a rushed human grader on a 60-second standard-tier pass:

  • Centering ratio. Deterministic pixel measurement. More accurate than an eyeballed read, more consistent run-to-run.
  • Major corner damage. Whitening ≥ 1mm shows up as a lightness delta we don't miss.
  • Edge whitening. Same contrast-delta logic as corners.
  • Obvious surface scratches. If the photo captures them, we catch them.
  • Creases and folds. Shadow-line signature across the surface.

What We Miss

The axes where a human with a loupe and a 5500K raking light will beat us:

  • Fine print lines on holo. Requires physical inspection at the right angle. We flag these as "surface ±1 uncertainty" every time.
  • Sub-pixel corner fuzzing. Visible under 10x loupe, invisible to a photo.
  • 3D defects. Binder dents, warping, raised ink. Flat photo can't see depth. Multi-angle flow partially helps; a video scan would help more. We're working on it.
  • Light holo scratches. Angle-dependent. A single top-down photo misses most.
  • Factory silvering vs. edge wear. Distinction requires domain knowledge we partially encode but don't perfect. We warn on pre-2003 Base Set cards specifically.

Our Current Accuracy

See the live accuracy dashboard for rolling 30-day numbers, updated nightly:

  • Overall grade exact-hit rate
  • Overall grade within ±1 grade
  • MAE (mean absolute error) across all grades
  • Centering tolerance (vs. ground-truth calipers)
  • Sample size

Every number on that dashboard is sourced from our feedback-form submissions (you tell us the real grade) plus public PSA/CGC cert-lookup cross-validation. We update nightly. When accuracy improves, we show you. When it regresses on a new set release, we show you that too.

Our launch-week numbers will be smaller and noisier than our month-6 numbers. That's not a bug, that's statistics. Treat the first two weeks as beta and give us feedback.

We're Not PSA / BGS / CGC / SGC / TAG

CardGradeCheck is an independent AI pre-grading tool. We are not affiliated with PSA, Beckett, BGS, CGC, SGC, TAG, Ace Grading, or Nintendo / The Pokemon Company. We do not slab cards. We do not issue certificates. Our pseudo-cert /verify/<id> link is a scan record — it is not a grading certificate and has no resale value.

Our predictions are opinions based on published standards and image analysis. The only grade that counts is the one on the slab.


FAQ

Is this official grading?
No. We're a pre-grading tool. We estimate what PSA / BGS / CGC / SGC / TAG will probably return. The only real grade is the one on the slab.
Can I resubmit with a different photo and get a different number?
Yes — and that's useful information. If our grade shifts a lot between your photos, the card is borderline and the real grader will probably disagree too. If it's stable across 3+ angles, you have a defensible read.
What if I disagree with the grade you gave me?
Use the "too-high / about-right / too-low" feedback button on the result page. If you later submit to a real grader, come back and enter the cert# — we'll update our accuracy stats. Your feedback literally makes the model better.
Do you keep my uploaded photos?
We retain scans for 30 days to compute the accuracy dashboard, then delete the image. The scan record (grade, timestamp, no photo) is retained indefinitely so your /verify/<id> link stays live.
How do you handle holofoil cards specifically?
Holos get extra surface-axis scrutiny. The #1 PSA-10 killer on modern holos is fine print lines across the holofoil, which we can only detect if your photo captures them at the right angle. We always add a "surface ±1 uncertainty" caveat on holo cards unless you upload a second photo at an angled light.
Why don't you auto-identify which card this is?
We will — in a future update. For now we ask you to type the card name + set so we can pull live eBay sold-comps and show a realistic market value. Auto-recognition is a Vision-API cost we don't want to pass on to a free tool.
Why is this free?
Because the market is full of tools that gate the first scan behind a signup or a daily cap. Unlimited free scans is our wedge. We may add an optional Pro tier later for features like batch upload and multi-card collection tracking, but single-card scanning will always be free.
Can I use your prediction to price my card for sale?
Use the cost calculator on the result page. It pulls live eBay sold-comps for both raw and graded prices and gives you a break-even and 3x ROI threshold. If the math says don't grade it, don't grade it.

Sources + Further Reading