CardSight

AI submission advisor

Know if a card is worth grading before you pay to grade it.

A guided phone scan becomes an explainable grade probability, verified market values and the grading math, so you can see the PSA 10-vs-9 cliff before you send it in.

Scan a cardYour scans

Capture modes

Quick check

Low to medium

Front + back. Card ID, obvious defects, rough centering.

Guided grade

Medium to high

Front, back, angle shots and corner close-ups. The main grading workflow.

Guided grade + video

Highest on a phone

Adds a short light-sweep video: best surface evidence and 9-vs-10 calls.

How it works

  1. 01

    Capture

    Guided front, back, angle and corner shots. Bad photos are rejected on the spot, not graded.

  2. 02

    Measure

    The card is rectified to a top-down geometry; centering, corners, edges and surface are measured.

  3. 03

    Compare

    Where a licensed reference exists, your card is aligned to it and differences are mapped.

  4. 04

    Predict

    A probability for each grade, with the measurements that drive it. No single forced number.

  5. 05

    Decide

    Verified completed sales and fees turn probabilities into Submit / Borderline / Hold math.

Refuses to guess

If the evidence is weak, you get 'Insufficient evidence' and a retake prompt, never fake precision.

Measurements first

OpenCV and vision models produce the numbers. The language model only explains them.

Real sales only

Comparable values come from completed sales with provenance. Asking prices are never shown as sales.

No enhancement

Photos are never generatively 'enhanced' before grading. What wasn't photographed isn't invented.