Accuracy by Grade Level
Accuracy by Card Era
Sub-Grade Accuracy
Verified Examples
Submit Your Verified Grade
Got your card back from PSA, BGS, or CGC? Help us improve by reporting the actual grade. Enter your CGK cert ID and the official result.
Optional: sub-grade breakdowns (if your label has them)
Methodology
How Our AI Analyzes Cards
Each card goes through a two-stage pipeline. First, OpenCV computer vision detects the card boundaries and performs a perspective correction to get a normalized top-down view. Then Claude Vision (Anthropic's multimodal AI) analyzes the corrected image for corners, edges, and surface quality. The two systems run concurrently for speed.
The 4 Sub-Grades
- Centering -- Measured by computer vision. We detect the inner art border and calculate the left/right and top/bottom border ratios. Converted to a 1-10 grade using PSA's published thresholds (50/50 = 10, 55/45 = 9, 60/40 = 8, etc).
- Corners -- Analyzed by AI vision. Checks all four corners for whitening, dings, bending, rounding, or wear. Even minor whitening drops the grade.
- Edges -- Analyzed by AI vision. Looks at all four edges for whitening, chipping, rough cuts, nicks, or any damage along the edge.
- Surface -- Analyzed by AI vision. Checks for scratches, scuffs, print lines, ink spots, haze, indentations, and other surface imperfections.
Capping Rules
The final grade is the floor of the minimum sub-grade. If your centering is a 10 but your corners are a 7.5, the final grade is 7. This mirrors how PSA works -- a single weak area caps the entire grade. The final grade is always an integer between 1 and 10.
Known Limitations
- Surface assessment from flat photos. A top-down photo cannot reveal every surface scratch, especially subtle ones visible only at certain angles. Tilted/angled supplementary photos help but are not mandatory.
- Holographic print lines. Print lines on holo cards are visible only under specific lighting conditions. Our AI may miss them if the photo doesn't catch them.
- Image quality matters. Blurry photos, heavy shadows, reflections, or low resolution can degrade the analysis. Good even lighting from directly above gives the best results.
- Edge cases with vintage cards. Older cards (Base Set, Jungle, etc.) have different printing tolerances and wear patterns that modern AI training data may not fully capture.
- Back of card. Currently we focus on the front face. Back centering and back surface defects are analyzed only when the user provides a back photo.