Grading AI — Analysis Snapshot

Upload front + back photos for automated vision analysis (deskew + lighting normalization + Claude grading), or score manually. Backend deterministic TAG-style scoring with PSA / BGS mapping.

Open Grading Viability Calculator →

1. Upload Card Photos

Each photo is deskewed (perspective-corrected) and contrast-normalized server-side before Claude vision grades it. Front is required; back is strongly recommended; an angled-light shot improves surface defect detection.

Click to choose…
Click to choose…

Close-up shots (up to 4)

Zoomed views of a single region. Skips card-detect but applies sharpening to expose micro-defects (hairline scratches, print dots, tiny edge nicks). Pick the region so detected defects get attributed correctly.

1b. (Alternative) Upload Walkthrough Video

Drop a short clip walking around the card (front → corners → back, ideally with angled-light passes for surface). Backend extracts ~6 of the sharpest, most-diverse frames via OpenCV (Laplacian-variance sharpness ranking, timeline-spread filtering) then feeds them to Claude as a unified profile. Max 500MB.

2. Card Metadata

3. Category Scores (0–1000)

Pre-filled by vision analysis when you upload photos. Tweak then click “Run Analysis” to re-grade with your manual adjustments.

4. Evidence Quality

5. Defects

None added. The engine treats missing evidence as risk, not clean condition.

Result

Upload photos or fill the form, then click Analyze. Backend at https://grading-ai-analysis.onrender.com runs preprocessing (OpenCV deskew + CLAHE) → Claude vision → grading_engine.build_engine_score.