How accurate is the prediction?
AuraGrade publishes a target accuracy band and residual notes; the underlying scoring pipeline is described in detail at /how-it-works.
About 3 in 4 predictions land within one PSA tier
On scans that pass our QC gate, AuraGrade's PSA-equivalent prediction matches the eventual grader call within ±1 PSA tier roughly 3 out of 4 times — if we predict PSA 9, you most often receive PSA 9 or PSA 10. The remaining ~1 in 4 may land a full tier away. That gap costs money: a $1,000 card predicted at PSA 10 that settles at PSA 9 can drop toward $400. Use AuraGrade as a pre-screen, not a valuation. This range is based on our internal test cards and will be refreshed with live data after the first 50 real PSA returns. Terms §2 carries a non-quantitative version of the same disclosure; your sole remedy if a prediction differs from the eventual grade is governed by Terms §8.
Where the error comes from
(1) Camera-hardware ceiling: phone optics cannot reliably resolve hairline scratches, which are the common wedge between PSA 10 and PSA 9; (2) Inter-grader consistency: PSA / BGS / CGC disagree on roughly 10% of borderline cases; AI prediction inherits this ceiling; (3) Model confidence: borderline 8.5↔9 calls flip more often than core-band predictions — users on "Low" and "Medium" confidence should keep a wider buffer.
How to lower the residual
Use a flatbed scanner (600 DPI). Uniform background. No glare. Tilt < 5°. Capture both sides. Pro subscribers can keep originals in Aura Vault for later re-analysis. Our scoring engine is deterministic — every deduction in your report carries a specific rule code and a bbox evidence image; email support@auragrade.com if you want a walkthrough of how a particular grade was computed.