PRECISION
BY LOGIC.
Should you send this card to PSA? A submission ties up $25–$600 in fees and up to 7 months of capital lock. Only 43% of submissions return a Gem Mint 10. AuraGrade tells you which side of that line your card is on — before you ship.
Garbage in, garbage out. Never.
A client-side validator gates every upload at the edge before any inference cycle is spent. Sub-threshold captures are rejected with a single actionable retake instruction. Image or video — the engine accepts both.
Required Captures PER SCAN
Quality Thresholds
Evaluated client-side via WebGL/HTML5 Canvas. No network round-trip required to reject a bad capture.
Edge Validator
Live // 60 HzRetake Feedback
Failed captures return one actionable instruction — never a generic error. Examples returned by the validator:
Record an 8-second walkaround. Skip the retake loop entirely.
A slow rotational pass captures every angle in one motion. Our frame-extraction pipeline scores each frame on sharpness, glare, and perpendicularity — then picks the optimal frame per pillar. Multi-frame fusion cancels specular reflections that defeat single-image captures, which is what makes surface-defect recall so much stronger here than on competitor platforms.
The Four Pillars of Grade
A ConvNeXtV2 backbone with separate Multi-Task Learning heads for each pillar. Each pillar is evaluated independently, then the Score Aggregator enforces PSA's hard-cap logic before a final distribution is returned.
Centering
Algorithmic pixel counting via Sobel edge detection — not a deep-learning guess. Enforces PSA's 55/45 front threshold for Gem Mint 10 with mathematical certainty.
Corners
Dedicated MTL head trained on 20K hard-negative crops. Detects micro-fraying and whitening at radii invisible to the naked eye.
Edges
Perimeter segmentation identifies silvering and rough factory cuts on vintage stock — defect classes most consumer-grade graders miss entirely.
Surface
Trained against synthetic glare and shadow augmentations so the network learns invariant defect features — dimples, print lines, scratches — even under bad smartphone lighting.
The 3-grade chasm your submission is rolling against.
The final grade is the floor of all four pillars — one bad corner caps the whole card. Only 43% of PSA submissions return a Gem 10. The valuation gap between a 10 and a 9 on the same card routinely runs 3× to 10×. AuraGrade tells you which band you're holding.
A wrong submission
is months of locked capital.
PSA charges $25–$600 per card and ties up your asset for 5 business days (Super Express) to 7+ months (Value tier). 57% of submissions return below Gem Mint 10. The expected value of submitting blind is brutal — AuraGrade screens the decision in seconds for the cost of a pack of gum.
below Gem Mint 10
per card // sub-10s
Avoiding even a single mis-submission pays for years of Pro tier.
The axes that actually matter.
First-generation AI graders validated the demand but failed the execution. We compete on the dimensions where their architectures collapse: surface-defect recall under variable lighting, sub-grade granularity, cryptographic transparency, and permanent data ownership.
A 20,000-image defensible moat.
Each label is cryptographically verified: OCR extracts the PSA cert number from the slab, then queries PSA's official Verification API to confirm the grade, set, year, and qualifiers. No seller descriptions. No mislabeled samples. No training contamination.
From pre-grader to grading authority.
The pre-grading product is the wedge. The proprietary corpus is the moat. The endgame is replacing the incumbent grading paradigm with cryptographic transparency.
AI Pre-Grader
SaaS pre-submission filter. Every paid scan generates cash flow and crowdsources real-world user images into the proprietary training corpus.
Hybrid Encapsulation
High-confidence pre-scans get invited to mail in for physical processing under controlled multi-spectral lighting. AI does 95% of evaluation; human technicians do final QA + sealing.
Defect Map →
Cryptographic Authority
Every slab carries a QR linking to a permanent on-chain ledger entry with the pixel-level defect map and per-pillar reasoning. Algorithmic transparency replaces the opaque legacy integer.