Technology
From raw scan to registered grade in under a minute.
The CardFax engine is a pipeline of specialized measurement systems — one for each quadrant of the 4Q — feeding one deterministic formula, wrapped in a closed loop that makes the whole pipeline sharper with every card it grades. Here is exactly how it works.
- 01
Capture and Isolation
Every submission starts with detection models that find the card inside the scan and isolate it precisely. All downstream crops are anchored to the card itself — not the photo — so a slightly off-center scan never distorts a measurement. Oversized scans are decoded within strict memory bounds so quality is never traded for speed.
- 02
Set-Aware Centering
Centering is measured as border geometry, guided by priors for the specific set being graded — the engine knows how wide a set's borders run before it measures yours. Front and back are held to published buffer standards, weighted 80/20, with tilt detection that catches diamond-cut miscuts by comparing corner deltas.
- 03
MOSES — The Model
MOSES — the Model for Optical Scanning and Evaluation System — is our proprietary assessment engine. Each of the eight corners (front and back) runs through its four-layer ensemble: a trained detection model flags defects, a similarity layer compares the corner against our curated and continuously calibrated sample library, an AI examiner layer reads the corner in context, and a silhouette-geometry layer measures the corner outline itself. The layers vote; disagreements are resolved by deterministic rules, and the method used is recorded with the grade.
- 04
Edges and Surface
All four edges are graded independently and combined as an average capped by the lowest edge. Surface follows a two-dimension doctrine: enhanced-signal analysis establishes whether a defect exists, and the natural view scores its severity — including a dedicated speck detector and magnified quadrant review on both front and back.
- 05
The Closed Loop
CardFax is a learning system. Every measurement, every calibrated corner, and every audited identity flows back into proprietary datasets — the corner library, similarity indexes, and examiner feedback. Blind calibration keeps the model honest, and improvements ship as versioned releases on the release log. The loop sharpens future grading only: grades already issued never change.
- 06
Deterministic Final Grade
Subgrades feed one published formula: the final grade is the minimum of the preliminary average, the lowest-subgrade cap, and the floor of the surface grade. The same inputs always produce the same output, and every grade is fully auditable back to its measurements.
- 07
Registered on the CardFax
The grade, the subgrades, and the measured evidence behind them are permanently registered on the CardFax Registry, along with population data and ownership history. A grade issued is a grade kept — formula improvements apply to new gradings only and are never silently applied to cards already registered.
Against the Pre-Screen Apps
Pre-screen apps stop at a prediction. CardFax is the grade.
- Their special capture mode is our default. Pre-screen apps treat macro corner close-ups as a separate scan you have to choose. Every CardFax grade runs all 8 corners through all 4 MOSES layers — 32 corner verdicts on every card, every time.
- “Does not automatically retrain” is not a feature. A frozen model drifts behind the hobby. The CardFax loop learns continuously — gated by blind human calibration, shipped as versioned releases, on the record in the release log.
- A confidence score is a hedge. A measurement is a fact. Pre-screens attach a confidence score to a guess about what a human grader might say. MOSES publishes the measured evidence behind a deterministic grade — the same card gets the same grade, every time.
- A report download expires. A registration does not. Pre-screens end in a PDF. A CardFax grade ends sealed in a Black Label Slab and registered on the CardFax — permanent, public, and verifiable by anyone.
See the standards behind the engine
The full grade scale, the centering buffers, and the final grade formula are published and versioned. Transparency is the product.