CardFax

Closed-Loop Grading Software

The only Card Grading Software that learns.AI Powered by MOSES

CardFax is not another AI grading tool. It is a closed-loop grading engine built on MOSES, our proprietary model — thousands of calibrated data points sharpen it with every card it grades, and deterministic math turns its measurements into a grade you can verify. Every card is registered on the CardFax Registry for life.

Konnor Griffin — CardFax 10
Gem Mint10

Konnor Griffin

2026 Topps Now #154

Serial 5437410

CardFax
CardFaxSerial 5437410
10Gem Mint

The 4Q

Cen10
Cor9.7
Edg9.7
Sur10

Model: MOSES · Registered on the CardFax

Konnor Griffin

2026 Topps Now #154

View Card

Click the card to flip it · live from the registry

$ cardfax grade --serial 1935851

[scan]card detected · isolated · 8 corners cropped

[moses]TL 10 · TR 10 · BL 9.5 · BR 10 · library match confirmed

[centering]front 58/42 · back 68/32 · set priors applied · 9.5

[edges]9.5 · surface 10 · no print lines detected

[loop]evidence captured · 8 data points into the training loop

[registry]registered on the CardFax · population updated

CARDFAX 10 · GEM MINT

1,711+

Cards graded and registered

12,000+

Corner measurements feeding MOSES

4

Analysis layers per corner

58s

Average grading time

The Engine

Meet MOSES.

Model for Optical Scanning and Evaluation System

There is a new AI grading app every week, and the pitch is always the same — “micron-level precision,” “trained on thousands of cards,” “eliminates human bias” — adjectives with no numbers attached, wrapped around someone else's general-purpose AI. Ours is arithmetic. MOSES — the Model for Optical Scanning and Evaluation System — renders 32 corner verdicts on every card (8 corners through 4 analysis layers), measures centering against published buffers (60/40 front, 75/25 back, weighted 80/20), and averages 58 seconds from scan to registered grade — measured in production, down from 223 seconds one engine release ago. Behind it sit more than 2,000 hand-calibrated corner samples we own and 12,000+ production corner measurements, refreshed by the training loop every 30 minutes. Every claim in this paragraph has a number attached — and every grade is auditable on the public registry.

Sub-Pixel Centering Analysis

Border ratios measured against published buffers — 60/40 front, 75/25 back, weighted 80/20 — with diamond-cut tilt detection at a 3% corner delta. The formula is public. Check our math.

Micro-Surface Inspection

Enhanced-signal analysis detects hairline scratches, print lines, and indentations invisible to the naked eye — then severity is scored in the natural view, front and back alike, with magnified quadrant review.

Instant Verification

58 seconds on average from scan to registered grade — measured in production — and the result is an immutable condition record on the CardFax, verifiable by anyone, for life.

Just another AI app

  • Rented, general-purpose AI under the hood
  • “Micron-level precision,” “95% accurate” — no methodology published
  • Predicts what a human grader might say
  • Answers drift with lighting, angle, and luck
  • No public track record

CardFax + MOSES

  • Proprietary engine and 2,000+ calibrated reference samples we own
  • Published math — buffers, caps, and formula on the Standards page
  • Grades against its own standard, and registers every result
  • Deterministic — same card, same grade, every time
  • 1,711+ auditable grades on the public registry, plus a release log

Anyone can take a picture and ask an AI to grade a card. Only one has a model forged by professional graders — MOSES — trained on a proprietary data library thousands of records deep, against published whitepapers and formulas, with every release on the record.

The Closed Loop

Software that learns. Grades that get sharper.

Most grading software is frozen the day it ships. CardFax runs a loop: grade, record, calibrate, release. The data never leaves the loop, and the loop never stops.

01

Grade

Every card is measured — eight corners, border geometry, edge and surface signal — and the evidence behind each subgrade is captured, not discarded.

02

Record

Those measurements land in proprietary datasets: a curated corner library, similarity indexes, and examiner feedback. Thousands of calibrated data points, growing with every card.

03

Calibrate

Blind calibration keeps the model honest — examiners review corners without seeing the machine's answer, and every disagreement becomes training signal.

04

Release

Improvements ship as versioned engine releases, on the record. Every card graded after a release runs on a sharper MOSES than the card before it.

The loop sharpens the future, never rewrites the past. A grade issued is a grade kept.

See the release log →

The 4Q

Four quadrants. One grade. Zero opinions.

Diamond grading has its 4Cs. Card grading has the 4Q — the four quadrants of condition MOSES measures on every card. Each quadrant is scored independently, every score is math, and the weakest quadrant caps the grade.

Q1

Centering

Border geometry is measured against set-specific priors — the engine knows how wide this set's borders should be before it ever scores your card. Front and back are held to published buffer standards and weighted deterministically.

Q2

Corners

MOSES, our proprietary model, runs every corner through a four-layer ensemble: object detection, similarity matching against our proprietary sample library, an AI examiner layer, and silhouette geometry measured on the corner outline itself.

Q3

Edges

All four edges are graded independently, then combined as an average capped by the lowest edge — a single rough edge can never hide behind three clean ones.

Q4

Surface

A two-dimension doctrine: enhanced signal detects whether a defect exists, natural view scores how severe it is. Front and back get equal treatment, including magnified quadrant review.

Deterministic by Design

The final grade is math. The same card gets the same grade — every time.

Subgrades feed one published formula. The lowest subgrade caps the ceiling, the surface grade sets a hard floor cap, and a Gem Mint 10 requires a subgrade average above 9.5. No mood, no Monday-morning grader, no mystery.

Read the full grading standards →

// The CardFax final grade formula

final_grade = min(

preliminary_grade,

lowest_subgrade_cap,

floor(surface_grade)

)

// avg > 9.5 → Gem Mint 10

The CardFax Registry

Registered on the CardFax. For life.

Every grade produces a CardFax — the card and its condition analytics in one permanent public record: the subgrade measurements, the evidence behind them, population context, and ownership history, designed to be shared. The record on the right is live — pulled straight from the registry, not a mockup — and the CardFax app puts the same analytics in your pocket.

Look Up a Serial Number

The CardFax App

The grading engine, in your pocket.

Point your camera at a card and MOSES takes it from there — corners measured, grade computed, card registered on the CardFax. Your collection, its condition analytics, and every card's population data travel with you.

Coming soon to the

App Store

Coming soon to

Google Play

Grade from your phone today →
CardFax
+
CardFax 10 · Serial 5437410
10Gem Mint
Cen10
Cor9.7
Edg9.7
Sur10
CardFaxPopMarketHistory

CardFax Registry · Model: MOSES

Konnor Griffin #154

Registered on the CardFax · 2026 Topps Now