Should you submit this card for grading? Run the math first.

Last year I pulled 47 cards from my long-tail inventory that I "thought I should probably get graded." After running the math on each one, only 11 were worth submitting. The other 36 would have lost money on the round trip — $25+ in submission fees, $5–10 in shipping, $1–3 in supplies, plus 4–8 weeks of capital tied up — for a grade that didn't move the price enough to clear those costs.

This is the most expensive mistake a card dealer can make at scale. Not the obvious "I graded a junk card" mistake — anybody can spot that. The expensive one is "I graded a good card that wasn't quite good enough." A raw modern Pokemon card that comps at $40 has to grade a PSA 10 to clear the math. A PSA 9 puts you underwater. And the difference between a PSA 9 and a PSA 10 — for most modern cards — comes down to centering and corner fray you can sometimes spot from a photo if you know what you're looking for.

Here's the math, and the new AI Card Vault feature that runs it on every raw card with a photo.

The math

Three numbers decide it. You need all three.

1. The raw comp. What does the card sell for ungraded, today, in the condition you actually have? This is your floor. If grading doesn't beat this number plus your costs, you don't grade.

2. The graded comps for each likely grade. Not just "PSA 10 sales" — what does this card go for at PSA 8, PSA 9, AND PSA 10? Each grade has its own market. A modern Pokemon card might be $40 raw, $55 at PSA 8, $90 at PSA 9, and $400 at PSA 10. The shape of that curve is what matters, not the peak.

3. The realistic grade probability. This is the hard one and the one most dealers eyeball wrong. "Looks mint to me" is not a probability distribution. If you submit 10 of your "looks mint" cards, you'll probably get back two 10s, five 9s, and three 8s. That's the data your math needs.

Plug those into a simple expected-value calculation:

EV(grade) = sum(P(grade) × graded_comp[grade]) - submission_fee - shipping - capital_cost

If EV(grade) > raw_comp, submit. If not, sell raw.

Why most dealers skip step 3

Because step 3 — calibrated grade probability — is the one that requires either a lot of experience or a system that can give you a reliable prediction. Most dealers eyeball it. They look at the card, decide "this is probably a 9 or a 10," and submit. Then they get back a 7 and curse the grader.

The graders aren't being unfair. The dealer's prediction is just uncalibrated. When you've graded a few hundred cards yourself you start to develop the eye for it — you know that the centering on this Charizard is off by 60/40 not 55/45 and that's the difference between a 9.5 and a 9. But that calibration is years of submission fees you've paid to learn it.

The AI Card Vault pre-grade feature

We just shipped (May 12, 2026) a button on every raw card's inspector page that runs a grade prediction in 1–4 minutes from your existing front + back photos. It uses CardGrade.io's grading pipeline — the same one that gave its 50,000+ users predicted grades during their open beta — and returns:

• A predicted grade with confidence (e.g. "PSA 9 · 87% confidence")
• Four sub-scores: centering, corners, edges, surface
• Cross-grader predictions — what the same card would likely score on PSA vs BGS vs CGC. Because some cards are PSA cards (modern slabs, sports), and some are BGS cards (vintage, basketball, some Pokemon WOTC), and some are CGC cards (manga, anime, modern Pokemon Japanese). The grader you pick changes your math.
• A "limiting factor" line that names the specific defect holding the grade back — "minor whitening on top-right corner," "centering off-axis 65/35 left-right." This is the part you actually act on.
• A market-value range for the predicted grade

Costs 25 credits per card (~$0.25 in beta pricing). Refundable on pipeline failure. Hidden on already-graded cards.

The point: it gives you the step-3 number you couldn't get before, in less time than it takes to brew coffee.

How I use it

The workflow on my desk, since we shipped this:

1. Identify the card and price it raw. This is unchanged — same Identify-with-AI button that's been there since launch.
2. Look at the raw comp. If it's under $20, I sell raw and don't bother with the pre-grade — submission fees eat $20 cards even at a PSA 10.
3. If raw is $20–$80, run the pre-grade. Look at the predicted grade and the limiting factor.
4. Apply the math. PSA 10 = submit. PSA 9 on a $50 raw card with $90 PSA-9 comps = sell raw (margin is thin and you take 6-week capital tie-up risk). PSA 9 on a $50 raw card with $400 PSA-10 comps but only a 30% chance of a 10 = sell raw or submit, judgment call.
5. Re-photograph and re-run if the limiting factor was photo-related. Sometimes "centering off-axis" is real and sometimes it's the camera angle. The pre-grade catches both. If the limiting factor moves between runs, that's a sign the issue is your photo, not the card.

This is not a replacement for the grader. The grader is the truth. The pre-grade is the prior — a calibrated starting point for the math.

What the pre-grade can't do

Three caveats up front:

It doesn't predict the value of the card. It predicts the grade. Pricing the graded comps is still up to PriceCharting + eBay sold comps. (We do show a market-value range, but it's a CardGrade estimate, not the authoritative comp.)

It can be wrong. The confidence percent is honest — when it says "87% confidence," that means there's a 13% chance you'll get something different. Don't bet the farm on a single pre-grade.

It's still a photo-based prediction. Hands-on grading checks for surface defects under a loupe that no AI can see from an iPhone photo. If your card has a hairline scratch or a print defect that's visible only at 30x zoom, the pre-grade won't catch it.

That said — for the modern dealer with a $40 Pokemon card asking "is this a 10 or a 9?", the pre-grade is the closest thing to a calibrated answer you can get without paying $25 to find out.

The cost frame

I priced the feature at 25 credits per card — which in the beta tier is about $0.25 per pre-grade. That's roughly 1% of a PSA submission fee. If running pre-grades on every borderline card stops me from submitting one card that wasn't quite a 10 per month, the feature has paid for itself 100x over.

That's the actual ROI question. Not "is the prediction perfectly accurate" (it isn't, and that's fine), but "does it tilt my submission decisions in the right direction often enough to justify a quarter per check?" The math says yes.

— Jamie