Grade only when the math clears.

Buy underpriced raw basketball cards from eBay and local shops, run every candidate through the verdict engine below, submit only the winners, sell the slabs. Grading is a capital-allocation decision — never the default.

Grading Calculator

Enter a candidate before you buy it, not after. Verdicts follow the Grading SOP: expected profit ≥ $25, ROI ≥ 15%, and a PSA 9 result should roughly get the money back.

Card Source Grading lane Raw cost $ PSA 10 comp $ PSA 9 comp $ Gem probability % Photo-only screen (no card in hand) — applies the −10 pt gem haircut

Fee assumptions

Regular fee $ Express fee $ Checkout fee $ (log real fee when seen) Sub shipping share $ Selling fee % Outbound shipping $

Batch Queue

$0 / $1,000 grading cap

Cards that passed the calculator, waiting on the batch rules: minimum 5 passing cards, grading spend under $1,000 per cycle.

Card	Lane	Raw	Exp. profit	ROI	Verdict

Nothing queued yet. Run a candidate through the calculator and save it.

House Rules

Buy

Sweet spot: raw $120–500, PSA 10 comp ≥ 4× raw.
Sub-$100 raw stays in the quick-flip pile — the fee eats it.
Front + back photos or no bid. Never raise a bid late.

Screen

Reject: dimples, print lines, scratches, wax, soft corners, whitening, centering past 60/40.
Photo-only buys take a −10 pt gem haircut.
Every rejected card sells raw immediately.

Submit

Regular by default. Express when the slab should top $1,500 (insurance cap) or the fast cycle pays.
5+ passing cards per batch, grading spend < $1,000.
Never pad a batch with marginal cards.

Sell

List every slab within 24h of arrival.
Gems at mid-comp. Misses priced to exit.
Log every grade vs. the predicted gem odds — that calibration is the moat.

Build Roadmap

Live now — the desk. Lane-aware grading calculator, SOP verdict engine, batch queue with the $1,000 cap.
Next — the scanner. DeepSeek bulk pass over live raw listings: title-vs-photo mismatch, seller quality, photo-quality gate. Kills ~95% of listings before a human looks.
Then — the eyes. Centering measurement and corner scoring from listing photos; confidence capped by photo sharpness.
Then — the boss. Claude verdicts on the survivors: comp sets for both exits, lane pick, EV, and a ten-second "why" per card. Push alert only on clear buys.
Always — the loop. Every returned grade scored against its prediction. After 2–3 batches the gem estimates stop being guesses.