scratchme

Collectible intelligence

Condition, price and sentiment analysis, built for institutional investors in alternative assets. We're developing a vision foundation model to read condition.

Augmenting human vision

1 · Centering

Measuring borders

Computer vision traces the cut edge, squares the card and measures every border.

Left / right
45/55
Top / bottom
55/45

Pink lines show where the border widths were read.

The whole Topps card, with the measured lines in pink: the card's cut edge, where the printing starts on each side, and short marks where the widths were read. The card's middle-left ninth, closer: the left border beside the photo. The left border at the scan's own resolution, from the card's cut edge across the white border to the photo's frame; pink lines mark where the border's width was read. The card's top-center ninth, closer: the top border above the team name. The top border at the scan's own resolution, from the card's cut edge down to the top of the team name's red letters; pink lines mark where the border's width was read.
From the whole card to the left border, then to the top border, each at the scan's own resolution.

2 · Corners and edges

Finding edge wear

Computer vision finds the chips along the edge, so you see the wear before you buy.

Pink circles mark the chips and whitening along the edge.

The whole Magic card. The card's lower-left ninth, closer: the black border, the corner and the lower left of the text box. The lower-left corner at the scan's own resolution: pale chips and whitening along the black border's rim, circled in pink.
From the whole card to the lower-left corner at the scan's own resolution.

3 · Surface

Mapping scratches

Computer vision looks past the sparkle to find the scratches underneath.

The whole Pokémon card. The card's middle-left ninth, closer: the left part of the holographic art and the text below it. The holographic foil on the middle left at the scan's own resolution: sparkle dots on dark foil, crossed by thin light scratches.
From the whole card to the scratched foil at the scan's own resolution.

Price analysis

Price time series analysis for collectibles.

Time series
Sale prices over time, by card and grade
Volatility
How far a card's price swings around its trend
Grade premium
At the top of the scale, one grade step can multiply a card's price

Sentiment analysis

Market sentiment analysis with large language models (LLMs).

Signal
What the collector market is saying, analyzed over time
Catalysts
Events that move sentiment, like a reprint, a new set or a player retiring
Structural breaks
Where a catalyst shifts a card's price series to a new level or trend

Condition analysis

Our computer vision pipeline.

Centering
How evenly the borders frame the card, side to side and top to bottom
Corners and edges
Chips and whitening around the rim
Surface
Scratches and specks, even under holo sparkle