1,838 Card Sales, 16 Passed the Filter

The set-tracker videos on Binderdex answer one question per card: what has it done since the set launched? To answer honestly you need a launch-week baseline of real raw sales, each one matched to the exact card. Last week the matcher doing that matching accepted 16 sales out of 1,838. The videos would have told you the market went quiet, when really the filter had gone blind.

The rejection ledger told the whole story:

Release week, one set, 1,838 raw sale listings:
  1,071 rejected: no condition named in the title
    ~660 rejected: seller repeated the card number
     ~60 rejected: rarity words (SIR, "Anniversary")
     29 rejected: "280 HP" read as heavily played
     16 accepted

Sellers write titles like "Mew ex 232/193 PSA 10 GEM MINT 10 Slab" or "Mew ex #151/165 Holo NM" or "2 PSA 10 Mew ex 232/193". Every one of those broke a rule somewhere: the "10" inside GEM MINT 10 looked like ten slabs, the # looked like a second card number, the leading 2 looked like a bundle. A dozen review rounds chased these, and each round fixed one title and broke another. Regex whack-a-mole, with the market's full vocabulary as the opponent.

The fix was to stop patching and go structural: strip the filler you can name (grade tokens, hit points, cert numbers, set names, years, languages), then decide whether what's left is one card or a list. One shared helper now backs both the slab filter and the raw-sale matcher, and graded baselines only pool for single-printing cards. After the rewrite: 533 of 1,838 accepted, and the cards with a sales-backed launch baseline went from 1 of 40 to 20 of 40.

The number I care about most: the top-10 medians didn't move. Loosening a matcher should widen coverage without changing the answer. When it does change the answer, you had a bug, not a judgment call.

If you match free-text listings to a catalog, steal the sequence: measure rejections with reasons first, fix the structure second, and re-run the old numbers every single time. The ledger is the feature. It's the difference between "our filter is strict" and knowing exactly which 99% you're throwing away.