Calibration log
When we say a card has a ~60% chance of a PSA 10, how often does it actually come back a 10? This is where we answer that — with verified returns, not marketing.
Early — 0 verified returns logged. We publish every one.
1,081 predictions frozen in the ledger so far (model v1.0-popmix). The log fills in as graded cards return.
How it’s built
Every prediction is frozen the moment you first see it — its inputs, model version, and the full probability distribution. Before a card ships to the grader, the prediction is pre-registered against that submission line item. When the card returns, we verify the cert and match the card identity, then join the real grade to the frozen prediction. Nothing is ever joined after the fact — a prediction and its outcome can only meet if the prediction came first.
What will appear here
As verified returns accrue, this page will publish, per grade: when we said ~X%, the grade came back Y% of the time (n=…). It will also publish its own coverage — verified outcomes divided by pre-registered submissions — and which verification classes make up the sample, so you can weigh the number honestly.
One limit we state plainly: we verify the cert and the card identity, not physical custody of your exact copy. Every return counts — the hits and the misses.