Measured

3 public benchmark observations across 3 boards

Estimated

1 reported rows against measured corpus ladders (61%); publisher-corpus-prior over 211 held-out anchors (39%); shrunk 4 toward the measured corpus