CiP ocrdiff2 in P1
This is piggy's manual classification of each change type which occurs in the Perpetual P1 dataset. The green lines are 1-degree polynomial least squares fit to the log-log data. Think of them as the average or expected value.
Exempt
These are changes which should not be counted as part of proofing.
Perfect
These are changes which go monotonically downward.
Good
These changes have a general downward trend with no apparent floor.
Marginal
Disappears quickly
These changes may not go down, but they are gone within a few rounds. It appears that P1 handles these almost completely.
Noise Floor
These changes go down for a few rounds then reach a noise floor, appearing to make no further progress.
Bad
These changes immediately reach a noise floor or are otherwise obvious errors.
Insufficient data
These changes are too rare to make generalizations about.