I think the estimator is using the count of all words: normal, ketiv, & qere. It should really only use normal & ketiv, since in the manuscript, qere is in the margin.

There are spuriously-large line estimate errors in a few places due to a different convention of line counting used in tanach.us change records.

The convention of tanach.us is to count non-blank lines only.

I count lines of fixed height, regardless of whether they are blank or not.

I.e. I treat the page more like a fixed grid or spreadsheet.

(This issue started as an issue in my private repository called "trope" but I will close that one, letting this one represent the issue.)

Petuḥah divisions involving a blank line are of course particularly significant.