Writing a fake Voynich by rules: copies, habits and the link between words

[Mauro]

I don't know if it is what dashstofsk did, but I calculate autocorrelation like this:

On each line of a text, for each consecutive word pair:
if there are no gallows (k, t, p, f) in either word, or there are gallows in both words, increase positive
else increase negative

The autocorrelation is (positive-negative)/(positive+negative).

Just out of curiosity (I'm sure what you did works as intended), isn't usually autocorrelation defined as SUM_for_every_word_i( Pgallows(i) * Pgallows(i-k) )?
When you say 'word pair', you mean overlapping pairs I guess? As (W1, W2) (W2,W3),(W3,W4) etc.?