Going back to the encoded text, for a moment...
I just tried adding totals from all the "chunks" together, without much success. Going at it paragraph-by-paragraph, I took either the position in the "halfhabet" (first time) as well as total number of characters in line up to that point(the second time), and then either subtracting or adding that number to a total depending on "polarity". I haven't attempted to use a mix of both halfhabet values and line positions.
As soon as I decided to post my results here, I realized my definition of a "chunk" was so poor that nobody could hope to agree to it. The flow of text is interrupted by images or quotations as often as by the end of a post, and indeed it is probably best to assume a continuous stream of lines.
That got me thinking about whether there are any truly sound ways to divide the text into "chunks", and one occurred to me that had not already: Line beginnings are ternary, being read as "+" or "-" or ">". Each line's result either adds or subtracts to that of the preceding line. When a ">" is reached (but only without an "-" or an "+"!), then the counter is reset. The ">" that follows "+" or "-" is to be disregarded for purposes of dividing the text.
EDIT: Gonna post an example here in a sec so you know what I'm on about. Brain requires more coffee first (besides, it's FFS's the doctor's orders)
I'm going to go and play around with the concept. Like I said, I don't think I have the brain power to solve this but it's definitely worth a shot.
EDIT: Gosh, double post! That means the rest of you aren't thinking / posting fast enough!