Need advice for testing of hypotheses related to the self-citation method

[ReneZ]

I like this approach. One of the usual complications arises, namely that the number of permutations to test quickly increases to impractical levels.

The restrictions you put seem reasonable:
1) No need to achieve 100%
2) Limit the 'history' from which to pick new words
3) Limit the number (and type) of permutations to allow

The initialisation problem (per page) remains, but this can be 'delayed' by just ignoring the first part of each page for the time being. 
The special characters (with preferences for certain locations) also remain as an issue. I have no quick solutions there, and since these happily appear in labels, I also don't yet see how labels can be addressed. (Also for other reasons).

In general, I would suggest to test on the basis of a few choices for (2) and a few choices for (3), and then see for each combination what is the percentage achieved for (1).

One approach might also be to start by replacing each character by its 'STA family'. This reduces an enormous amount of signal (but of course we can't be 100% sure if it is signal or noise).
If this leads to much higher percentages of coverage, at least we will have learned something...