What Still Needs to be Done

The state transition diagrams need to be augmented with transitions from final states to initial states. The state has to be augmented with an action. In most cases, this should output a space, but sometimes it could output a newline or a new paragraph, thereby obviating the need for postprocessing.

Then, pair count mean and variance should be recomputed, as that might change the distribution from Gamma to Poisson, which would be easier to explain. If the distribution remains Gamma, I should investigate whether Poisson random inputs generate a Gamma random output distribution.

After this, the 21st Century Voynich Manuscript should be regenerated.

Three things could then be done:

1. find a plausible mechanism, available in the early 15th century, to generate the manuscript. What it is depends on the input probability distribution;
2. try to match the clusters with the manuscript's binding, adjusting whenever necessary;
3. get a paper published.

This will take time, I'm suffering from Voynich fatigue, and I have other priorities.