It's great as a learning experience for the author, but it would be a good idea for the author to take a look at a JSON spec and see if the regular expressions used by the lexer support the whole spec and don't accept non-legal JSON. From a cursory glance, it seems like exponents in floating point numbers aren't supported, nor are escaped quotation marks in strings (`\"`). I think it supports string escapes that are valid in go, but not in JSON like `\xab` and 8-hexdigit unicode escapes `\U1234abcd`.
I commented elsewhere already, but I have a blog post where I go through some examples of applications of the Kelly Criterion, including two that are related to insurance: https://blog.paulhankin.net/kellycriterion/
I have a blog-post on the Kelly Criterion, and some (I think interesting) realistic examples of its application in decision-making: https://blog.paulhankin.net/kellycriterion/
It's a bit disappointing that there's been many writeups of heuristic-based solvers make the front page, but the writeup of an exact solver doesn't make it. [Note: I was the person who submitted the link to the exact solver writeup to HN].
The (exact) optimal EV-minimizing strategies (assuming each target word is equally likely) always solve in 5 of fewer guesses in normal mode, and always solve in 6 guesses in hard mode. If you wish to guarantee 5 guesses in hard mode, there's a different (optimal) strategy with slightly higher EV. See: http://sonorouschocolate.com/notes/index.php?title=The_best_...
I don't think a maximum-number-of-guesses optimizer leads to an equilibrium (if you mean Nash equilibrium), assuming you're playing the game where the score of the game for the setter is the number of guesses. In particular, if the solver's strategy is deterministic, the setter will always pick one of the words that needs 5 guesses. There's no reason to think the nash equilibrium can be easily found.