I was following a recent discussion on this topic on alt.python. I was a bit disappointed with the apparent general support to the idea of creating these solvers by mimicking human reasoning.
Note that the following isn't a criticism of your decision to do it that way. Everyone should choose how they wish to use their time. The personal satisfaction of reaching a solution that performs reasonably well is a worthy feeling if one is doing it to have fun. It is instead a criticism of the idea that these type of algorithms gather support and are worth a debate on their merits.
Sodoku solvers are a dime a dozen, but the problem is that these mostly all fall down in the category of brute-force solutions of "do it as I play it". These are relatively fun and make up a good addition to the list of "Things I Must Code before I Die". But certainly they aren't worth much of a debate. There's very little to argue about brute-force sudoku solutions:
1. Performance can't be qualitatively measured because solutions aren't based on the puzzle difficulty.
2. They are rigid algorithms incapable of easily adapting to the large number of Sudoku variants because they are logic-derived instead of deductive-based algorithms.
3. Speed and scalability is entirely delegated to the system capabilities.
That so many people in alt.python chose to happily discuss the merits of a number of sudoku brute-force solvers left me with a bad taste in my mouth. Things like "my algorithm gives me a solution in x microseconds" feels like someone saying he's a pilot because he flies a drone.
However -- and this is the disheartening bit I'm trying to express -- there's very little mathematics involved in many of the less-logic-and-more-deductive algorithms that can also solve sudoku puzzles. These algorithms offer better performance, are more generic and can be more easily modified to support other variants and often the solutions are based on the puzzle difficulty, making for a better scaleable solution and indirectly providing important data on how difficult the puzzle could be for humans.
Most of the proven deductive algorithms only require a modest amount of computer science knowledge in fields that we would do well to have a modest knowledge about. And yet, it seems, no one in alt.python, including seasoned veterans, remembered to advise towards better solutions that put to the test our CS knowledge since we are all, you know, programmers...