html tidy's parsing strategy

I'm curious what strategy tidy uses to parse and correct input with
errors.  I'm interested both because I want to use tidy with some
modifications, and because I have some input that fits this general
description and I'm not familiar with how to approach the problem.

I have the source, but there's a lot of it, so I'm hoping for some help.

Thanks.
Ross Boylan

P.S. I notice the tidylib interface is "very rough" according to the
comments.  Is it serviceable?