When Flow imports a wikitext talk page it is possible to attempt parsing that into a set of topics (based on sections), and splitting even further into comments (by signatures).

We can imagine many edge cases where this can run into issues, such as when user A makes a list, and user B replys in-line to a particular element of the list. We arn't sure how prevalent these are though.

It seems plausible that a significant number of wikitext talk pages might be able to be cleanly converted, especially less trafficked talk pages that only have a couple of sections. There is already a place in the Flow conversion process to report this information, we just have not written anything to attempt this.

I'm thinking we should write something that works at a very basic level as a first attempt. It would read only the current version of the talk page, split it into sections, and the split by signatures into individual attributable comments. Some sort of score should be generated representing our confidence in this being a complete conversion. If this looks to produce reasonable results with proper conversions on some pages we can make it a little more robust by starting with the first edit to a talk page and stepping through the revisions, but this gets more complicated due to archiving and such.

The main goal is not to convert all wikitext talk pages to Flow, many are just too complex and need to be archived, but to have a process that can attempt the conversion, and then we convert only pages with a confidence interval > some pre determined level.