At some point in development mode, an "index" directory was created in my rails root with stuff like "index/development/model_name/segments_8" and I'm not sure yet if that is sphinx related. I do know that when testing, I see nothing like "index/test/model_name/…".

Thinking Sphinx used to not support faceting, time deltas, or excerpting, but now does; those were the "enterprise" features that kept people on Ultrasphinx. Since the addition of those to Thinking Sphinx, Ultrasphinx is no longer being developed, so the upgrade instructions are especially useful.

The basic issue with string sorting in deltas is that you need to map the strings into an absolute space, whereas str2ordinal maps it into a relative space. So the correct algorithm needs to be as so:
* define a bigint Sphinx field for each string field you need to sort
* downcase and strip your strings of symbols
* map them into the lower ascii space (0 to 35), which is 5 bits per character
* bitpack that into a bigint, truncating the extra (this gives us up to twelve sort characters per string)
* store that bigint in the additional Sphinx field at index time

Now your client can request sorted results from the additional field, with a little overscan, and post-sort to achieve strict order within the group. Of course this will break down if all your strings start with the same 12 characters.

Ideally Sphinx would let you declare a sortable characters limit and store the truncated portion of the string internally.

Thanks for the article.

The post has been updated; thanks for taking the time!

Thanks you for bringing that point up as I had not touched upon that in the post!