Having got rid of the annoying flaw in disambiguation (I hope), I’ve posted the latest build of Alabaster. The plan at the moment is to do a little more beta-testing to make sure that the conversation is sufficiently rounded out; then to remove the conversation-building machinery and do the last speed tests and refinements once that is gone. If you want to play along, transcripts are welcome.
Currently the biggest between-turn lags — sometimes very long indeed — continue to be in response to disambiguation questions or when the parser can’t match a quip at all. I am not sure why it’s doing this, but I suspect that the quip-creation machinery may be slightly interfering with the efficiency.
Then we’ll do some profiling.
The startup delay should be gone completely, though, and between-move delays reduced in most other cases.
Emily Short's Interactive Storytelling 
I think I must need to update Git or Glulxe or something, because I can’t even get Alabaster to load on my machine. It just goes straight to “Press any key to exit” in the former and complains about the version with the latter.
To the best of my understanding, your best bet for Windows interpretation right now would be Windows Git, recently updated as announced here. Zoom should work on the Mac. I don’t know what the current state of Linux Glulx interpreters is.
That did it; thanks.
Just out of curiosity, are you planning to release Conversation Builder as a public extension? Right now I’m working on a conversation-based project myself, so I’m fascinated paging through the source and seeing the engine you put together for Alabaster, but I’m having some trouble piecing together how it works since a lot of the key terms seem to be defined outside the source.
That is the plan in the long run, yes. Alabaster is partly meant to help refine it, but I don’t intend to release the final project until (a) the core code is adequately speedy under a range of different conversation conditions; (b) I have written, tested, and released multiple full-sized works in it, with different artistic aims; (c) other people have beta-tested the extension package; (d) a full set of documentation has been written (I have some now, but it needs a lot of revision); (e) (optionally) I have created some of the visualization tools I would like to have accompany this package.
This is, regrettably, like to take a while longer, but I’m committed to producing something that is genuinely robust and usable. My impression is that the chief obstacle to more conversation-based IF is not the technical problem of producing flexible chat systems — there have been a wide range of extensions available to do this for some time, and TADS 3 has a pretty strong set of options built into the core system — but the work load for authors. So I’m focusing more than I otherwise might on the tools that will go into creation, testing and debugging with this system, in an attempt to make that process as painless for authors as possible.