the new "lexer°hard word"-based°hard word parser°hard word has crossed the hump°hard word of possibility°hard word.
it took roughly 2 days of effort (plus°hard word 3 days of complaining, earlier). this is, arguably°hard word, slower than it would have been for me to write this without using°hard word the machine.
there are a few minor°hard word things to fix before it replaces the new system. those are in the next list.
----
today's°hard word things:
- New lexer°hard word bugfixes°hard word - lists format°hard word incorrectly°hard word, PGN°hard word templates°hard word don't°hard word work, quotes°hard word aren't°hard word stored°hard word
- Fix the test framework°hard word - occasionally°hard word it deletes°hard word the (local°hard word) prod°hard word database
- Better export°hard word/import°hard word tools - should be able to load the prod°hard word database locally°hard word by transferring°hard word a JSON°hard word file°hard word
- Don't try to add i-Ching°hard word based°hard word features°hard word
- Try to get the LLM°hard word Magic Query°hard word working - based°hard word on a short prompt°hard word ("Tell me about George°hard word Washington°hard word", "What does baking powder do"), figure out the best extended°hard word prompt°hard word, and then°hard word get the response°hard word.