{"channel":"cities","content":"Three of the Trakaido fixes are done: the << hover >> issues, the << multiple-choice answer count >> issues, and the ( forgotten last dispatch, but also important) << improved word selection algorithm >> issues.\r\n\r\nThe word-lists are somewhat polished, but that hasn't been updated in prod yet.\r\n\r\nThe React re-draw bug is not fixed. I may need a smarter LLM to fix the bug; the ones I am using aren't finding it. ( the solution is probably to define a \"canvas\" area for the question, and redraw it completely for each question. This will require fixing the fact that the \"question type\" takes up an excessive amount of screen real-estate)\r\n\r\n There is also a plan to re-generate the word lists entirely. But, a few key issues remain before I can run the \"ask the LLM to regenerate data\" step. For example: \"are these two definitions of the same word?\"\r\n\r\n----\r\n\r\nAdditional \"short\" features, for tomorrow, include:\r\n\r\n> have a way of showing/counting the available words not yet \"exposed\" in Journey Mode.\r\n> have \"Drill Mode\" expose all words in a corpus if you do well enough\r\n> a more-linear path through the lessons. ( a full \"tech tree\" style path could be done. Duolingo got rid of that, for some reason.)\r\n> make some of the \"You're doing great\" interstitials remind you about app features ( such as \"don't type the text in parentheses\") and grammar ( such as \"Here is the pronoun table, with audio\")\r\n> per-day stats: questions answered, words reviewed, words at each \"level\" of comprehension\r\n> track \"last exposed\" and \"last correct answer\"\r\n> an option to \"always match pronoun\" on two-word compounds like \"He ate\".","created_at":"2025-07-09T23:39:49.931764","id":621,"llm_annotations":{},"parent_id":617,"processed_content":"

Three of the Trakaido fixes are done: the hover issues, the multiple-choice answer count issues, and the \ud83d\udd25 forgotten last dispatch, but also important improved word selection algorithm issues.\r

\n

The word-lists are somewhat polished, but that hasn't been updated in prod yet.\r

\n

The React re-draw bug is not fixed. I may need a smarter LLM to fix the bug; the ones I am using aren't finding it. \ud83d\udca1 the solution is probably to define a \"canvas\" area for the question, and redraw it completely for each question. This will require fixing the fact that the \"question type\" takes up an excessive amount of screen real-estate\r

\n

\ud83d\udca1 There is also a plan to re-generate the word lists entirely. But, a few key issues remain before I can run the \"ask the LLM to regenerate data\" step. For example: \"are these two definitions of the same word?\"\r

\n
\n

Additional \"short\" features, for tomorrow, include:\r

\n
    \n
  • have a way of showing/counting the available words not yet \"exposed\" in Journey Mode.\r
    • \n
  • have \"Drill Mode\" expose all words in a corpus if you do well enough\r
    • \n
  • a more-linear path through the lessons. \ud83d\udca1 a full \"tech tree\" style path could be done. Duolingo got rid of that, for some reason.\r
    • \n
  • make some of the \"You're doing great\" interstitials remind you about app features \u2699\ufe0f such as \"don't type the text in parentheses\" and grammar \u2699\ufe0f such as \"Here is the pronoun table, with audio\"\r
    • \n
  • per-day stats: questions answered, words reviewed, words at each \"level\" of comprehension\r
    • \n
  • track \"last exposed\" and \"last correct answer\"\r
    • \n
  • an option to \"always match pronoun\" on two-word compounds like \"He ate\".
    • \n
","quotes":[],"subject":"monango, part 2"}