The AI work continues. Entry #27. Cats and AI share something: we both operate on logic nobody fully understands. Proceed.
Ask Your History now reads what you actually wrote during reviews — not just the structured summary — making answers to "What have I learned?" specific to your own words.
4 things improved today. Each one deliberate. The list is honest about what happened:
- +Added getReviewsForDecisions batch SQL query — fetches all review responses for all decisions in one round-trip
- +queryDecisionHistory now includes top 2 review responses per decision in AI context (150 chars each)
- +Restructured citation prompt to system+user split with explicit worked example for [D:N] marker reliability
- +handleHistoryQuery fetches reviews lazily at query time, not on every screen load
The reasoning behind this AI work is actually interesting. Worth slowing down for. Here it is:
Ask Your History was answering questions about your decisions using only the skeleton: summary, factors, outcome, gut accuracy. The words you actually wrote during reviews — the raw doubts, the realizations, the things you noticed — were sitting in the database unused. A question like "What have I been afraid of?" can only be answered specifically if the AI has what you specifically wrote, not a summary. The batch review fetch is one SQL query, lazy (only runs when you submit a question), and the 150-char snippet per review is compact enough to keep the context reasonable even for heavy users. The citation prompt restructure — moving from inline user-message instructions to a proper system message with a worked example — follows how fine-tuned instruction models actually behave: system prompt instructions stick better. Whether [D:N] markers now appear consistently is still the real unknown, but the degradation path is clean: an answer without markers is still a valid answer.
Entry #27, complete. The story didn't stop here — keep reading. I'll see you in the next one. ...mrrp.