root
83e75e7c24
Both are Section B (needs-context): B11 touches the AnalysisResult contract and the question of where an LLM is needed at all; B12 touches scoring coefficients and the DB schema, which project rules keep off limits to automated edits. B11 records the key finding that the LLM does not compute priorities -- the math already lives in geo_service/scoring_service/psychotype_service and analyze_with_fallback already returns a full AnalysisResult with no API. The only unique in-hot-path LLM role is parsing free-text reports, so a strict intake form and a local provider are independent levers. B12 notes ISRID supplies empirical priors for the currently hand-tuned zone weights, with the legal status of the database flagged as a blocking question to resolve before any data enters the repo. Summary table updated to B1-B12. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>