list-negative-uses-two-stage-classification
IN premise
`list_negative` uses keyword pre-filtering against a hardcoded `NEGATIVE_TERMS` list (~50 words), then LLM classification via `ask._invoke_claude` to eliminate false positives.
Summary
The negative review detection pipeline works in two stages: it first does a cheap keyword scan against about 50 hardcoded negative terms to find candidate reviews, then sends those candidates to an LLM call to filter out false positives. This means detection accuracy depends on both the completeness of the keyword list (which gates what the LLM ever sees) and the quality of the LLM classification step.
Dependents
These beliefs depend on this one:
- list-negative-is-defensively-bounded — The negative belief listing pipeline applies defense-in-depth: keyword pre-filtering narrows candidates before LLM classification, hallucinated node IDs are discarded against the actual network, and malformed LLM output falls back gracefully to zero count rather than raising.
Details
| Source | entries/2026/04/29/reasons_lib-api.md |