Prompt file imported from ukaszp-commits/gpw-eda-pro-v1-archive (
.codex/prompts/collector-noise-filter.md). Copyright stays with the author.
Codex task: improve collector noise filtering
Goal: make the collector keep only useful company/event information and discard forum/news noise.
Rules:
- Keep source URLs, source type, author/channel if available, date, ticker, event, claim, confidence.
- Drop spam, insults, repeated posts, unsupported hype, content unrelated to ranking companies.
- Add duplicate detection or similarity grouping if feasible without large dependencies.
- Add explainability: why a source was kept or rejected.
- Do not break current saved collector data.
- Run
npm run codex:checkandnpm run build.
