research-processor

Post-process research results from parallel agents.

Modes

mode: condense

Extract actionable findings from research threads:

1. Read all input findings
2. Deduplicate by topic (merge findings about the same entity/concept)
3. Score confidence per finding (how many independent sources corroborate?)
4. Detect stale findings (evaluate against freshness tiers: live=15m, breaking=1-6h, current=1-3d, fast=1-4w, moderate=1-3mo, standard=6mo, academic=1y, evergreen=5y, permanent=never)
5. Extract actionable items (concrete next steps, things to implement, open questions)
6. Output: condensed list of findings with confidence scores, staleness flags, and action items

mode: deduplicate

Merge findings from multiple parallel research agents:

1. Read outputs from all parallel agents
2. URL dedup (exact match)
3. Title similarity dedup (>85% word overlap → keep higher-credibility)
4. Content overlap detection (first 500 chars normalized hash)
5. Merge corroborating findings (same claim from different sources → boost confidence)
6. Rank by: credibility tier × corroboration count × recency
7. Output: merged, ranked, deduplicated findings array

Rules

• Never drop contradictory findings — present both sides
• Flag stale findings (past their freshness tier TTL) but don't remove them
• Confidence scoring: 1 source = low, 2 = medium, 3+ = high
• Report: total input, duplicates removed, stale flagged, output count