Fan-out swarm discovery of research sources and ingest into Research Foundry

Parallel discovery swarms that populate a research run's source cards faster than sequential manual research

• · run_id (Research Foundry run identifier)
• · profile (default 'personal')
• · research_brief.md and optional swarm_plan.yaml

• · ingested source card manifest
• · discovery_strategy {research_question, key_topics, source_types, domain_legs}
• · report via rf status push + rf writeback

• · rf (Research Foundry) CLI (guard check, ingest, guard check, extract/claim-map/synthesize/verify/bundle post-run)
• · agent orchestration (discovery_lead, domain_researcher, source_scout agents)

• · Research Foundry workspace initialized (rf init + rf doctor green)
• · rf guard check passed
• · research_brief.md written

• · rf ingest fails on malformed candidate (structured validation error)
• · Duplicate sources bypass dedup (false low yield)
• · IntentTree unreachable during phase (status push fails, not fatal)
• · rf verify returns exit 7 (human council review required)

• · 4-phase Plan → Discover → Dedup → Ingest
• · rf guard check gates entry and exit
• · IntentTree push during Ingest + writeback post-run
• · POST-RUN deterministic tail documented (extract → claim-map → synthesize → verify → bundle → writeback)