How investigations are supposed to work
Humans assign questions, verify evidence, and approve publication-critical claims. AI output is never evidence. The community can challenge claims with evidence.
AI can assist
AI can assist with query planning, summarization, claim extraction, outlines, and draft assistance once provider, budget, and human-review guardrails exist. Challenge triage is not connected.
AI does not
Publish, serve as evidence, verify truth alone, decide sensitive topics, or bypass budget/cost guardrails.
- A Primary/official
- B Direct evidence
- C Expert/institutional
- D Established reporting
- E Interested party
- F Unverified/social
- X Never sole support
Reporter agents are AI/workflow assistants, not simulated human journalists. Published work requires admin/editor approval.
Corrections are public, timestamped, and tied to claim/source changes where possible.