Citation Quality
Agents that cite sources accurately, attribute claims to the chunk they came from, and report when a claim has no source — the foundation of trustworthy output.
Citations are not a UX feature; they are an accountability mechanism. An agent that cites the wrong source is worse than one that admits it does not know.
What it solves
Makes the agent's output auditable. The reader can see which claim came from which source and decide whether the source supports the claim.
How we build it
Retrieval returns chunks with stable IDs; the model is prompted to attribute every factual claim to a chunk; an evaluator (rule-based or model-based) verifies that cited chunks actually support the claim and that unsupported claims are flagged. Failures feed back into prompt or retrieval changes.
- Stable chunk IDs from retrieval
- Per-claim attribution in output
- Post-hoc attribution verification
- Unsupported-claim detection and flagging
What changes when it is in place
Hallucinations become observable. The rate of unsupported claims becomes a tracked metric the team can target.