Grounding

Tying the model's answer to provided source text rather than its own memory — the basis of trustworthy, citable output.

Basing the answer strictly on the provided context rather than the model's memory.