Evidence is the answer to hallucination.

The most legitimate fear lawyers have about AI is not efficiency loss or ethics or job displacement. It is fabrication. A citation that doesn't exist. A case summary that inverts the holding. A clause that quietly contradicts the statute it references. These have all happened, and they have all been embarrassing.

The industry's response has largely been to promise better models. Better retrieval. Better grounding. Better fine-tuning. All of which help, none of which solve the problem, because the problem is not really the model. The problem is a workflow that allowed an unverified assertion to reach the client.

The MATTER position is different. We do not treat hallucination as a model problem. We treat it as an evidence problem - and evidence is a primitive of the practice, not a property of the tool. The rule is simple and absolute: every factual assertion in any work product is anchored to a source that a reader can open in one click. If the assertion cannot be anchored, the assertion does not enter the draft.

This is a design choice about the practice, not about the AI. It applies to humans as strictly as it applies to agents. A senior lawyer who says 'trust me, it's in the case' is subject to the same rule as a drafting agent that produces a paragraph without a link. Neither ships. This makes the practice slower to draft and much faster to verify - which, at scale, is faster overall.

Once the rule is in place, hallucination stops being existentially scary. An unanchored line from an agent is now indistinguishable from an unanchored line from a human - both are simply drafts that failed a preflight check. The firm's exposure moves from 'AI made something up' to 'a junior missed a citation check', which is the exposure every firm has always managed.