Hallucinated certainty is more dangerous than hallucinated citations.

A fabricated citation is a self-announcing error. Opposing counsel cannot find the case, a citator does not recognize it, and the sanctions docket has made the consequence familiar. The failure has a natural adversary and a natural detector.

Hallucinated certainty has neither. It occurs when a system presents a genuinely contested question as settled: a circuit split rendered as a single rule, a balancing test reported as a bright line, an open analogical question answered as though precedent had closed it. Every case cited may be real and accurately quoted. The error lies in the confidence, not the citation, and nothing in the output reveals it. The client is advised as if the law had one answer, the brief is written as if the other side had no authority, and the court is not told about the split. The first person to notice is often opposing counsel, in an opposition citing the courts that went the other way.

That is why the Zeroth Law of System Design, which takes precedence over the other three, concerns confidence rather than accuracy: no output may be presented with greater confidence than the system can defend, and uncertainty must be surfaced, not smoothed over. A system can pass every citation check and still violate it.

The remedy is not a hedging disclaimer appended to every answer. Blanket caveats are as uninformative as blanket confidence. The remedy is a system that knows where the law is actually divided, can say so specifically, and can distinguish a gap in its own research from a gap in the law. Fabrication is a defect of retrieval. Hallucinated certainty is a defect of epistemics, and it is the harder one to engineer away.