A law library tells you something before you open a book. The shelves disclose a jurisdiction, a period, a court. You know when you have crossed from a report of Zimbabwean decisions to one of South African decisions; the change has a physical shape. In a search box, the shelves disappear. Authorities from different places and decades arrive in the same neat column, their differences compressed into a line of metadata.
That compression is a remarkable convenience. It is also why the question behind legal AI is not simply, “Can it find a similar paragraph?” Similarity is only the beginning of relevance, and relevance is not authority.
Consider a lawyer researching a dismissal. A judgment from another country may describe strikingly similar facts. It might even express the argument more clearly than a local decision. Yet the lawyer still has to ask what law the court applied, what question it decided, and whether the decision belongs in advice to this client. An answer that borrows its best sentence from the wrong legal setting can be persuasive at precisely the moment it becomes unsafe.
The collection a system searches therefore shapes its answers before the model writes a word. Our Law Reports archive gathers the Zimbabwe Law Reports, the South African Law Reports, and Zimbabwe Labour Court decisions—the source collections guiding Zalari’s developing research database. Their value is not that they can be poured into one undifferentiated pool. It is that a practitioner might search across a wider body of material while keeping in sight which court spoke, when, and in what legal conversation.
This is an old principle in a new interface. Lawyers have never treated a library catalogue as a judgment. The catalogue helps them reach a case; it does not tell them what the case means. AI adds the ability to search by the substance of a problem rather than by the words one hoped a judge had used. That can uncover an unexpected line of authority. It cannot decide whether the line is binding, distinguishable, overtaken, or merely interesting.
Nor can an empty result settle the matter. A case may be absent from the collection, poorly described, expressed in unexpected language, or hidden by the way the question was put. A statute may have changed after the reported decision. “I found nothing” is a statement about a search, not a proposition about the law. The difference is easy to forget when the search speaks in complete sentences.
So the first encounter with Zalari need not be a leap into an unfamiliar matter. It is an early preview, and its drafting is not yet reliable enough for production work; fuller capabilities are still to come, and we will share progress as they do. Choose an issue you know well, perhaps one for which you remember a decision that took an afternoon to locate. Ask in the jurisdiction and forum that matter. Look at what appears, then at what does not. Open the judgments. Notice whether the case cited for a proposition actually decides it. If an expected authority is missing, tell us: a useful research system must be corrigible by the people who know its field.
That is a more exacting measure than the number of documents in a database or the elegance of an answer. A good tool expands the lawyer’s field of view while leaving the edges of that field visible. The library is part of the answer; knowing where its shelves end is part of the work.
