A citator answers one question about a judgment: has anything happened to it since. Not whether it exists, and not whether it is well reasoned. Whether later courts have followed it, applied it, distinguished it, doubted it or overruled it. Solicitors who came up with KeyCite or Shepard's know the reflex, and the reflex is the point: you do not cite a case in a document that leaves the firm without checking what was done to it afterwards.
The mental model behind KeyCite and Shepard's
A citator is built by reading the citing judgments, not the cited one. Somebody, or something, works through every later decision that mentions the case and records what that later court did with it. Aggregate those records and you get a signal that saves the reader the work of doing it themselves.
The commercial versions add editorial judgement on top: a person decides that a passing reference in a footnote is not really an application, and that a single sentence in a Court of Appeal judgment quietly undercuts the whole ratio. That editorial layer is a large part of what a Westlaw or Lexis subscription buys, and it is genuinely hard to reproduce.
Treatment is not the same as outcome
The most common misreading is treating a signal as a verdict on whether you win. Distinguished does not mean bad. It means a later court found the facts different enough to decline to follow it, which may make it stronger authority on facts like yours, or weaker, depending entirely on what the distinction was. Applied does not mean safe, either. A case can be applied twenty times on a point that has nothing to do with your argument.
Doubted is the signal that deserves the most attention and gets the least, because it is the one that never resolves itself. A case nobody has overruled but several judges have described as difficult is exactly the authority you do not want to build an opinion on without saying so to the client. Overruled is the easy case: you stop, and you find something else.
How our Citator sub-agent works
The Research Agent is an orchestrator with eleven sub-agents, and Citator is one of them, sitting alongside Primary Law Retrieval, Regulatory Retrieval, the Cross-Jurisdictional Comparator and the Regulatory Change Tracker. It assigns a treatment signal from a fixed set: good law, followed, applied, distinguished, doubted, overruled, or unverified. It also lists the citing cases it found, which is the part a supervising solicitor uses most.
It works from the same fixed list of 74 approved public sources as everything else: BAILII, the National Archives Find Case Law service, legislation.gov.uk and the rest. That is both its strength and its ceiling. The signal comes with the citing judgments attached, so you can open them and disagree with it, which is not something a proprietary signal usually lets you do.
Why unverified is a real answer
Most systems that produce a signal are under pressure to produce one for every case. That pressure is where the damage happens, because a confident green marker on a case the system could not properly check is worse than no marker at all. So there is a seventh value in the set, and it is used: unverified means the citing history could not be established from the approved sources.
The same principle runs through the whole product. Nothing degrades upward. A check that cannot complete becomes needs a check, never verified, and a treatment the sub-agent cannot establish becomes unverified rather than good law by default. Silence is reported as silence, because a reader who knows the system is uncertain can act on that. A reader who has been told everything is fine cannot.
Where it sits in the run
The Research Agent is governed by 14 protocols, with a third layer of output gates covering citation verification, confidence scoring and style formatting. The citator signal feeds the confidence layer: an answer resting on a doubted or overruled authority cannot carry the confidence of one resting on a case that has been followed three times since.
Practically, that means the treatment signal is not a decoration on the authorities table. It changes how the answer is written. The confidence-floor gate only ever marks work down, so a weak treatment can lower the confidence of a section but never raise it, which is the correct direction for a control to move in.
How to read a treatment signal properly
Three habits. First, read the citing cases, not just the signal, on any authority the argument actually rests on. The list is there for exactly this, and two minutes in the citing judgment tells you more than any label. Second, treat unverified as work rather than as an absence: it means somebody has to check, and that somebody has a name.
Third, remember that a citator tells you what courts have done, not what a regulator or Parliament has done. A perfectly good case can be made irrelevant by an amending statute or a new regulator handbook provision without any judge ever mentioning it, and no citing-case analysis will show you that. Statutory currency is a separate check.
What this does not do
It is not a licensed editorial citator, and we will not describe it as an equivalent to one. It has no editorial team reading every judgment, it has no access to unreported decisions that never reach a public source, and it cannot see anything behind a paywall, because Westlaw, Practical Law, LexisNexis, Bloomberg Law, Justis and vLex are refused by name. A firm whose litigation practice depends on a commercial citator should keep it.
Coverage is England and Wales, with EU, Irish and ECHR primary law in the source list, so treatment analysis of US, Canadian or Australian authority is not something this does at all. And a signal is a starting point for a lawyer's judgement, never a substitute for it. The question of whether an authority still supports the proposition you are making is one no label has ever answered.
