The matter

Before anyone judges the sentence, it has to exist as something an institution can act on. In Runcible Oversing it is not a line in a document. It is an object, and other things attach to it.

The trial data attaches to it. So does the definition of “recovery time” the claim depends on, the market it is intended for, the regulatory regime that governs it, the campaign that wants to use it, and the deadline that campaign is working to.

That is the difference between a claim and a matter. A claim is text. A matter carries its own context, and it keeps carrying it.

The judgment

Runcible AI works on the matter through the platform, not beside it.

It reads what is attached. It identifies that “recovery time” is defined one way in the trial protocol and another way in the draft copy. It notes that the 30% figure holds for one patient subgroup and not for the population the campaign describes. It records what supports the claim, what does not, and what is unresolved.

None of that becomes a separate report. It attaches to the matter, where the next person to open it will find it.

The decision

Someone has to own this. Runcible Oversing already knows who.

The matter sits in a workflow state, and that state determines who may act on it. Regulatory review holds authority over the subgroup question. Marketing holds authority over the campaign. Neither can quietly override the other, because the permission that would let them do it does not exist.

Human validation is the default here. Runcible AI has produced a judgment; it has not made the decision. Where an institution has explicitly delegated authority — with scope, conditions, and escalation defined in advance — Runcible AI can decide and act within it. Publishing a therapeutic claim is not that kind of decision, and the platform is configured accordingly.

The decision is recorded with its grounds: not just approved with amendment, but on what evidence, by whom, and against which objection.

The action

The amended claim goes out. Runcible Oversing supplies the surface it goes out through, so the action stays attached to the matter that authorised it.

What was published, in which market, on which date, under which approval, in which version. Not as an audit log filed somewhere else — as part of the same object the evidence and the decision are on.

The consequence

Six weeks later a regulator asks a question about the subgroup.

This is the point where most institutions begin an internal excavation: who approved this, what did they see, what were they told, which version went out. The excavation is expensive, it is slow, and it usually produces a partial answer.

Here the question arrives at the matter. The regulator’s query attaches to the same object that holds the trial data, the qualification, the objection that was raised in review, the decision that answered it, and the exact copy that shipped. The answer is not reconstructed. It is read.

The memory

The claim is corrected. That correction is the part that usually evaporates.

The definition mismatch that started all of this is now a known failure mode attached to a matter anyone can find. The next therapeutic claim that reaches review meets it. The reviewer who raised the subgroup objection has a record that they raised it and were right.

An institution that can do this has changed what it is capable of. It is no longer only as good as who happens to still work there.