Inference without Authority: the three authorities that govern an output

Every program produces output, and it is taken for granted that the code deciding what to emit also decides where it goes — to a database, a log, a topic, a screen. This paper argues that these are distinct authorities, and that a program speaking both with one voice has fused decisions belonging to different competences. Separating them reveals a third: what an output is — its projection — is no more the domain's to shape than where it lands is the producer's to name. An output is governed by three authorities — the domain, which knows what exists; the actor, which knows what becomes observable; and the assembler, which knows where observation occurs.

Their collapse has a signature. When one authority is made to pronounce another's decision, it must reach a conclusion it has no standing to reach — an assertion beyond its warrant. This unwarranted assertion is the sibling, one level in, of the infrastructural symptom the series treats elsewhere: it compensates for an authority that is absent and dissolves when that authority is restored. The symptom appears at both ends of an output — a producer denied a voice for the destination commits to a where it cannot know; an observer given only snapshots infers a how it was never told — one defect, absent information, met once by stipulation and once by inference. Restoring the authorities dissolves both: the destination gains its own voice, and the record reaches the observer as an account it is told rather than one it must reconstruct. The observer comes to know by testimony, not by inference — within a trust boundary. Where the producer sits across one, as it does for anything leaving its own trust domain, testimony must be verified rather than simply believed; that boundary the paper marks as the bound of its scope, not a caveat to its result.

A journaled actor system serves as a worked instantiation in which the three authorities are kept apart — a demonstration that the separation is buildable, not a measurement of its cost or benefit. The contribution is analytic — a theory for analyzing in the sense of Gregor's (2006, Type I): a criterion by which any architecture may be read, sound not by how it divides code, but by whether the authorities within it match the decisions each part has the standing to make. The labs are an existence proof of realizability, not a design-science evaluation of cost or benefit.