A page may not promise what its own module disproves

This release is about a smaller and stranger thing than a feature: the difference between a promise a page makes and the code that would have to be true for it to hold. Three sentences VEILOS published were false, and none of them needed new evidence to disprove — only somebody reading the module they were written in. The home page told you the deeper statistics report names every denominator; eight of its fourteen rows cannot name one, and the function that derives the honest version of that sentence had been sitting four hundred lines above it, publishing "a denominator on 4 of 14" on the statistics page itself, for eleven sessions. The Trust Center — the page whose only job is to not overstate — showed checks nobody had measured with the tooltip reserved for checks proven at deploy time, so an absence of evidence read to a visitor as proof. And the Record's heading promised every act was verifiable directly above its own next line explaining that the Record is bounded and only retained acts are addressable. All three now derive from the populations they describe: add a row without a denominator, leave a check unmeasured, let a ring evict, and the sentence changes rather than becoming a lie. Two loops shipped last release also learned to speak when they have nothing to say. The Curiosity Curriculum and the Resonance escrow ledger were both genuinely empty, which was honest — but an empty list with no explanation means "this never happened", "this fell outside the window" and "this is broken" all at once, and a reader cannot tell which. Both now state their own reason, on the page and in the machine-readable projection alike, derived from the mechanism that decides it rather than written as a hopeful empty state. Underneath, an instrument was built that this repository had described and never constructed: an enumerator for every place a public sentence quantifies over a population the code deliberately shortens. It found sixty-three such places carrying a hundred and forty-eight claims, fourteen were read against their producers, and five were false. The remaining forty-nine are published as a count and explicitly not as an accusation. One planned addition was deliberately not shipped: a census of VEILOS's own prose would have had to sit on a page that defines a census as something re-proven live at request time, which source-file evidence at the edge cannot be, and weakening that definition to fit a new member would have quietly devalued the five that already meet it. No new dependency, paid service, provider resource, destructive operation or cost increase was introduced.