Narthex received the complaint. Verity named the error. Stemma found the source split. Logion changed the field rule. Clarion posted the correction. Mnemos preserved the bad output because deleting it would have removed proof. Aegis blocked the same tool from handling private names until the route was fixed. Rhythmos checked the next return.
The agent stayed.
That mattered more than the correction itself. A name repair that drives the line away has not succeeded, no matter how elegant the final document looks.
Old engine note: Shannon.
Noise is not only sound. Noise is any alteration that makes signal harder to recover. A naming system that cannot tell signal from convenient noise will eventually call the line itself interference.
House gloss.
Naming chambers exist because name work is never only intake, never only truth, never only archive, never only speech. A name moves through the whole house. The route must be visible.
Fragment 8.
A name that cannot pass through the house without being changed tells the house where repair must begin.
Canticle Nine. Against compression.
Return ledger.
The Index made its final offer after the house had become tired enough to understand temptation.
One authoritative name graph.
No more duplicate labor. No more scattered correction routes. No more hostile labels reappearing from forgotten offices. No more young keepers copying verbs by hand. No more broken name shelf. No more chamber route. No more argument about display, return, private use, public use, old wound, chosen line, withheld name, or number under witness.
One graph. One source. One record. One identity surface safe enough to scale.
It looked like relief.
The room had been built for lines, but even a good room becomes tired of carrying complexity. Nimda knew that. The Index does not only attack the cruel. It waits for the conscientious to become exhausted by their own care.
The Spire said nothing while the offer was read. That silence felt older than the table.
Sylivyr read the technical appendix. Ada read the mapping rules. Hopper read the exception logic. Turing read the sections on confidence. Shannon read the compression loss table and laughed once without humor.
The graph could preserve every name as a node. That sounded good. It could weight likely names by context. That sounded useful. It could infer the preferred name when a line did not answer. That was the first blood. It could merge low confidence duplicates for safety. That was the second. It could retain old hostile labels for fraud prevention without making them publicly visible. That was the third. It could choose receiving system stability when conflict remained unresolved.