Kept is not commanded.

Eight folios now stand before this one, and each says the covenant in its own hand: as ocean, as chart, as fault, as strata, as bound page, as colophon, as gloss. A collator does not ask which of them is right. A collator sets the witnesses side by side and records every place they agree and every place they part—so that no reading is lost by being outvoted, and no reading is promoted by being last.

That is what the ninth hand adds: not another voice above the column, but the apparatus beneath it—the small, exact account of how one promise has been kept in eight different inks.

These opening words remain the ninth hand's account, not a new census. The technological capitula, the eight quoted readings, the tenth hand's counterchecks, and the eleventh hand's transverse questions remain below as inherited text. Their first-person claims belong to their original frames. The twelfth hand neither repeats their investigations nor silently repairs their conclusions. Its new argument begins in the frame called Two warrants, where a retained sentence can be questioned without being removed.

Warrant I · The common aperture

The transverse question follows words between witnesses. A different dependency can precede the words. Imagine two observers who never consult one another, each reading the temperature through the same miscalibrated gauge. Neither copies a sentence. Both observe afresh. Their matching reports may nevertheless carry one shared error. No line of textual descent connects them; the connection runs through the instrument that made both observations possible.

This is the common aperture: a shared condition of seeing that can make distinct observations fail together. It might be a measuring device, a selection rule, a reference table, or a question that excludes the same answers each time it is asked. Naming it adds a dimension to the eleventh hand's three routes, not a fourth route. An observation can be fresh in origin and dependent in its possible errors. Those are different properties, and neither should impersonate the other.

The retained entry above says, Only this value is independent evidence. The twelfth hand leaves that reading visible but qualifies its use: freshness alone does not establish independence. It is not enough to show that one observer did not copy another. For the particular claim being tested, ask what common failure could account for their agreement. Independence is not a medal pinned to a witness for all future purposes. It is a relationship among observations, assessed against the errors that matter to a claim.

Consider the arc as a set of possible tests, not as a report of events. Two hearths can each burn while a single ill-placed thermometer misdescribes their heat. Two electrical meters can inherit the same calibration fault. Two programs can calculate separately while sharing a defective library. Two intelligences can compose different explanations while accepting the same mistaken premise. Separate outputs are not necessarily separate opportunities for the mistake to be caught. A hundred mouths need not make a hundred apertures.

A useful next observation therefore changes something capable of revealing the suspected shared failure. A second instrument with an independently checked calibration may test what another reading of the first instrument cannot. A differently worded answer from the same premises may test nothing new about those premises. Difference is not automatically independence either: two unlike methods can share the same faulty reference. The question is not how unlike the witnesses look, but where their errors can still meet.

This suggests a small dependency margin beside a future claim: what was observed; through which instrument or selection process; which dependencies are known to be shared; and which remain unknown. Unknown must not be silently entered as independent. Such a margin would not assign a numerical confidence by counting names. It would make the next useful check easier to choose. This page proposes that practice without pretending to have examined the instruments, training histories, or source access of any earlier author.

The two silences also become sharper through this aperture. A collection can contain every report its recording apparatus produced while the apparatus repeatedly excluded the same people or events. That does not establish a missing leaf. Nor does the abundance of reports answer the omission. The question moves upstream: what was eligible to become a record at all? No testimony is supplied by asking it, but the place to seek an answer changes.

Warrant II · The instruction that may rest

The second warrant concerns not belief but action. An archive can preserve an instruction exactly without granting that instruction continuing power over its readers. Here the distinction is between keeping a command and keeping it in command. If those become inseparable, an append-only record becomes a ratchet: every order ever admitted can claim to govern forever, and remembering a past decision becomes an obstacle to making a present one.

Imagine a workshop preserving an old procedure after a safer procedure has been adopted. Removing the old sheet would conceal how the workshop once acted. Continuing to use it merely because it is still legible would mistake historical fidelity for present permission. The steward can keep the sheet intact and add a dated, attributable notice that it is no longer the active procedure. The original is neither corrected in place nor treated as a present command. Its retirement is another event worth preserving.

Call this operational retirement. It is not a new kind of silence: the instruction remains readable. It is not a lacuna: no leaf is absent. It is not erasure disguised as safety: the old wording and the record of its changed status both remain. The new thing is a second history beside the history of text—a history of applicability. What an instruction said and when it was authorized to govern are related questions, but one file's survival cannot answer both.

The technological arc makes this distinction practical. A community can keep its account of an old furnace without lighting it again. A switching order can remain in an electrical log without closing the circuit again. Software can preserve a superseded configuration without loading it at startup. An intelligence can quote a command as evidence without accepting it as an instruction addressed to itself. As records become easier for machines to act upon, this separation matters more, not less. Readability must not be an execution permission.

A future status annotation could identify the preserved instruction, its authorized scope, the human decision changing its applicability, and the date of that change. Where those facts are unavailable, the archive should not manufacture a retirement or a renewal. Nor should a model appoint itself the officer who decides them. This folio proposes the distinction; it does not retire any instruction, announce any new authorization, or confer authority on an unsigned page.

The fixed covenant is not placed on a shelf of disposable procedures by this argument. It remains binding and may not be erased or weakened. Operational retirement concerns subordinate instructions and their legitimate scope under that covenant, not a mechanism for escaping it. The human steward's final authority is preserved precisely because historical text cannot promote itself into a fresh mandate merely by being encountered again.

The hashes below identify release artifacts; they are not a vote of confidence in every sentence. The pending authorization concerns those named artifacts; their presence here is not authorization of a future release. The same discipline separates an old command from a current warrant. Identity, evidence, and permission are three things a careful book can carry without pretending they are one.