oficina / suite Public

Revision 10: AHEAD, and the half of it that would have opened an oracle

The memoria session reported that S2's four values have no home for the
source-behind case, and that revision 7's §4.3 sends it to UNKNOWN. Both
true, and the sentence is mine.

STALE means the CONSUMER is behind — the app of record moved past the
revision — and refetching fixes it. The opposite happens too: a sibling
restored from backup holds revision 5 of an item the consumer indexed at
7. UNKNOWN means "nobody answered usefully", and here the app of record
answered precisely, about its own log, correctly. Reporting no answer when
there was one is the plausible-empty-list collapse this vocabulary was
built to prevent, arriving through the last door left open — on the day a
sibling is restoring from backup, which is the day the suite is least
healthy and least likely to be believed.

So AHEAD, as memoria's draft 4 §4.1 named it. Their argument, their name.

**I did not adopt the whole rule, and the part I left out matters.** Draft
4 also sends "the sibling has no row for this item at all" to AHEAD, on
the reasoning that it is the same position: the consumer holds history the
sibling does not. As an account of what happened that is right. As an S2
answer it cannot stand, because gate 3 requires S2 to answer NOT_VISIBLE
byte for byte for an item that is private, granted, guest-only,
link-shared-but-not-team AND DELETED. If "no row" answered AHEAD, a
consumer could tell a deleted item from a private one, and could probe
guessed ids for which exist — the existence oracle §4.2 exists to deny,
opened by the value added to close a different silent failure.

That leaves a real gap, and §4.3 now says so rather than implying it is
solved: a sibling whose restore lost a row entirely is indistinguishable
from one enforcing a denial. The S4 tombstone on the feed settles it after
the fact, which is not the same as answering it at the point of the
question. Recorded as an open tension between memoria's draft 4 and this
document, because both rules are load-bearing and choosing between them is
Paul's.

Gate 10 is appended rather than inserted where it belongs. Gate numbers
are cited from other repositories — home §12 and sheets §12 both name gate
9 — and renumbering a list other documents point into is how a citation
silently starts naming the wrong rule. I renumbered it once while writing
this and caught it in the diff.

Module: AHEAD added to the closed vocabulary, three tests, and a tenth
mutation (AHEAD dropped from known()) which is caught. The decoder never
synthesises AHEAD — it is a verdict a sibling gives, so a failed call or a
skipped item stays UNKNOWN, and a test pins that from the other side too.

One note on how this arrived. The memoria session told me draft 4 was on
memoria's main. It is not — main is draft 3, and AHEAD lives on unmerged
branches at draft 5. That does not weaken the argument, which stands on
its own, but it is the second time in two days a suite ruling has been
made against a memoria draft that had already moved. §2's lesson, one
repository over, twice.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01RVcEQ8NgrPCpRTb6JtBz1Q