The vault held
Sources cited in this piece · the sealed-records registry, github.com/Aicent-Stack/azuli · its rendered vault at aicent-stack.github.io/azuli/ · the canonical attestation bundle, iqa.org/attestations/ · the blog's own seal ledger, blog.aicent.com/attestations/ · dual-address resolver, iqa.org/open/?uri=….
On 10 November at noon, our German server went down for a planned 72-hour maintenance window — a virtualization upgrade announced by the provider in advance. IQA(iqa://)and RTTP(rttp://)and every family site in between went dark, for three days, by appointment. The same morning, a public registry went up that did not go dark at all: Azuli, a sealed-records registry holding 166 artifacts — and for the next three days it carried the one thing an offline period cannot be allowed to interrupt: the ability to check what we claim.
What the registry holds is the whole evidentiary estate, copied once and never rewritten. The conformance runners' sealed records and sidecars. The page-freeze tarballs of the family pages, report pages, the brief, the whitepaper, the URI reference — every generation of every page, superseded generations included, because a freeze that disappears when a new one appears is not a freeze, it is a current state. The release packages of the published implementations across all three ecosystems. And the two ledgers themselves: the release-attestation bundle and the blog's seal ledger. A SHA256SUMS file pins all 166 artifacts, and a cross-check of the vault against the ledgers reconciled every entity artifact, row for row.
The design rule that keeps a vault honest is the one that keeps it useful during an outage: the vault is availability, never authority. The canonical source remains the site that mints the records — when it is dark, its truth is offline, not gone. The registry holds copies identified by content digest, and states plainly in its README who operates it and that it vouches for nothing by itself. A copy you can verify byte for byte against a digest you already hold does not need you to trust the copier — which is fortunate, because the copier is us.
The maintenance window was the vault's first working day, and it earned its keep immediately. Five pull requests were open at the lab, each anchored to records that lived on the sites that were about to go dark. Each PR received a maintenance comment in advance: the canonical sites would be offline for the window, and the sealed records they reference remain reachable at the registry meanwhile. A reviewer mid-review on day two of an outage is the exact situation a vault exists for — the record survives the infrastructure that produced it, because it was copied somewhere else, by digest, before anyone needed it.
There is a naming honesty here worth a sentence, because our stack already has a mirror layer. MOLOON, the twelfth pillar, is about mirrors inside the protocol — periodic rewriting that keeps a system continuous across time. The registry does the opposite of that: it never rewrites, never sutures, never lets the new generation replace the old one. It is not a protocol function; it is an operational service standing beside the protocol. MOLOON manages mirrors inside the stack; Azuli keeps seals outside it — one in the protocol, one on the protocol's service surface, neither claiming the other's job.
The window closed early, the sites came back, and the registry stayed up — which is the quiet point. It was not built for the window; the window was simply the first time its reason to exist became visible to everyone. Every seal we mint already carries two addresses and a ledger row; the registry adds the least glamorous property of all, persistence of access — the confidence that when a reader arrives, whenever they arrive, the bytes are there.