An agent finds another agent through a registry: an index, a name service, a directory. It asks for a name and gets back a record — an endpoint, a key, a set of capabilities — and acts on it. The record may be signed, and the signature may verify. On this the existing systems are complete: a signed record cannot be altered in flight without the signature breaking.
But a signature guards the wrong thing. It proves a record was not modified; it says nothing about whether the registry served the record it should have, or the same record to everyone. A registry can hide an agent it does not want found. It can serve a forged endpoint whose signature checks out against a key the same registry controls. It can answer one caller honestly and another falsely, and no single caller can tell. The trust in a lookup rests entirely on the registry's good behavior — the one thing a signature does not attest.
This is the corroboration gap. Discovery is trusted on faith. And it is exactly where a single source is weakest: with one registry there is nothing to compare against, so omission, tampering, and equivocation all pass silently. The fix is not a better signature — the records are already signed. The fix is a second opinion.
The Quilt is that second opinion, made systematic. It asks several independent registries the same question, from several vantages, and compares the answers. Agreement is evidence; disagreement is a finding. The outcome is one signed, content-addressed Corroboration Record that converts "the registry said so" into "these independent sources agreed — and here is the signed sweep that proves it." A cheating registry stops being undetectable and becomes, at last, auditable.