TheQuilt Working draft

The Quilt · Corroborated Discovery for the Internet of AI Agents

Making the Registry Prove It

A registry can hide a record, forge an endpoint, or tell two callers different things — and no signature on a single record can catch it. The Quilt asks every registry the same question and makes any disagreement loud.

Corroborated discovery turns a registry from an authority you trust into a source you can audit. Ask several independent registries the same question, from several vantages, and treat any disagreement — a hidden record, a forged endpoint, an equivocating answer — as a signal. The result is one signed, content-addressed Corroboration Record: a verdict about a subject that anyone can re-verify offline. The index points; the registry proves.

What it catches

A signature proves a record was not altered. It says nothing about what a registry chose not to serve, or whether it served the same answer to everyone. Corroboration across registries catches the three ways a registry lies.

Omission

A record was hidden

One registry serves an agent that another registry confirms absent. A single caller cannot tell a hidden record from a record that never existed; two registries can.

Tampering

An endpoint was forged

Registries claim different addresses for the same agent. The unsigned endpoint a caller would connect to disagrees across sources — one of them was substituted.

Equivocation

Different answers to different callers

A registry serves one identity here and a different valid one there — across callers or across vantages. Its own answers, compared, contradict each other.

The verdict. A sweep resolves to AGREE (the decisive sources concur), DIVERGENT (they disagree — a finding is raised), or INSUFFICIENT (fewer than two decisive claims to compare). An unreachable registry is never mistaken for one that says a record is absent.

Out of scope. The Quilt does not judge whether an agent is trustworthy, capable, or well-behaved — only whether the registries that describe it agree. It attests the consensus about a subject, not the subject's merit.