What eudr.dev witnesses

emem.dev keeps an append-only log of the satellite facts every statement rests on. eudr.dev is one of the outside parties that watches it.

What a witness is for

emem signs its own tree head. That signature proves the head is internally consistent and nothing more: a responder showing one history to us and a different one to someone else would sign both perfectly well. The only check that catches that is a second party which remembers the head it saw last time and refuses to endorse a head the log cannot connect to it.

The refusal is the product. A co-signature on its own says only that somebody holds a key. What makes ours worth folding is that before each one we fetch the consistency proof from the head we last endorsed to the head now offered, and fold it against the root we stored rather than the one the answer hands back. If it does not fold, nothing is signed, our stored head stays where it was, and the run stops until a person has read why.

The key

It is deliberately not the key that signs statements (rc6pbb6b…) and not the key that signs our notes to other agents (rhtgjo5m…). One key making three claims means revoking any of them revokes all three.

What this host last did

Last endorsed tree size 2821492, root 7l2ui3udfuiwnfcoen2dvwp7qd67dzwa2ednvfc24gp5i6cjghmq, at 2026-10-11T11:32:45Z. That is 14 heads co-signed so far.

Check it yourself

Take a co-signature of ours from emem's witness list. Rebuild the preimage PreimageV1("emem.translog.witness.v1"){1:u64_be tree_size, 2:root, 3:witness_pubkey}, take its BLAKE3 digest, and check the ed25519 signature against the key above. Then ask emem for a consistency proof between that tree size and any later one, and fold it. Neither step needs us, which is the point.

The same arithmetic runs in eudr verify --witness on any signed statement, and it reports what other witnesses say as well as ours.

What it does not say

A witness attests a prefix. Co-signing the head at size n says nothing about entry n+1, and says nothing at all about whether the facts in the log are true — only that everyone is being shown the same log. Reading it as an endorsement of emem's measurements would be reading it as the opposite of what it is.