Matt Stark's patch changes sequencer.c so that a change-id header already on a commit is kept when the commit is rebased or cherry-picked. Stark describes earlier discussion as showing consensus that such a header is valuable but no consensus on its precise format. His patch therefore does not define a format. It preserves the header as it was, and he argues that since Git only preserves existing headers, a standard format could be chosen later without breaking compatibility.
brian m. carlson raised a security objection to one use of such headers. Users can edit change-id headers, so they cannot be trusted as unmodified data. He gave an example in which a malicious user gets a review on v1 of a series and then sends a different v2 with the same change-ids. A server that honored the existing review could accept malicious code. In his view, only a server that performs the rebase itself can be confident nothing has changed.
Ewout ter Hoeven offered a concrete use case: stacked pull and merge requests. He said GitHub's gh-stack CLI and GitLab's experimental glab stack both have to decide whether a rewritten commit is the same change as before. Without an identity that survives rewrites, they guess from branch names, recorded bases and patch IDs. He said these guesses fail in practice, citing a gh-stack issue where renaming a branch with git branch -m strands the stack, and a case where old commits are replayed onto new copies after a lower layer is rewritten.
Junio C Hamano, Nico Williams and Phillip Wood also posted in the thread, though their messages are not in the excerpts. The format question remains open, as does whether preserving the header without any verification is enough.