Antoine,

Yes I think you're a bit confused on where the actual sponsor vector is. If you have a transaction chain A->B->C and a sponsor S_A, S_A commits to txid A and A is unaware of S.

W.r.t your other points, I fully agree that the 1-to-N sponsored case is very compelling. The consensus rules are clear that sponsor commitments are non-rival, so there's no issue with allowing as many sponsors as possible and including them in aggregate. E.g., if S_A and S'_A both sponsor A with feerate(S*) > feerate(A), there's no reason not to include all of them in a block. The only issue is denial of service in the mempool. In the future, it would definitely be desirable to figure out rules that allow mempools to track both multiple sponsors and multiple sponsor targets. But in the interest of KISS, the current policy rules are designed to be minimally invasive and maximally functional.

In terms of location for the sponsor vector, I'm relatively indifferent. The annex is a possible location, but it's a bit odd as we really only need to allow one such vector per tx, not one per input, and one per input would enable some new use cases (maybe good, maybe bad). Further, being in the witness space would mean that if two parties create a 2 input transaction with a desired sponsor vector they would both need to specify it as you can't sign another input's witness data. I wholeheartedly agree with the sentiment though; there could be a more efficient place to put this data, but nothing jumps out to me as both efficient and simple in implementation (a new tx-level field sounds like a lot of complexity).

> n >=1 ? I think you can have at least one vector and this is matching the code

yes, this has been fixed in the gist (cred to Dmitry Petukhov for pointing it out first), but is correct in the code. Thank you for your careful reading.