Right now BOLT 3 is trying to describe the keys derivation from the view point of both nodes, when it should only describe from one side, i.e from a local node building and signing a remote commitment transaction.
In a RevokeAndACK, Bob send his per_commitment_point to Alice, which uses it to derive a set of keys for next commitment transaction:

* `local_delayedpubkey` to encumber  to_local output and HTLC-Success/HTLC-Timeout outputs

* `remote_pubkey` to encumber to_remote output

* `local_htlcpubkey` and `remote_htlcpubkey` to encumber offered/received HTLC outputs

At no moment, she need to derive localpubkey or remote_delayedpubkey ones.

Yes, when Bob is going to verify the sigs, he is going to do the same derivation and from his viewpoint, he may consider local_delayedpubkey is remote_delayedpubkey and remote_pubkey as localpubkey but that implementation-dependent and shouldn't be in spec.

No, that is absolutely required by the spec. You need to be able to generate both local and remote commitment transactions.

Unfortunately, that does make it more complex! Perhaps the description can be simplified somehow?

Yes, you need to generate both local and remote commitment transactions, but even when you build a local commitment transaction for yourself, there is no such thing as a localpubkey or remote_delayedpubkey. Transactions description should be described from a unique viewpoint, the one for which the commitment transaction is intended, and so pubkey names should be same on both sides.

On the implement-side. at no moment you need to derive the set of 6 keys like the spec let it suppose, every commitment transaction is going to only use 4 of these keys.

Is there a better way to describe this ambiguity than my current diff ?

This PR does not matter substantially for the spec.