Coming to this relatively new and reading CIBA with the perspective of potentially implementing it (due to FAPI profiling CIBA), I feel there's some ambiguity around the format of the Authentication Request (sec 4.1 in the published version and sec 7.1 of the latest from source). The example shows a POST with application/json in the body while the text defers to authorization/authentication request of OAuth/OIDC, which uses application/x-www-form-urlencoded parameters. The text (in the latest source) also says that OpenID Connect Signed Request Object (referencing OIDC §6.3.2) can be used, which could be interpreted to mean form-urlencoded parameters with request=[JWT] or request_uri=[URL to JWT] and also requiring the scope, response_type and client_id parameters to be present as form parameters and that their values match the same named claims in the JWT. OIDC also says that parameters MAY also be passed using the OAuth 2.0 request syntax (regular form parameters) even when a Request Object is used and that both are considered with the values contained in the JWT Request Object superseding those passed using the OAuth 2.0 request syntax. I rather doubt that the intent of CIBA was to bring in all that baggage from OIDC's request object but that's how OIDC defines it and so is a not unreasonable interpretation of the CIBA draft, particularly for someone familiar with Connect or pedantic about reading its details.
I see the idea behind CIBA wanting to just kinda defer to the OAuth 2.0 Authorization Request / OIDC Authentication Request in this back-channel context but those are designed to pass through a user's browser via redirection, which is rather different than direct a client to AS request like CIBA needs. I think basing the CIBA request on the OIDC/OAuth front-channel requests actually causes more issues than it solves. And it'd be more appropriate and straightforward for CIBA to directly define the parameters and format/encoding. The relevant parameters are already defined in the Authentication Request section. The format/encoding just needs to be more clearly stated. From what's there now, I suspect that the intent is to have the request be JSON in the body with the parameters as top level members of that JSON and the content-type would be application/json. That's what the example in the draft has now but the normative text doesn't unambiguously say it. For a 'signed request' the content-type should probably be application/jwt with the body of the request being just the JWT that's payload is the claims that would be the JSON of the actual request parameters. The JWT should probably be required/encouraged to also have an audience and expiration.