> I hope this argument is more or less convincing.

It's a very well laid out argument. Thank you for writing it out in
detail, and sorry for not having particular comments on the details.

I don't have a good grasp on how complex ASN.1 is to implement if you
leave out all the schema stuff and pick suitably simple encoding rules.
It'd be interesting to see a prototype implementation.

Do you think the simplest case is simple enough that my separate binary
S-exp format should be abandoned? I'm not quite convinced that something
of approximately equal simplicity can be crafted by suitably subsetting
ASN.1, but I do not know this.

Crucially, those binary S-expressions are also easy to use from C (well,
as easy as any hierarchical dynamically-typed structure can be). I like
to implement new formats in C; if the pain is tolerable there, it's
tolerable anywhere.

ASN.1 seems to use OIDs (dotted names). How much of a core part is that?
With hindsight, we can now say that it might be simpler to use repurpose
the internet DNS for a name hierarchy.

Adding custom datatypes needs to be baked into any such format. It needs
to be possible for an implementation to skip advanced datatypes that it
doesn't understand. This means that values of those types need to be
prefixed with their length somehow, or be built entirely from
length-prefixed parts.