The word "entities" in the quote does not mean "entities" as specified in [basic.pre] p1, I think (for example, a typedef-name is not an entity).

And since the pattern can transform a pack into a different syntactic thing (for example, from type to expression), the "considered to be" wording is just wrong, because it mixes levels.

I think we should not try to differentiate semantic and syntactic rules here.

In the origin rule, the "entity" is just misleading since we basically refer to these entities defined in [basic.pre] p3 when we mention it in this document. This is also implied by this phrase

regarding entities other than packs

A pack is an entity defined in [basic.pre] p3.

The clarification is just want to make the pack that appears in a construct can be first validated by the relevant grammar(syntax), such as an id-expression cannot appear in a type-specifier, and not try to differentiate semantic and syntactic rules here. The thought is to give the pack a grammatical category, in other words, clearly know whether the pack is interpreted as a typedef-name, or id-expression, or a template-name. For example, initializer-list is defined as:

initializer-list

• initializer-clause ..._opt

{ T... };

Given the construct, it is ok if T is a non-type template parameter pack that is interpreted as an id-expression but it is wrong if T is a type template parameter pack declared without template that is interpreted as a typedef-name since a typedef-name cannot be an initializer-clause. This is the intent here. The semantic rules can also be checked for the construct in both template definition context and instantiation context.

And since the pattern can transform a pack into a different syntactic thing (for example, from type to expression), the "considered to be" wording is just wrong, because it mixes levels.

As above said, consider this example

template<class...T>
void show(){
    auto c = {T...}; 
}

it is just wrong, T is interpreted as a typedef-name regardless of what the pattern is.

We agree about the intent of the rule, but what exactly is wrong with the current formulation other than the unfortunate mention of "entity" (which is defined to mean something else)?

Except for "entities", the remaining wording seems good to me. We may fix it with other approaches.