With the single exception of your first word, I entirely agree with what you said. I just don't see the equivalence between the quoted paragraph and your explanation. Perhaps because English is not my native tongue.
As far as I can figure out what you are arguing, you essentially claim that the words "that model" are an ambiguous, or incorrect reference to "the code set independent model". Well, since there isn't any other "model" in the preceding sentence, the reference is rather specific and clear.
The summary of the what the code set independent model attempts to do may be a bit terse, but I see no actual discrepancy with the expanded version of the explanation provided by van. Unicode certainly does not provide "late-binding" of locale-specific character properties, because most character properties that Unicode deals in are locale-independent and therefore 'early-bound' to use that term here.
String handling in Unicode is also not done on "byte streams of unknown semantics" - any well-formed stream of code units always resolves to known Unicode characters with known semantics under the Unicode scheme of string handling.
Leaves, as the only point of confusion the phrase: "to protect implementations from the details of hundreds of different character encodings". Under the code set independent model, applications are directly exposed to different character encodings - the data they process are always represented in the original encoding. Hence the need to "protect" by attempting to let implementations process these byte streams blindly up to the point where locale-specific properties are bound to characters.
A Unicode-based implementation may have an outer layer that accepts other encodings, but only in terms of providing a mapping to and from. The core of the implementation always works in Unicode, hence there's no need for the Unicode scheme of string handling to "protect" the implementation. Instead, implementations can fully rely on the identity and properties of the Unicode characters they process at any point in the implementation.
There are some operations, such as special case transformations or sorting that do need locale-dependent processing (and therefore properties) and Unicode doesn't provide any obstacle to implementing those kinds of operations. The difference is that it doesn't force the locale dependency onto every single operation, as the code set independent model would do.
FYI, there are a number of contributors on the Unicode Editorial Committee whose native language is not English. They usually speak up if some draft text is worded in a way that might be unnecessarily difficult to understand by non-natives.