Thanks for the detailed proposal. Since you opened this, the whole Factory/Accessor chapter was reworked (audit pass against the source), and I think it now addresses your main concerns — mostly in the spirit you intended:

• Corrected tag example (your key point): the tag-based definition is now clean and correct — create: and tags: as siblings, tags: {multi: article} where the tag value becomes the method suffix, plus MultiFactory(tagged: multi). I verified tagged: against LocatorDefinition. So the misleading sample you flagged is gone.
• get() vs create() and instance semantics: the parametric section now explicitly says create($name) returns a new instance on each call while get() returns a shared one — which covers your "new instance per create()" note.
• The "disadvantage of the alternative syntax" sentence: I kept it deliberately — it's genuinely useful design guidance (you can't tell which $name values are supported, and you can't differentiate return types), so I'd rather not drop it.
• The "Definice seznamem" / "Definice pomocí tagů" subheadings: kept — they help readers jump to the two definition styles, and the chapter boundary is clear from the heading levels.

A couple of the remaining ideas I didn't take: the "dependencies are created lazily at create() time" note isn't quite accurate (a generated factory receives its own injected dependencies when it's constructed, not at create() time), and "configuration is required for a multiple factory" isn't universally true (a factory with typed create<Name>(): Type methods can be resolved from the return types), so I'd rather not state either as a blanket rule.

Closing as superseded by the reworked chapter — thanks for pushing on this section, it clearly needed the attention.