Parlotype's localization gate landed before roughly 450 keys left the markup

The checker is four file rules and a counter. `scripts/check-localization.ps1` verifies key parity across locales, placeholder parity (the `{0}` counts must match, or `string.Format` throws in front of a user), that every `{loc:Tr}` key in markup resolves, and it scans `.axaml` files for hardcoded literals that have not moved yet ^[11]. The counter is `scripts/localization-baseline.json`, which records how many hardcoded literals each `.axaml` still has ^[15].

Both cases are spelled out in the hook. A stale baseline with nothing broken tells the agent to re-baseline; a new literal, a missing key or a placeholder mismatch exits 2 and blocks ^[16]. Exit code 2 is what makes a Claude Code hook blocking, and the same guard exits 0 on any internal failure of its own, so a bug in the tripwire cannot wedge a session ^[17]. I think that default is right. A guard able to hang the session it is protecting gets switched off within a day.

The ordering argument is about cost. An agent doing bulk extraction across 26 markup files will drift: miss an attribute, invent a key naming scheme halfway through, translate one locale and forget the other ^[10]. "A human reviewer catches that at review time, which is the expensive moment. A checker catches it in the loop, which is free," Parlotype's developer wrote in the dev.to post ^[9][26]. He also broke each check on purpose before trusting it, deleting a key from one locale, adding a literal to markup, mismatching a placeholder, and says that habit is what exposed the first blind spot ^[18].

One architectural call had nothing to do with strings. `x:CompileBindings="True"` is mandatory in this codebase and `{ReflectionBinding}` is banned, but live language switching wants a binding to a localizer lookup, exactly the indexer-or-method shape the ban targets, and he went in expecting to write a narrow documented exemption ^[19]. Avalonia 12 removed the need for one: `CompiledBinding.Create<TIn, TOut>` takes an expression and a source object, so a plain property access on a known type compiles ^[20]. `Localizer.Entry(key)` returns one cached `LocalizedString` per key, an `ObservableObject` whose `Value` re-reads the `ResourceManager` at the current culture, and that per-key object exists so the binding expression can be `s => s.Value` instead of an indexer ^[21].

Some sizing, since the post gives enough to do it. 389 keys in three languages is 1,167 translated entries ^[22]. The diff nets 10,849 lines across the 122 files, about 89 each ^[23]. The starting inventory of 211 literal attributes and roughly 200 view-model literals comes to about 411, against the roughly 450 keys the guardrails were built to precede ^[24].

The post says three user-visible defects got past all four guardrails, that each was a different category of blind spot, and that none was a translation error ^[6]. It does not identify them in the sections that describe the guardrails. The reach of those checks is bounded by what they read: all four open files, so a defect that only exists once a window is rendered sits outside them by construction ^[25].