Validation & quality

Quality linting

Catch awkward, informal, or inconsistent copy — statically, and optionally with an AI pass.

Running quality#

quality lints a locale's text (the source locale by default) and prints suggestions grouped by severity. It runs entirely offline unless you pass --ai.

shell

bin/lingo quality                 # lint the source locale
bin/lingo quality --locale de     # lint a target locale
bin/lingo quality --ai            # add an AI review pass

Static rules#

A set of regex rules flags common issues without any model call:

Terminology — cannot vs can not, verb log in vs noun login, click here, doubled please.
Clarity — vague words (stuff, things), abbreviations (ASAP, FYI), double spaces.
Business tone — sorry, oops, awesome, cool.
Accessibility — positional references (see below, above), colour-only meaning.
Placeholders — spaces inside %{…}, uppercase placeholder names.

quality.terminology selects a built-in list — business (the default) or banking — or a path to your own YAML mapping terms to suggestions (a nil suggestion marks a term as reviewed and acceptable). Flagged terms produce info-level suggestions, e.g. "wire transfer → consider 'bank transfer'".

British spellings#

With quality.british_spellings: true, the source locale is checked for American→British drift (organization → organisation, color → colour, analyze → analyse, …). These are auto-fixable.

The AI pass#

--ai samples the locale and asks your quality.model to suggest improvements for clarity, professionalism, and friendliness, each tagged with a severity. It samples rather than reviewing everything, to keep the cost bounded. Needs provider credentials; without them it warns and skips.

Auto-fixing#

fix-quality rewrites the fixable suggestions — the universal fixes (can not → cannot) and British spellings — back into the locale files, preserving the original case. Preview first with --dry-run.