i18n readiness check

The check reads your project on your machine and reports, per target market, what is done, what is broken and what has not been verified. Source code never leaves the machine. Only the report is uploaded, and only if you pass --upload.

#Run

npx threefold check . --markets es-MX,de-DE
npx threefold check ./app --markets es-AR,ja-JP --json > report.json
npx threefold check . --markets de-DE --runtime web --url http://localhost:3000

Exit code 1 means the verdict is incomplete; 0 means complete or complete_static.

The platforms are detected from the project: iOS (String Catalogs, .strings, .stringsdict, InfoPlist.strings), Android (res/values-*, locales_config.xml), web (i18next, next-intl, FormatJS JSON, JSX), Flutter (.arb, l10n.yaml), React Native with Expo, and store metadata in fastlane folders or a JSON file.

#The report

• checks: every applicable check with pass, fail, warn, unverified or not_applicable, evidence (file, line, snippet) and a fix hint.
• verified_coverage: the share of applicable automated checks that actually ran. It is not a pass rate.
• verdict, overall and per target market:
  • complete: every applicable static and runtime check passed and nothing required is unverified;
  • complete_static: static checks all pass, runtime checks did not run;
  • incomplete: at least one check failed.
• human_review: market legal items and anything else that needs a person. They are listed next to the verdict and never counted as done.

There is no blended readiness percentage. Checks whose measured precision or recall is below the bar are labelled (beta) in the report.

#Checks

37 checks. Each declares its platform, severity (blocker, error, warn or review), how it is verified (static, runtime or human) and its source.

Store limits checked include App Store name 30 characters, subtitle 30, keywords 100 bytes, promotional text 170, description 4,000; Google Play title 30, short description 80, full description 4,000.

Runtime checks for iOS and Android build the commands and parse the results, but without a simulator or emulator they report unverified.

The full catalogue, with sources, is served by the API: GET https://api.threefold.eco/api/i18n/rules (filter with platform and market), or via the MCP check tool with what: "repo".

#Benchmark

The checks are measured against four open-source projects (pinned commits of Next-js-Boilerplate, IceCubesApp, Fossify Gallery and flutter/gallery) and copies with hand-inserted defects. A check needs at least 90% precision and 90% recall on at least five verified positives to ship as production. In the run of 7 November 2026, 19 of the 27 measured checks met the bar and 8 are labelled beta; the iOS and Android runtime checks were not measured. See Translated isn't localised for the details.

#In CI

The threefold repository includes a GitHub Action (action/) that runs the check on pull requests, posts the report as a single comment it keeps up to date, and fails the job on incomplete. Inputs: path, markets, runtime, url, fail-on-incomplete (default true), github-token, upload. Outputs: verdict, verified-coverage.

#With a coding agent

The localise-app skill gives an agent the loop: run the check, fix every fail, translate missing strings with the market resources, lint variants, check the store copy, re-run until the verdict is complete or complete_static, then show you the human-review list. It tells the agent never to invent pass results or claim legal compliance.

Coding agents run the check with npx threefold check in the project, or over MCP using the check tool with what: "repo" (docs/agent-tools.md). On the local stdio server (node mcp/server.mjs), check what: repo scans files on the machine. On the hosted server (https://api.threefold.eco/mcp), it returns the market readiness rules and criteria for the agent to apply directly to open repository files.

#Stored reports

POST /api/i18n/reports stores an uploaded report for your workspace; GET /api/i18n/reports/:id reads it back.