Market resources

Market resources come from the library, so they reflect what is in each storefront now. Each cites the apps it learned from. Markets, niches and stores are the same as everywhere else (see the overview).

(See Agent tools for the MCP tool definitions).

#Market glossary

Inputs: niche, market (required); variant, store, format (json, csv, tsv, tbx), include_foreign.

The terms and calls to action that top-ranked listings in that niche and market actually use. Phrases are n-grams over stored listing descriptions, split at sentence punctuation, with stopwords trimmed only at the edges so phrases such as “tarjeta de crédito” and “meses sin intereses” survive; Japanese is segmented with Intl.Segmenter. Scores are document frequency, lowered for terms the same niche uses in other markets and for filler that every niche in the market uses.

• Only listings detected in the target language variant count; excluded_non_matching_listings reports how many were left out.
• Terms that name another market (“colombia” in es-MX) are flagged foreign_market_reference and left out unless include_foreign is true.
• Every term and call to action lists the app ids it was found in.
• tsv is laid out for DeepL glossaries; tbx is the standard terminology exchange format.

#Variant lint

Inputs: text, variant (for example es-ES, es-MX, es-AR). Free.

Deterministic rules with a source for each: vosotros forms in Latin America, tú imperatives where Argentine calls to action use voseo (descargá, suscribite), and vocabulary that marks a variant (móvil and celular, ordenador and computadora, coche and auto). Every issue is a warning with the matched text, its position, a suggestion and the source URL. It is the same rule set the i18n check uses.

curl -X POST https://api.threefold.eco/api/variant-lint \
  -H "Content-Type: application/json" \
  -d '{"text": "Comparte tus fotos con vuestro equipo desde el móvil.", "variant": "es-MX"}'

That text gets two warnings: VAR-ES-VOSOTROS-LATAM on “vuestro” and VAR-ES-VOCAB-MOVIL on “móvil”.

#Listing check

Inputs: niche, market (required); the proposed listing (title, and over MCP the other fields such as subtitle, description, keywords, screenshots); store.

Validates the copy against the store's field limits (App Store name 30 characters, subtitle 30, keywords 100 bytes; Google Play title 30, short description 80), then compares it with the market's top listings: title and subtitle length, screenshot count, how many competitors localised at all, and overlap with the market glossary. Every comparison cites the competitor app ids.

#Price points

Inputs: niche, market (required); store.

What the top apps in a niche charge in that storefront, from stored listing prices: counts and shares of free, paid and unpriced listings, the price distribution (minimum, quartiles, median, maximum) in the local currency, and the rounding convention in use. When every top listing is free to download, it says so. Partial Google Play rows are flagged.

#Market launch checklist

Inputs: market (required); report_id (a stored i18n report).

One list per market, merging the market profile (language variant, channels, monetisation note), the B2B entry checklist, the legal review items from the i18n check and, if given, the status of your stored report. Statutory items are prompts for a person; the checklist never certifies compliance.

#Translate

Inputs: texts (required); target, variant, formality (for German: formal Sie or informal du), glossary (your own term pairs, or a niche and market to use threefold's market glossary), niche, market.

Batch translation that preserves ICU tokens ({count}, {name}) and printf specifiers (%s, %@), then runs the placeholder check on the output and flags anything it broke. It is an AI analysis, counted against your plan. Use it for drafts and for reading evidence; have a person review copy before it ships.

#The loop

The localise-app and launch-in-market skills chain these workflows over MCP using market and check (docs/agent-tools.md). threefold provides the evidence and criteria, and the agent grades the material and cites the links.