AI-AID: model of the languages/ folder
For Kiro and any other AI assistant. Read this before editing or summarizing the folder. Each subfolder has a DEGREE.MD stating how complete and valid it is.
Purpose
languages/ holds three independent things. Do not merge them.
| Part | Files | Model |
|---|---|---|
| Localization packs | *.lang (12: en es fr de pt it ja zh hi ko th ar), KEYS.md, settings.conf | key = value UTF-8 text. English is the reference key set. Missing keys fall back to English. Read by lib/languages (SLLanguage, SLLanguageCatalog). Documented in README.md. |
| Language research profiles | 80 folders named by language (english, mandarin-chinese, ... maithili) | Top 80 languages by total speakers (2025 ranking). Each holds README.md (research template), alphabet.json, ALPHABET.xml. Statistics are per country and year, never traits of speakers. |
| FRUCHST dictionaries | FRUCHST_FORMAT.md, FRUCHST_SOURCES.md, FRUCHST_STATUS.md, FRUCHST_TARGETS.json, datasets/, tools/ | One binary fruchst.<code>.bin per profile: 1,000 frequency-ranked native headwords, each with aligned English, French and Spanish definitions. |
Ground rules
- A profile folder is not "done" until its corpus, definition sources and licenses are reviewed and its binary validates (
tools/fruchst.py). - Never invent ranks, frequencies, definitions, statistics or sources. Unknown means "to be researched" or "none".
- Never present frequency candidates as approved definitions. Every candidate is
review_status: required; approved count is 0. - Keep the three parts independent. Alphabet files and
.langpacks do not depend on FRUCHST. - Social indicators (education, fertility, savings, well-being) belong to geographies, not languages. Do not compute a combined "aptitude" score.
- Respect upstream licenses recorded in
FRUCHST_TARGETS.json(license_review). Do not copy third-party text that is not listed there. - Source of truth for numbers:
FRUCHST_STATUS.mdandFRUCHST_TARGETS.json.DEGREE.MDfiles are derived snapshots (2026-11-16). On conflict, the sources win and theDEGREE.MDshould be refreshed.
Pipeline stages (used by every DEGREE.MD)
| Tier | Meaning |
|---|---|
| C0 | Scaffold only. No licensed frequency source found. |
| C1 | 1,000-word list staged. Definitions negligible (under 50 words with all three). |
| C2 | Word list plus partial EN/FR/ES candidate definitions. |
| C3 | Near-complete candidates, still unreviewed. |
| C4 | Reviewed and approved definitions. (none yet) |
| C5 | Binary built and validated. (none yet) |
Degree percent = 10 (scaffold) + 20 (word list) + 30 x (all-three definitions / 1,000). Review and binary make up the other 40 points, currently 0 everywhere. No folder can exceed 60% until review starts.
Overall state (2026-11-16)
- 51 of 80 profiles have a 1,000-word list; 29 have none (no licensed source).
- 0 approved definitions, 0 binaries built.
- Profile
README.mdresearch fields are untouched templates in all 80 folders. - Alphabet inventories are
provisional.
Where to look
- Format:
FRUCHST_FORMAT.md. Targets:FRUCHST_TARGETS.json. Status per profile:FRUCHST_STATUS.md. - Pipeline scripts and tests:
tools/(seetools/DEGREE.MD). Data:datasets/(seedatasets/DEGREE.MD).