# Indonesian dictionary modifications

1. The pinned LibreOffice `id_ID.dic` independent entries were screened with the unchanged pinned `id_ID.aff` flags. NEEDAFFIX and unknown flags were excluded; no affix engine or generated inflection was used.
2. Only lowercase NFC a–z single-token forms of length 2–30 with at least five surface occurrences in Leipzig Indonesian Mixed 2013 100K were retained. Leipzig count descending, then lexical word order, assigns rank.
3. Wiktionary Indonesian via Kaikki supplied whole-dataset POS, sense and gloss screening. MALINDO Morph supplied lexical witness and exclusion evidence. The 844 reviewed removals in `curation-v4.json` bind exact words to raw dictionary line and surface, evidence and reason.
4. The 4,604 resulting rows were written to the preferred editable `candidate.tsv`, then converted to the 29 content-addressed IAG2 files using the repository IAG2 generator. The data asset is GPL-3.0-only. Raw upstream files and their notices remain unchanged.

The source identities, retrieved versions, URL and hashes are in `source-identities-v4.json`; the exact generation metadata is in `generation-lock.json`.