# Changes to BÍN data

The publisher's 15-field Kristínarsnið CSV is filtered to `alm`, lemma rating 1, form rating 1 and core visibility `K`. Historical or rare lemma/form style markers, phrase-bound lemma/form markers, and weak or unsettled form-value markers are excluded. Only exact lowercase NFC single-token written forms with 2–30 code points from the locked Icelandic search alphabet are retained; punctuation, spaces, numbers, unsupported letters and uppercase-only names are rejected. Duplicates are collapsed by exact written surface. No letters are removed from source forms to create a word. No new inflected form is generated. `curation-v1.json` is empty for this lock and binds the exact publisher CSV hash; later changes require individually sourced operations. All retained rows are assigned rank 1 and standard flag 1, then encoded into 59 sequential IAG2 files using the corresponding source included here. The list is an adapted selection, not an unmodified BÍN export or a complete dictionary.