# Icelandic Words source

The selected source is Kristínarsnið, a downloadable CSV export of Beygingarlýsing íslensks nútímamáls (BÍN), published by the Árni Magnússon Institute for Icelandic Studies. Author and editor: Kristín Bjarnadóttir. The publisher's download page is https://bin.arnastofnun.is/gogn/mimisbrunnur/ and the exact archive URL is https://bin.arnastofnun.is/django/api/nidurhal/?file=KRISTINsnid.csv.zip . The source was fetched on 2026-09-26T12:51:31.511Z. The publisher ZIP contains `KRISTINsnid.csv` and a SHA-256 checksum member. The ZIP is 35,656,625 bytes, SHA-256 `fa2468a941fe288a10fa541b86422a4cb0432199590d6d4b411a32a34971e1cf`; the CSV member is 470,870,708 bytes, SHA-256 `bf477c19b9f8245a4f0dfc4c8a61a53d89aa01b499d3eb0dbb373a9e15c4190b`. The ZIP member timestamp is 2026-09-10 21:09:14 UTC. The member has LF line endings and strictly decodes as UTF-8; the publisher did not declare a charset on the HTTP download. A strict decoder is used, without replacement or ignored errors. No encoding conversion is applied. The archive and CSV bulk are intentionally not redistributed here.

The source format and field meanings are documented by the publisher at https://bin.arnastofnun.is/gogn/k-snid . Fields 4, 5, 9 and 12 are used for `alm`, lemma rating 1, core visibility `K`, and form rating 1. Style, grammar and variant fields 6, 7, 13 and 14 exclude old, rare, bound and weaker forms as detailed in `policy.json` and `generate-candidate.mjs`. Each retained row supplies a complete written surface; no Hunspell affix or compound generation is used. The candidate has 686,597 unique lowercase NFC surfaces of 2–30 letters. The candidate and package are adaptations; see `MODIFICATIONS.md`.

`provenance.tsv.gz` is a deterministic gzip of a UTF-8/LF TSV with one selected BÍN source line, BIN.ID, lemma, word class, inflection tag and admitted-row count for every candidate surface. It is a source-traceable projection of the pinned CSV, not another lexical source. The full CSV can be re-downloaded and hash-verified to inspect any row.

LibreOffice Dictionaries Icelandic files at commit `32b006a2c22a4ac7e8ed3f03346f7b3d85a970a4` were evaluated but not used for candidate membership, rank or curation. Their mixed provenance and noisy literal-entry results are described in the PREQ audit. Hyphenation and thesaurus material are unused. No frequency data is used.