Datasets:

endomorphosis
/

ipfs_namibia_laws

pretty_name: ipfs_namibia_laws
license: other
license_name: official-public-legal-texts-hub-rows-and-cc-by-nc-4.0-namiblii-rows
language:
  - en
task_categories:
  - text-retrieval
tags:
  - legal
  - law
size_categories:
  - 10K<n<100K
configs:
  - config_name: articles
    data_files:
      - split: train
        path: data/articles*.parquet
    default: true
  - config_name: laws
    data_files:
      - split: train
        path: data/laws*.parquet

Namibia Laws

Legal corpus for Namibia from the ipfs_datasets_py legal scrapers (endomorphosis).

Instruments: 1,609
Hub baseline: 530 instruments (endomorphosis/ipfs_namibia_laws@cffa06694596, last updated 2026-09-16)
Source of the baseline: official gazettes and legal portals; see source_url per record.
This revision (2026-10-30) adds 1,079 NamibLII instruments and 22,047 article rows on top of the baseline: laws/articles 530/5,791 -> 1,609/27,838 (1609/27838). The 530 baseline instruments and 5,791 baseline article rows are all kept (see "Changes to baseline rows").
Schema (laws): id, title, text, source_url, source_type, jurisdiction, country, language, eli, date, date_issued, retrieved_at, license, law_status, identifier, official_identifier, article_count, json_path, metadata_json. Schema (articles): law_id, id, title, text, source_url, document_number, article_number, record_type, metadata_json. Both files are ZSTD parquet.

Licence: read this before reuse

The baseline rows are public legal texts from official Namibian sources. The 1,079 rows whose id starts with na-namiblii- come from NamibLII (https://namiblii.org) and carry CC BY-NC 4.0 (non-commercial) in their license column. That licence was stated in the owner brief and could not be confirmed on a live document page (behind a Cloudflare challenge). Treat those rows as non-commercial until confirmed. The official gazette / authentic text prevails. Research compilation of public legal texts. Not legal advice.

NamibLII update (2026-10-23 collection, 2026-10-30 revision)

Adds 1,079 instruments / 22,047 article rows from archived copies of namiblii.org (Internet Archive Wayback id_ snapshots; Common Crawl was a configured fallback but every row here is archive_source=wayback). Each row's metadata_json records archive_source, archive_snapshot_timestamp, archive_url, original_url and content_sha256. Roughly 1079 rows cover ~46% of the NamibLII catalogue (about 2,298 documents in its January 2026 "All legislation" index), so this is partial coverage, not a full mirror.

Site terms and robots signals (disclosure).

NamibLII terms s4: "Scraping and bulk downloading from this website is strictly prohibited ... There are more efficient ways of obtaining access to our data in bulk. Please email us if you would like to access our data in bulk."
Terms s5: "You must adhere to the instructions in the /robots.txt file."
robots.txt (read 2026-10-23): Crawl-delay 5; disallows /api/, /accounts/, /judgments/, /gazettes/, /search/, /akn//provision/, /akn/na/judgment/, /akn/na/officialGazette/ among others; and Content-Signal: ai-train=no, search=yes, ai-input=no, which is an EU DSM Directive 2019/790 Art. 4 rights reservation against AI training and AI input use.
Acquisition: no live namiblii.org document page was fetched; text comes from archived copies only. The collection was directed by the owner on 2026-10-23 with awareness of the terms. Publication of these rows in a public dataset that is usable for text retrieval and ML sits against the site's ai-train=no / ai-input=no reservation; the owner has not obtained NamibLII's permission or a bulk-access arrangement. Anyone reusing the NamibLII rows must make their own determination.

Row semantics for NamibLII rows.

source_url of the Communications Act 2009 row had a ?ref=broadbandbreakfast.com tracking parameter, now removed (its archive_url/original_url in metadata_json keep the exact archived URL).
retrieved_at is the time of the archived snapshot that supplied the text (from archive_snapshot_timestamp), not when this pack was built and not when the live page was visited. Baseline rows keep their original retrieved_at.
Titles had site suffixes ("| NamibLII", "Laws.Africa" etc.) stripped; title_note in metadata records this.
Article rows are an exact partition of the law text at numbered article/section markers; laws with no clean markers have no article rows (435 NamibLII laws, mostly from gazette-PDF snapshots).
All NamibLII rows are tagged language=en. Some gazette-PDF rows are bilingual (Afrikaans/English) scans with OCR noise, so the language tag and text quality are imperfect there.
Duplicates: 24 rows that were URL-form aliases (eng-DATE / -0 vs eng@DATE) of the same work with identical text were dropped (23 byte-identical, plus the Deeds Registries Act 2015 eng-2015-12-31 row, 98 articles, whose text is identical to the eng@ row after whitespace/case normalisation) (listed in provenance/changes.json alias_duplicates_dropped).
Shared-text Government Notices (flagged, not dropped): 33 groups (95 rows) are distinct Government Notices / regulations that share one identical gazette-PDF text. Their titles are per notice but text is the whole shared gazette text, so those rows can carry several instruments. They carry shared_text_group in metadata_json; the groups are listed in provenance/changes.json.
Works with multiple versions: four works appear in more than one dated version; each version is a separate row.
27 long or keyword-flagged Acts that the automatic quality filter had held (size over about 200k chars, "Bills", "Forms", "Press", "High Court" in the title, etc.) were restored by owner decision; they are real instruments and are listed in provenance/changes.json owner_approved_restores.
Still held out (3): Communications Act, 2009 (/akn/na/act/2009/8/eng-2009-11-16) - superseded_older_version; Electoral Act, 2014 (/akn/na/act/2014/5/eng-2014-10-08) - superseded_older_version; Namibliii (/akn/na/act/p/1979/ag3) - title_unresolved_Namibliii.

Changes to baseline rows (metadata only; no text, title or row removed)

Both parquet files rewritten with ZSTD.
json_path on 364 rows was a box path and is now null.
Article ids that collided within a law (1,463 rows, 700 ids, 86 laws) now carry -2, -3 ... suffixes in file order; the original id is kept as id_original in the article metadata_json. No row dropped. A full old-to-new map (rekey_map_na_20261030.csv) accompanies this revision.
171 baseline laws declared article_count >= 2 but have 0 article rows in the articles table (the baseline has the same gap). article_count is now 0 for them; the former value is article_count_original in metadata_json, with an explanatory article_count_note (the articles table has no rows for the law; embedded slices are low quality and were not promoted). Their low-quality embedded slices in metadata_json.documents were not promoted to article rows.
Known baseline issue left as is: 303 of the 530 baseline titles are generic ("REGULATIONS", "GOVERNMENT GAZETTE", "REPUBLIC OF NAMIBIA", ...).

Per-row provenance (source URL, snapshot, dropped candidates and reasons, held/restored lists) is in provenance/changes.json; file hashes in MANIFEST.json; code in scrapers/.