Programme · Policy

What Bayanat Labs publishes, and on what terms

How Bayanat Labs chooses, licenses and publishes open Arabic data: three release tiers, a licence matrix and the rubric we use to choose what to build.

By Bayanat Labs· Updated · اقرأ بالعربية

Documentation, not legal advice. This page states Bayanat Labs' own rules. It describes other platforms' terms only to explain why we follow them.

Bayanat Labs open-programme policy, version 1.0.1 · Reviewed by counsel, October 2026

Bayanat Labs publishes open Arabic data only where it can license it. Our commercial moat is rights clarity, so every release says what it contains and on what terms before anyone downloads it.

This page summarises the open-programme policy, version 1.0.1. The full policy in English and Arabic, its machine-readable files and a validator are in the open-policies repository.

The three release tiers

Tier What we publish Licence What it needs
Tier 1: open content Full text or audio, plus our annotations CC0 for manifests and metadata; CC BY 4.0 for our annotations; CC BY-SA 4.0 where derived from Wikipedia; a source_license field on every record An open licence, public-domain status or the contributor's consent
Tier 2: manifests plus our annotations IDs, URLs, timestamps, our labels and a one-command rebuild script; no verbatim content CC BY 4.0 on annotations, CC0 on IDs A named collection basis: an academic partner, written permission from the rights holder, or a recorded legal decision that permits it
Tier 3: partner-licensed Content released under a written permission from the rights holder As agreed with the partner A signed permission on file before collection

Licences we release under

We release under open licences only, and we keep what we release separate from what we may collect. Every record carries its own source_license field.

ID Content Licence tag
OUT-01 Manifests, IDs, timestamps, metadata cc0-1.0
OUT-02 Our annotations, transcripts of CC content, segmentations, labels cc-by-4.0
OUT-03 Structured reference data (gazetteer, lexicons, validators' test suites) cc-by-4.0 / cdla-permissive-2.0
OUT-04 Wikipedia-derived text cc-by-sa-4.0
OUT-05 Web-text packaging (crawl-derived text) odc-by
OUT-06 GPL-licensed inputs gpl-3.0
OUT-07 Verbatim-only inputs whose terms forbid modification other
OUT-08 ODbL-derived data odbl
OUT-09 CC BY media whose licence version is shown per item (audio clips) cc-by-3.0 / cc-by-4.0
OUT-10 Bayanat Labs-authored data and docs (synthetic data, written text, guidelines, templates, reports, policy text) cc-by-4.0
OUT-11 Corrections to non-commercial sets source licence
OUT-12 Partner-licensed content (Tier 3) as agreed

Code is Apache-2.0. On our own work we never use a non-commercial (NC) or no-derivatives (ND) licence, never Apache or MIT for data, and never a custom licence. Where sources mix, the dataset card says: "Underlying content is not licensed by Bayanat Labs; see per-record fields".

How we score ideas

Every candidate is scored from 1 to 5 on five criteria. Anything that fails the legal gate is dropped, whatever its score.

Criterion Weight
Impact (citations, press, developer use) 30%
Claim strength 20%
Ease 20%
Legal cleanliness (gate) 20%
Commercial pull 10%

Legal gate: red means not built; yellow means IDs, URLs and our own annotations only; green means full content may be released under an open licence. Where the legal light and the legal score differ, the more cautious of the two decides.

Suggested priority: score 75 or more with effort S or M gives P0; score 60 or more gives P1; score 45 or more gives P2; otherwise P3. These thresholds are adopted for v1.0, provisional; the coordinator decides each case.

Contact, corrections and takedowns

Write to hello@bayanatlabs.com. We remove items within 30 days of a takedown request, and within 24 hours for platform requests and personal-data reports. Changes to this policy are listed in the changelog in the repository.

Others may adopt the licence matrix under CC BY 4.0, with the attribution "Policy by Bayanat Labs".