Data provenance and release license
This file records the conservative release position; it is not legal advice.
| Dataset | Local evidence | License/status used for this release |
|---|---|---|
| Muharaf public line images | selection manifest records DOI 10.5281/zenodo.11492215 | CC BY-NC-SA 2.0 |
| RASAM | dataset repository includes LICENSE |
Apache-2.0 |
| TariMa | Hugging Face/GitHub source records | Apache-2.0 |
| Agapet | Zenodo record 15473122, DOI 10.5281/zenodo.15473122 | CC BY 4.0 |
| Omar Al-Saleh manuscripts | Hugging Face revision e6e8cd92661b3e208091bb86820f769494d00602 |
CC BY 4.0; gated access |
Why the model is CC BY-NC-SA 2.0
Muharaf is the most restrictive recorded training source. The release therefore adopts CC BY-NC-SA 2.0 as a conservative, non-commercial ShareAlike license rather than using Apache-2.0 or MIT for the weights.
Before any commercial release, obtain a dataset-by-dataset legal review or train a clean replacement using only data compatible with the intended commercial license.
Separate application content
The Athar retrieval library is not bundled into this model repository. Its texts have separate provenance and license records. In the current application, only one of four indexed book records has fully documented digital-text licensing; the UI exposes that limitation.