HUDA-Net Sources is the private, version-controlled source repository used to build the HUDA-Net bilingual question-answering, semantic retrieval, and text-ranking system for Hajj and Umrah jurisprudence.

This repository preserves the original book datasets, cleaned source files, provenance metadata, and integrity manifests required to reproduce and audit the HUDA-Net corpus.

Access status: Private research dataset
Primary language: Arabic
Secondary language: English
Redistribution: Subject to the rights and licenses of the underlying source materials

Dataset Overview

Item	Value
Source collections	23
Original book source files	23
Cleaned source files	34
Total source data files	57
Approximate source size	10.22 MB
Structured corpus records	4,138
Languages	Arabic and English
Primary domain	Hajj and Umrah jurisprudence
Integrity verification	SHA-256
Repository visibility	Private

The source collections in this repository are used to construct the unified HUDA-Net corpus containing approximately 4,138 structured records.

Repository Purpose

This repository serves as the authoritative source layer for HUDA-Net.

It is intended to support:

Rebuilding the HUDA-Net corpus from its original source files.
Reproducing cleaning and normalization operations.
Auditing source provenance and file integrity.
Developing Arabic and bilingual semantic retrieval systems.
Conducting question-answering and text-ranking experiments.
Supporting academic reproducibility.
Preserving the source materials independently from Kaggle.
Verifying transferred files using SHA-256 checksums.

Runtime indexes, embeddings, trained academic artifacts, benchmark results, and application code are maintained separately.

Source Collections

The repository contains source material associated with the following 23 collections:

#	Canonical ID	Source collection
1	`abdullah_bin_muhammad_manasik`	Manasik al-Hajj by Abdullah bin Muhammad
2	`dalil_al_talib`	Dalil al-Talib
3	`permanent_committee_fatwas`	Fatwas of the Permanent Committee
4	`fawzan_hajj_part1`	Hajj Lessons and Fatwas by al-Fawzan, Part 1
5	`fawzan_hajj_part2`	Hajj Lessons and Fatwas by al-Fawzan, Part 2
6	`al_furu`	Al-Furu by Ibn Muflih
7	`al_irshad`	Al-Irshad ila Sabil al-Rashad
8	`jami_masail_ahmad_1`	Jami Masa'il Imam Ahmad, Hajj Part 1
9	`jami_masail_ahmad_2`	Jami Masa'il Imam Ahmad, Hajj Part 2
10	`masail_imam_ahmad`	Masa'il Imam Ahmad
11	`masail_kawsaj`	Masa'il al-Kawsaj
12	`mukhtasar_al_khiraqi`	Mukhtasar al-Khiraqi
13	`al_hidayah`	Al-Hidayah ala Madhhab Imam Ahmad
14	`al_ifsah`	Al-Ifsah by Ibn Hubayrah
15	`ibn_taymiyyah_manasik`	Manasik by Ibn Taymiyyah
16	`al_iqna`	Al-Iqna
17	`al_kafi`	Al-Kafi by Ibn Qudamah
18	`kashshaf_al_qina`	Kashshaf al-Qina
19	`al_mughni`	Al-Mughni by Ibn Qudamah
20	`rawdat_al_murbi`	Rawdat al-Murbi
21	`umdat_al_fiqh`	Umdat al-Fiqh
22	`uthaymeen_manasik`	Hajj and Umrah Guidance by Ibn Uthaymeen
23	`zad_al_musafir`	Zad al-Musafir

Repository Contents

The repository contains two principal source layers.

Original Book Sources

The original book files preserve the datasets as they were supplied to the HUDA-Net development environment.

Each collection is assigned a canonical HUDA-Net identifier to support consistent provenance tracking across preprocessing, runtime construction, and academic evaluation.

Cleaned Sources

The cleaned source layer contains 34 processed files prepared for corpus construction.

Cleaning and preparation may include:

Column-name normalization.
Arabic text normalization.
Removal of empty or unusable records.
Consolidation of question and answer fields.
Standardization of source identifiers.
Schema normalization.
Preparation for bilingual processing.
Preparation for semantic indexing.
Preservation of source attribution.

Processing Pipeline

Original book datasets
        ↓
Source discovery and validation
        ↓
Cleaning and schema normalization
        ↓
Text normalization
        ↓
Source attribution
        ↓
Unified bilingual corpus construction
        ↓
Semantic embeddings and retrieval indexes
        ↓
HUDA-Net runtime
        ↓
Academic evaluation and benchmarking

This repository contains the source and cleaned-data layers of the pipeline.

Generated embeddings, indexes, model artifacts, benchmark outputs, and application files are stored in separate HUDA-Net repositories.

Data Fields

The exact fields may vary between original source collections.

The normalized corpus may contain fields conceptually corresponding to:

Field	Description
`question_ar`	Arabic question
`answer_ar`	Arabic answer or source passage
`question_en`	English question or translated field
`answer_en`	English answer or translated field
`book_id`	Canonical HUDA-Net source identifier
`book_name`	Source book or collection name
`source_file`	Original source filename
`source_path`	Provenance path
`language`	Record language
`record_id`	Internal record identifier
`text`	Normalized retrieval text
`metadata`	Additional provenance or processing information

Not every original file necessarily contains all these fields. The unified corpus construction pipeline standardizes the records before runtime indexing.

Integrity and Provenance

Every uploaded source file is recorded in:

_hudanet_metadata/source_manifest_sha256.json

The manifest records:

Repository-relative file path.
File size in bytes.
Source role.
Canonical book identifier.
SHA-256 checksum.

The transfer summary is stored in:

_hudanet_metadata/transfer_summary.json

These metadata files make it possible to verify that the Hugging Face copy matches the files transferred from the original Kaggle environment.

Verification Summary

Verification item	Status
All 23 source collections identified	Verified
Original source files transferred	Verified
Cleaned files transferred	Verified
Runtime assets excluded from this repository	Verified
Academic assets excluded from this repository	Verified
SHA-256 manifest generated	Verified
Source file paths preserved	Verified
Kaggle files deleted during transfer	No

Intended Uses

Authorized uses include:

Arabic question answering.
Bilingual question answering.
Semantic search.
Dense retrieval.
Hybrid retrieval.
Passage ranking.
Text ranking.
Sentence similarity.
Retrieval-augmented generation.
Corpus cleaning and normalization research.
Arabic natural-language processing.
Islamic studies information retrieval.
Hajj and Umrah jurisprudence retrieval.
Academic reproducibility experiments.
Source provenance and integrity auditing.

Out-of-Scope Uses

This repository is not intended to:

Replace qualified religious scholarship.
Independently issue religious rulings.
Serve as a definitive fatwa authority without source verification.
Remove attribution from original sources.
Support unauthorized redistribution of the underlying works.
Train unrelated commercial systems without authorization.
Present translated text as more authoritative than the Arabic source.
Provide legal, medical, or safety-critical advice.

HUDA-Net retrieves and organizes information from its corpus. Outputs should be checked against the cited Arabic source passages.

Limitations

The dataset has several important limitations:

Source formatting differs across collections.
Some materials were supplied as spreadsheet datasets rather than original digital editions.
Arabic orthography, punctuation, and spacing may vary.
Source files may use different schemas and column names.
English fields may include normalized or generated translations.
Translation quality may vary between records.
Retrieval quality depends on corpus coverage and preprocessing.
Source coverage is focused on selected Hajj and Umrah jurisprudence materials.
The presence of content does not establish ownership or redistribution rights.
Religious questions may require interpretation by qualified scholars.
Original source errors may remain present in cleaned or normalized records.

Languages

Arabic

Arabic is the primary language of the source texts and remains the authoritative language for source interpretation.

English

English fields may be used for:

Bilingual retrieval.
Translated questions.
Translated answers.
Cross-language evaluation.
Application-facing labels.
Academic experiments.

Where an English field differs from the original Arabic source, the Arabic text should be treated as authoritative.

Data Quality

The source pipeline applies multiple quality-control stages, including:

File discovery.
Schema inspection.
Required-field validation.
Empty-record detection.
Duplicate detection.
Source-name normalization.
Text normalization.
Book-level attribution.
Record-level provenance.
Corpus coverage checks.
File-integrity verification.

Cleaned files are preserved alongside original source files so that transformations can be audited.

Privacy

The dataset is composed primarily of book-derived and research corpus material.

It is not designed to contain private personal information. However, the repository owner should review source files before any future public release.

This repository remains private while source rights, redistribution permissions, and publication conditions are being evaluated.

Licensing and Rights

No general open-data license is asserted for the complete repository at this stage.

The HUDA-Net project maintains the compilation structure, preprocessing workflow, manifests, canonical identifiers, and project-specific metadata.

The underlying books and source texts may remain subject to their respective:

Copyright terms.
Publication rights.
Distribution restrictions.
Website terms.
Author or publisher conditions.
Jurisdiction-specific protections.

Access to this repository does not automatically grant permission to republish, redistribute, sublicense, or commercially exploit the underlying works.

A license should not be assigned until the rights of the source materials have been reviewed.

Repository Separation

HUDA-Net uses separate repositories to keep each system layer clean and auditable.

Repository	Purpose
`dakheel/hudanet-api`	Application code, interface, and live HUDA-Net Space
`dakheel/hudanet-sources`	Original books, cleaned sources, and provenance metadata
`dakheel/hudanet-runtime`	Unified runtime corpus, embeddings, indexes, and retrieval assets
`dakheel/hudanet-academic`	Training, validation, testing, models, and academic reports
`dakheel/hudanet-kaggle-archive`	Complete private archive of the original Kaggle inputs

This separation prevents generated runtime files from being mistaken for original sources.

Relationship to HUDA-Net Runtime

This repository is not loaded directly by ordinary HUDA-Net users during daily operation.

The production application primarily relies on the separately built runtime repository.

The source repository is used when:

Rebuilding the corpus.
Updating source materials.
Repeating preprocessing.
Auditing provenance.
Reproducing academic experiments.
Verifying the relationship between source and runtime records.

Version Information

Field	Value
Initial Hugging Face migration	July 2026
Source collections	23
Original book source files	23
Cleaned files	34
Total source data files	57
Structured corpus records	4,138
Integrity algorithm	SHA-256
Repository visibility	Private
Primary project	HUDA-Net

Reproducibility

To reproduce the corpus-building process, an authorized researcher should:

Download this private source repository.
Verify source files against the SHA-256 manifest.
Run the HUDA-Net source discovery process.
Apply the documented cleaning and normalization pipeline.
Construct the unified bilingual records.
Generate retrieval texts and metadata.
Build semantic embeddings and indexes.
Publish the resulting runtime assets to the runtime repository.
Run the academic benchmark and evaluation pipeline.
Compare the resulting manifests and metrics with the recorded project outputs.

Citation

For authorized academic use, cite the dataset as:

@dataset{hudanet_sources_2026,
  title     = {HUDA-Net Sources: Hajj and Umrah Jurisprudence Corpus},
  author    = {{HUDA-Net Project}},
  year      = {2026},
  publisher = {Hugging Face},
  note      = {Private Arabic-English research dataset}
}

Contact

Questions regarding access, provenance, corrections, licensing, or authorized research use should be directed to the HUDA-Net project owner through the Hugging Face repository.

Status

Migration status: Complete
Source integrity: Verified
Books identified: 23 of 23
Cleaned files: 34
Runtime files included: No
Academic files included: No
SHA-256 manifest: Available

Downloads last month: 8

Total file size:

10.7 MB