Data Sources & Credits

▪ Verbatim preservation — Hadith text is stored exactly as received from the source database. No paraphrasing, summarising, or reformatting of the original Arabic or translation.

▪ Non-destructive versioning — Original source files are archived in data/raw_baseline/. Any processing scripts produce new files, never mutating the originals.

▪ Open-source pipeline — All import and processing scripts are available in the project repository for independent audit.

▪ Grade transparency — Hadith grading (Sahih, Hasan, Da'if) is displayed exactly as provided by the source database. No editorial grading is applied.

▪ Translation disclaimer — English and Indonesian translations are those provided by the source databases. For scholarly use, always verify against authoritative printed editions.

Many collections on this site quote the same hadith. Riyad as-Salihin, Bulugh al-Maram and Mishkat al-Masabih are compilations: nearly every hadith in them was selected from an earlier source such as Sahih al-Bukhari, Sahih Muslim or the Sunan. This page documents how each collection's hadiths are linked back to those originals, so a reader can move from a compilation entry to the full chain of narration and every translation we hold.

The mapping is the deliverable. Once a hadith is identified as the same hadith as one in Bukhari, everything attached to that original — sanad, narrators, jarh wa ta'dil, grading, translations — becomes reachable by a simple lookup. We therefore publish a thin, auditable set of links rather than a merged corpus.

How a link is established

Each candidate is scored by evidence, and the evidence is stored alongside the link so the basis can be audited later:

1. Normalise the Arabic. Diacritics, hamza forms, alef/teh-marbuta variants, punctuation and invisible directional marks are folded. Indo-Persian orthography (ک ی ہ) is folded to Arabic (ك ي ه) — without this, a genuine match can score zero.
2. Strip editorial apparatus. Compilers append provenance notes such as ((متفق عليه)) or رواه أبو داود (775). These are removed from the text being matched and parsed separately as a provenance hint.
3. Rare-phrase fingerprinting. Five-word sequences that are rare across the whole corpus are the fingerprints; common phrases such as صلى الله عليه وسلم carry no signal and are discarded automatically.
4. Score and rank, then guard. Candidates are ranked by the number of fingerprints found, not by a percentage — a short accidental overlap must not outrank a long genuine one. A minimum evidence floor is required before anything is accepted.
5. Short texts get a different method. Some collections are maxims of two or three words (الحرب خدعة). Fingerprinting cannot work there, so the content phrase is searched verbatim instead, and coverage for that collection is correspondingly lower.

Translations are attached by resolving a source hadith to its own Arabic edition first, then reading every language at that resolved position. This is necessary because language editions of the same collection are not number-aligned: an Urdu edition can hold a different hadith at the same number as the Arabic edition. Where the Arabic cannot be resolved, no translation is attached.

▪ Duplicate source reconciled — Dehlawi 40 (Fawaz `ara-dehlawi`) and Shah Waliullah 40 (AhmedBaset `forties/shahwaliullah40`): Identical Arabic in all 40/40 rows; metadata author is “الشاه ولي الدين الدهلوي”. Same book, two files. Counted once.

What is not linked, and why

Coverage varies because the collections themselves differ in kind, not because effort was uneven:

▪ Compilations (Riyad, Mishkat, Bulugh) quote sources near-verbatim, so coverage is high. Independent musnads (Ibn Hibban, Ibn Khuzaimah, Mustadrak, Daruquthni) narrate through their own chains with their own wording, so the matn can differ even where the hadith is the same.

▪ Bulugh al-Maram contains variants and cross-references (“this is a portion of the long hadith mentioned earlier”) rather than standalone hadiths. Where it cites a hadith by number, that number is recorded as unverified provenance unless the cited text actually contains the matn — those are counted separately and are not presented as links.

▪ Dehlawi 40 is a list of maxims rather than full matn, so it will always sit lowest.

▪ Rows left unlinked are retained in the dataset with an explicit unmapped status and are not hidden.

▪ Known source gaps. Some anchor files we hold are partial. Sunan Daruquthni is present as a 236-row extract (hadith numbers 1–239, with gaps) while the full work numbers some 4,790 entries; Musnad Ahmad's enrichment source covers only a 1,374-row fragment. Coverage percentages for those collections describe the downloaded portion only, and will rise when complete files are ingested. They are published as measured rather than extrapolated.

For scholars, translators and reviewers

This mapping is offered as an aid to study, not as religious authority. It is a set of proposed correspondences between collections, produced by textual comparison, and it is published so that it can be checked.

▪ Every link carries a confidence tier (confirmed / probable / review), the evidence count behind it, and the method used.

▪ Links at probable and review are the ones most worth your attention. Collections reporting lower confirmed shares (Al-Adab al-Mufrad, Mustadrak, Daruquthni, Musnad Syafii) are the suggested starting point.

▪ The full machine-readable mapping is available as data/mapping-stats.json with per-collection detail, and the per-collection link files are in the project repository.

▪ Corrections and disagreements are welcome and can be applied per link without re-running the whole pipeline — the mapping is additive and auditable by design.