Transparency

Coverage

We're explicit about what our current indexing covers. Three numbers matter — each measures something different:

Surah coverage: chapters with at least one Tabari passage

114 of 114 (100%)

Tabari passages (per ayah, canonical layer)

6,226

Curated topical ayat with coverage

119 of 119 (100.0%)

Full English translation of Tabari's own text (word-for-word, not a summary)

6,226 of 6,226 (100.0%)

Translated via Claude with a strict anti-softening protocol: abd → slave, rajm → stoning, no softening of violence or punishment. Isnad chains written out in full, Quran verses kept in Arabic. Tabari's full text was translated word-for-word from the Arabic into Dutch, and the English edition was rendered from that Dutch text.

All surahs have at least one Tabari passage. For a few short surahs (e.g. Al-Fil 105) the slice runs through transition text from a neighbouring section — those are flagged as "Inferred" to be honest about the mixed content.

Since filling the remaining gaps from Turath, every verse — including every ayah in our topical curation — has a canonical Tabari passage. Coverage is now 100%.

Our two text layers

Our tafseer text has two layers. The canonical layer — what you see in the app — is fetched per ayah from the quran.com API, with exact verse-linkage; the few verses quran.com didn't provide were filled from Turath (the same edition). Alongside it is an older, OCR-derived layer we are phasing out:

• Canonical — 6,226 passages (6,196 from quran.com + 30 from Turath, filling quran.com's remaining gaps): exact verse-linkage, no guessing. Covers all 6,236 verses of the Quran (100%) and is fully translated into English and Dutch. This is what the app shows.
• Legacy OCR layer, kept — 1,797 passages: OCR'd from the printed edition with fuzzy verse-linkage (verified or inferred).
• Legacy OCR layer, rejected — 1,838 passages: a language model verified the passage is actually about a different verse, due to OCR segmentation errors. These are fully filtered out of the app.

Methodology

The canonical Arabic text is fetched per ayah from the quran.com API (Dar al-Tarbiya wa-l-Turath edition: vols 1–16 following Mahmud Shakir's tahqiq, vols 17–24 the Halabi printing) and verified against app.turath.io (the same edition); the few verses quran.com didn't provide were filled directly from Turath. The Arabic was first translated word-for-word into Dutch by Claude under a strict anti-softening protocol. The English edition was then rendered from that validated Dutch text rather than directly from the Arabic — a pivot translation — with every segment passing an automated fidelity check against its Dutch source, and the Arabic of Quran verses, isnad chains and technical terms carried through unchanged. Code and migrations are open and reviewable.