For scholarly review · 30 August 2026
Every āyah of the Qurʾān, the commentaries that speak on it, and — for 4,603 āyāt — the ḥadīth those commentators quote there, each link re-derived word for word from both sources before it was allowed onto a page.
All figures below were counted on the built artifact, not taken from the code that wrote it.
Open any āyah and the page answers, in this order: what the commentaries agree it means; which of them speak on it and in what words; and which ḥadīth those same commentators quote there — ranked, with the exact phrase of the āyah each one bears on.
Six phases, each one depending on the one before it and refusing to start until that one passed its gate.
Link, dedupe, cap at nine
Shingle-index every ḥadīth against every commentary cell; cap nine reports per āyah so no page becomes a corpus dump.
Re-derive every link from both sources
A run of twelve consecutive words must appear in both the ḥadīth text and the cited edition's commentary on that āyah — recomputed from the files as they stand, not trusted from the index that produced them.
Complete the missing language
No published Urdu exists for the Shīʿī collections in this project's sources. 827 renderings were written from the Arabic and are labelled as this project's own work, in those words, beside the Arabic so any reader can check them.
Render the layer and the reverse index
The layer under each āyah, and 5,538 pages that let a researcher walk the other way — from one report to every āyah a commentator cited it on.
Gate, archive, checksum, publish
Nine gates on the built pages, a script-free archival copy, SHA-256 over every file, and the rebuild instructions beside them.
Rank the reports and anchor each one
The nine reports on an āyah read together and ordered by what a reader most needs first, each anchored to the exact words of the āyah it bears on.
On Āyat al-Kursī the old order was mechanical — most editions first, then grade, then collection name. It is a defensible default and it is not a reading. The eight reports there are now ordered by what a reader most needs first, and the one that had sat last under the old rule now leads. Each carries an anchor naming the words it bears on, such as ٱللَّهُ لَآ إِلَٰهَ إِلَّا هُوَ — a phrase copied out of the āyah, which a gate re-derived against the Arabic before allowing it in.
Every claim on these pages is either a quotation re-derived from the corpus, or a line marked as this project's own. The checks below run as commands and refuse the build on failure.
| Check | What it proves | Result |
|---|---|---|
| Layer gate | Nine gates measured on the built pages — resolution, coverage, provenance, one list, three languages, links, idempotence, alignment | 36 pass · 0 fail |
| Verbatim | Every tafsīr and ḥadīth quotation re-checked character-for-character against the corpus | 234,319 · 0 wrong |
| Idempotence | The same input renders byte-identical output, including from a freshly built index | identical |
| Archive gates | No script, no external reference, no unbalanced tag in the permanent copy | 115 files · 0 fail |
| Checksums | SHA-256 over every archival file and site page, then re-verified | 251 · 0 changed |
| Corpus | 6,236 āyāt at the canonical Kūfan count; 197 source files byte-for-byte against the manifest | verified |
| Page audit | All 114 sūrah pages, structural and consistency | 0 failures |
Stated here rather than discovered by a reviewer.
| Gap | Extent | Why |
|---|---|---|
| The reason a report was cited | 236 of 22,188 | Present on 1%. The commentary states no reason in most cases and inventing one was refused. The pages therefore rank and anchor rather than explain. |
| The limit a commentary puts on a report | 10 | Same rule, and rarer still. |
| Independent reading of the renderings | 827 | Written by the same kind of model that reviewed them. Every one is labelled “not yet checked by a second reader” and stays so until a human scholar reads a sample. |
| Quotation budget | 3 works > 90% | Three modern works are reproduced at over 90% of their source text. This blocks public release, not scholarly review. |
| One Urdu rendering | 1 | Kitāb al-Ghayba #285 — its Arabic field in the corpus is a single character. Recorded, not hidden. |
| Arabic of seventeen reports | 17 | Corrupt in the source corpus. The archive declines to reproduce a damaged quotation and says so on the page; the translations are unaffected. |
In the order it should happen, and why that order.
Private links, six questions already on the register page: the balance of the selection, the similarities-only rule, sparse works, the disputed attribution of Tanwīr al-Miqbās, and the double-counting of translations. Nothing else on this list should move until at least one scholar has read a sample and said so in writing.
The one defect class no gate can catch is a rendering that is a faithful translation of the wrong report. Only a second reader finds that. Window N checks window N+1's work and never its own — the discipline already proven on the Qurʾān lemmas.
Cap each edition's share of its own source text across the corpus, quoting less per āyah rather than dropping a work. Two of the three may be exempt by their own licence — that must be read, not assumed.
CC BY-SA on the readings, MIT on the code, a Zenodo DOI so it can be cited, and an Internet Archive mirror. With the checksums and rebuild instructions already written, anyone in any century can reconstruct this and prove it is what it claims to be.
The 1% figure is the largest remaining gap in usefulness. A second pass over only those āyāt whose commentary does state a reason would raise it substantially without inventing anything.
The reading and the layer already carry Arabic, English and Urdu on every āyah. Turkish, Indonesian, French and Bengali are the largest unserved readerships; each needs a published translation source, not a rendering.