The Geometric Encyclopedia of the Qurʾān
Build Plan and House Discipline
Working handout. Derived from the Abrogation entry, which was built end to end as the pilot: topic harvested, spine forged, entry authored, page rendered, defects found by looking at the page. Everything below is either a rule that held under that test or a defect that cost something and is now fenced.
0 · Standing
What this is. A reference work on the Qurʾān whose unit is the headword and whose method is structural. Each entry defines its term, excavates its root, reads the corpus on the corpus, exhibits the geometry the term carries, and fences what the term will not bear.
What this is not. It is not a revision, translation or condensation of any existing encyclopaedia. From Brill's Encyclopaedia of the Qurʾān we take two things and nothing else: the topic list, as a map of what a complete reference work must cover, and the entry schema, as a solved problem in reference-work architecture. The argument, the analysis, the selection of loci, the structural reading and the prose are ours and are written from the corpus outward. Where the two works reach the same conclusion that is convergence and it is recorded as convergence. Where the bibliography overlaps that is because the primary sources are the primary sources.
Grade of the base. The Qurʾān is carried at lā rayba fīhi grade and is the only seal-eligible register. Everything else in the entry, without exception, is witness.
1 · The governing law
Corpus-internal primacy. The trusted reading of any term is built from inside the corpus: the root across its whole attested range, the immediate context of each locus, and the relation of parallel and contrasting passages. That reading is the only one eligible to seal.
The Provenance Firewall. Every post-canonical interpretive corpus of every tradition is compromised-until-corroborated and never seal-eligible. This is not a judgement on any school. It is a standing prior justified by dated gap, documented capture and sectarian contest, held as a prior and never inflated to a finding. Classical commentary, tradition-witness and mystical readings enter as witnesses, each flagged, each read against the internal reading, each marked contrastive where it diverges.
Consensus carries nothing, in either direction. That a reading is standard is not evidence for it. That a reading is ours is not evidence for it either.
No fabrication. No invented citation, date, page number, attestation, count or frequency. A number that cannot be verified is marked unverified. A hadith of uncertain attribution is marked. An uncertain root is omitted rather than guessed.
Honorifics are components of the name. Allah ﷻ, the Prophet Muhammad ﷺ, other prophets AS, preserved through every gloss, index row and summary.
Silence. The verification apparatus runs in full and stays off the page. One verdict shows, the geometric seal, and nothing else. Where a reading does not close, the entry says so in words and shows no seal.
2 · The entry architecture
Ten movements plus apparatus. Order is fixed. Movements 1, 2 and 5 are mandatory; the rest fire where the term warrants. This is the schema the Abrogation entry established and it should be treated as the template until a term breaks it.
1 · Head and vocalisation. # Term (translit, عربي). The head sets the guide word.
2 · Definition. First paragraph, unindented. Defines before it discusses, and carries the entry's thesis in one sentence so a reader who stops here has the claim.
3 · The literature. An independent survey of the scholarship on the term, run by register and not by chronology. This is the first substantial part of the entry and it is written from our own reading, never condensed from another encyclopaedia's version of the same survey; another work's entry is used to learn what the field considers in scope and is then closed.
Six registers carry most terms. The indigenous genre, its major works and, where one exists, its quantitative history. Legal theory, including the vocabulary the jurists built alongside the term and often at its expense. Theology, including the polemical setting that provoked the doctrine, since most doctrines of this kind were formulated against an objection from outside. Modern Muslim scholarship, which for most contested terms runs in a single direction and that direction is itself a finding. Western scholarship, with the disagreements named rather than averaged. And the contemporary register, political or manuscript or both, where the term does live work now.
Close the movement on what the literature has not asked. That paragraph is the bridge into the analysis and it earns the entry its reason to exist. If it cannot be written honestly, the entry is a summary and should be reconsidered.
Two disciplines bind here and neither is optional. Verify before citing. A literature review is the single place in an entry where invented citations appear, and memory is not a source; check the author, title, date and claim. Look for the quantitative history. Where a doctrine's own literature counted something, the shape of that count over time is usually the most informative datum available, and it is normally reported as scholarly disagreement rather than read as a symptom.
4 · Root excavation. The opening analytic move and usually the strongest. Recover the root, run the concreteness hierarchy, somatic then environmental then kinetic and only then abstract. Cross the cognate field. Run the bivalence check: where a root carries two senses opposite in what they do to their object, say so, because that is normally the explanation for why the term's later doctrine never stabilised. This was the single most productive paragraph in the pilot.
5 · The qurʾānic base. Every locus, read on its own terms. Note what the verse actually says as against what the doctrine built on it says. Watch for disjunctions, which are routinely read away, and for the rhetorical situation, since a verse answering an accusation is not a technical definition.
6 · The structural reading. The term's geometry, in plain human language, naming no machinery. What stands, what moves, how the two bind, and which of the three the term makes move. Close on the one shown seal where the reading closes; state the break or the gap in words where it does not.
7 · Discriminator and worked cases. Where the term covers two mechanically distinct operations, give the test that separates them and work at least one case each way from the corpus.
8 · Tradition as witness. The received taxonomy is reported and mapped, never adopted. Mapping it usually produces the entry's sharpest result, because a category the architecture cannot certify shows up here.
9 · Fences. What the term will not carry, each inflation named with the verse or mechanism that refuses it. Three to five is the working range.
10 · Scope. The domain guard. Name the adjacent question the term is regularly confused with and separate them.
Apparatus. Signature @ Name. Bibliography with Primary and Secondary runs, hanging indent. Cross-references woven inline: (q.v.) is a pointer meaning the word just read has its own entry, a small-cap see-reference is a delegation meaning this matter is treated there and not here. The two must never collapse, or the work becomes overlapping essays with an index attached.
3 · Tiers and budget
Entry length distributes with a violent right skew and a flat words-per-entry plan cannot produce that shape. Declare the tier at commission, not at draft.
| Tier | Movements | Words | Share |
|---|---|---|---|
| Redirect stub | head plus one line | under 30 | ~30% of headwords |
| Standard | 1-2, 4-5, 9-10, apparatus | 900 to 2,000 | ~55% |
| Major | all ten, literature review included | 3,500 to 8,000 | ~15% |
The pilot ran about 4,100 words at major tier, of which the literature review is roughly a third. At 500 substantive entries plus stubs the corpus lands near 1.1 to 1.4 million words, which is roughly 1,700 to 2,100 pages in the edition, two volumes. Budget the apparatus, indices and author list at about a fifth of total extent; that ratio is what makes the rest reachable and it is the number a first edition most often fails to plan for.
Fix mean length per tier at the first fifty entries and audit every fifty against it. Scope drift is visible in every multi-volume set built over years, entries growing longer and fewer as the editors go on. Catch it at fifty, not at four hundred.
4 · The filing law
The sort key is fixed before the first entry is written, because re-filing a finished corpus is the one edit that cannot be automated safely.
Strip a leading article and any bracket, decompose to NFKD, drop combining marks, drop hamza and the modifier letters, drop non-alphanumerics, lowercase. So al-Baqara files at B, ʿĀd files at A, Ṣāliḥ files at S.
Entries are stored sorted. This is not cosmetic. The renderer emits a letter opener whenever the initial changes, and an opener that appears out of sequence breaks multi-column flow and silently discards everything after it. An unsorted build dropped 97 of 125 entries in one volume of the restoration project and still exited zero. Sort at the source, not at the build.
5 · The typographic contract
The edition is a dense two-column reference page and its virtue is a uniform grid. Anything that opens the leading of one paragraph and not its neighbours breaks the page.
The Arabic rule, tested. Vocalised Arabic with full tashkīl is taller than the body line box, and there are exactly two ways to handle it. The wrong way is to open the leading of any paragraph containing Arabic, which stops the collision and destroys the grid. The right way is:
- Inline Arabic is unvocalised, set at 0.98em, and inherits the body line box. It then sits inside the grid and the leading stays uniform across the page.
- Vocalisation is reserved for display citations set on their own line, where the extra height costs nothing because the line is theirs.
- Transliteration carries the running prose. Arabic script inline is for the term itself and short phrases, not for extended quotation.
| Wrong | Right | |
|---|---|---|
| Arabic size | 1.06em to 1.12em | 0.98em |
| Line box | paragraph override, 1.6 | inherits body, 1.235 |
| Inline vocalisation | full tashkīl | stripped |
| Result | grid breaks per paragraph | uniform page |
Fonts, pre-flight. EB Garamond for the serif, Amiri for Arabic, Noto Serif Bengali for Bengali, Noto Sans Math for the seal glyph. Variable faces are instanced before the renderer consumes them and the cache is refreshed after placement. Name the math face explicitly in the serif stack rather than relying on fallback. Verify the seal at 300 dpi before calling it tofu; at page resolution ⟀ and a stray arrow look identical, and I have made that mistake once already.
No em-dashes anywhere. Unicode only, never LaTeX. The ﷺ ligature is one glyph.
6 · The render pipeline
Split into volumes. A full corpus of this size will not render in one pass and is killed by the memory reaper without an error message. Roughly 500 pages per pass is comfortable. Split on letter-range boundaries and merge.
Verify by counting headwords in the output. Never by exit status. A build that drops most of a volume exits zero and writes a valid PDF. Extract the text, probe for each head, compare against the source count. This is the single most valuable gate in the pipeline and it costs seconds.
Treat kerning splits as artifacts, not losses. Extraction turns Tyrant into T yrant. Probe a substring, not the whole head, or the gate will report phantom losses.
Letter openers must be monotonic and unique. Guaranteed by the filing law above.
7 · Topic harvest
The restored Brill corpus is the trigger list, not the content source. It carries 669 headwords in filed order, which is a complete map of what a reference work on this subject is expected to cover.
Working policy. Harvest the headword. Read the entry once as a topic brief, to learn which loci and which controversies the field considers in scope. Then close it and build ours from the corpus. The overlap should be the subject and the primary sources; if the argument overlaps, something has gone wrong.
Not every headword earns an entry, and some of ours will have no counterpart there. Selection is ours. Aim for 500 substantive entries plus a generous stub layer, since under-stubbing is the standard failure of a first edition. Budget roughly one redirect for every two substantive entries.
8 · Per-entry workflow
- Resolve the register from git. Never from memory; version recall is routinely stale.
- Anti-duplication. Search the live register for a resident or nearest coordinate. Apply by reference where one exists, extend under supersession where near-identical, forge new only where the register does not hold it.
- Forge the spine where the term needs one. Not every entry does. Where the structural reading is genuinely new, the coordinate is forged first and the entry then reads from it, which is the order the pilot used.
- Survey the literature independently, verifying each work rather than recalling it, and close the survey on the unasked question.
- Draft to the ten movements.
- Audit. Fences present, honest typing, no fabrication, corpus-internal reading seal-eligible and everything else witness-flagged.
- Render and look at the page. The pilot's three real defects, Arabic collision, grid break and a suspected tofu, were all invisible in the source and obvious on the page.
9 · QA gates
Run before an entry is called done.
- Definition states the thesis in one sentence.
- Every qurʾānic locus verified present and quoted accurately.
- No invented citation, date, count or attestation anywhere. Every work named in the literature review verified, not recalled: author, title, date and the claim attributed to it.
- Honorifics on every prophetic name in body, bibliography and index rows.
- One seal at most, and none where the reading did not close.
- Fences name their mechanism, not just their conclusion.
(q.v.)and see-references not collapsed.- Filed position correct under the sort key.
- Renders inside the grid with no paragraph-level leading break.
- A reading of ours that departs from the tradition is flagged as a reading.
- The literature review closes on what the field has not asked, and that gap is the one the entry fills.
10 · Publication path
Entries accumulate as individual source files in the edition format, so no conversion is needed at assembly. At 500 substantive entries, assemble, build the apparatus, index and author list, and render. Single volume if the extent permits, otherwise split on the natural letter boundary.
The pilot file EQG_Abrogation.md is the reference implementation of the schema and the format. Copy its shape.
Inshaallah.