Nine Verified Openings

Each searched against the literature, each buildable from the corpus already on disk, each with the work that occupies the neighbouring ground named.

Verified 17 August 2026 Fully open 7 Narrow opening 2 Buildable today 9

How these were found. The first list of 27 was built from broad, obvious questions, and eight of the fourteen checked turned out occupied — by corpora larger than this one and tooling already public. Searching wider would have produced more of the same.

These nine come from searching narrower: formally marked features that the large projects have not catalogued, and questions about method rather than about findings. Seven returned nothing comparable at all.

Every one is buildable from what is already on disk. Where a topic needs a corpus layer not yet extracted, the extraction is named and costed.

Fully open

Twenty-one defects: auditing in humanities computing

nothing comparable found

Searched: Reproducibility and audit practice in humanities computing. Found methodological advocacy; found no published defect log from a working project
Contribution: Twenty-one real defects, the split between machine-caught and human-caught, and the capture–recapture argument that the search is nowhere near finished
You already have: The register, the audit code, the dates
Build: None. It is written
Venue: Journal of Digital Islamicate Research; DSH

Counting: root, lemma and sense

field “in its infancy”

Searched: Quantitative Arabic linguistics and word-frequency method
Sharpened by searching: Published counts for yawm now include 365, 92 and 475 — three figures from three counting decisions. The disagreement is the paper
You already have: The three-level engine, the sense gate, the worked failures
Build: Two weeks of writing
Venue: JDIR; DSH

A sense-disambiguated layer for the Qurʾanic lexicon

no such layer published

Searched: Qurʾanic datasets and annotation layers. MASAQ and the 2025 treebank supply morphology and syntax; ontologies supply topic. None supplies sense
You already have: The declaration format, the exhaustiveness gate, worked examples for barr and ākhir
Build: About a year of annotation. The long project
Venue: LREC; JOHD; JDIR

Computational detection of iltifāt

no computational study found

The feature: Iltifāt is the Qurʾan’s unexpected shift of grammatical person — from third to second, singular to plural — while still referring to the same subject. It is central to its style and far more frequent than in pre-Islamic poetry
Searched: Abdel Haleem’s Grammatical Shift for Rhetorical Purposes (BSOAS) is the standard study and is qualitative. Searching for computational or automated detection returned nothing
Why it is open: It is a formally detectable feature — a change in person or number tag across adjacent segments referring to one referent — that nobody appears to have extracted at scale
You already have: 37,317 person and number tags and 24,681 pronoun features, verified present in the cached raw morphology
Build: Re-parse the raw file to keep person/number alongside root and lemma — half a day. Then the shift detector
Venue: JDIR; Journal of Qurʾanic Studies

A complete catalogue of the oaths (qasam)

not found computationally

The feature: Oath formulas — wa- plus a sworn-by noun, the la- emphatic, the response clause — are formally marked and cluster heavily in the early sūrahs
Searched: Oaths are treated in the literature on Qurʾanic speech genres, qualitatively. Computational work returned only audio — recitation recognition and reciter identification. No textual catalogue found
Why it is open: Formally marked, exhaustively extractable, and it connects to chronology: oath density is a candidate phase marker alongside verse length
You already have: Particle tags, the morphology, revelation-order metadata
Build: ~2 weeks
Venue: JQS; Journal of Arabic Literature

Who is speaking: speaker attribution across the text

not found for text

The question: The Qurʾan shifts speaker constantly — divine first person, the command qul, angels, prophets, opponents quoted and answered. Which verses are attributable to which speaking voice, and how is that distributed?
Searched: All computational “Qurʾan speech” work found is audio: recitation recognition, reciter identification, tajwīd checking. Textual speaker attribution did not appear
Honest risk: Attribution needs human judgement in disputed cases. Publish the clear cases as data and the disputed ones as a list — the disputed list is itself useful
You already have: Person tags, the qul occurrences, the morphology
Build: ~2 months, most of it annotation
Venue: JDIR; JQS

Which words did the tradition find difficult?

no such study located

The question: Measure how much commentary each word attracts across 21 tafsīrs, then set that against the classical gharīb al-Qurʾān lists. Did the lexicographers single out the same words the commentators actually laboured over?
Searched: Tafsīr text mining, lexical difficulty, and gharīb. The 80-commentary corpus paper (JDIR 2025) examined metaphor, not lexical difficulty. No study correlating commentary volume with word difficulty was located; the search returned it as an “emerging area”
Why it is open: It requires crossing three layers at once — morphology, commentary, and a classical genre — and it is unglamorous. That combination is why it has been left
You already have: 21 āyah-aligned tafsīrs, the morphology, the 399 hapax roots
Build: ~1 month for the measurement; the gharīb comparison is reading, not code
Venue: Journal of Qurʾanic Studies; JDIR

Narrow opening inside occupied ground

The neighbouring work exists and must be read first. The specific angle named appears unclaimed, but each of these could still collapse on closer reading.

Ring composition tested against a null model

falsification unclaimed

Occupied by: Cuypers, Farrin and Robinson qualitatively; a genetic-algorithm study of Sūrat al-Baqara using the QuranTree.jl toolkit
The opening: All of that identifies structure. Sinai’s published criticism is that such analyses are subjective and never tested against chance. A null model asking whether the symmetry exceeds what shuffled text produces was not found
Discipline: If the answer is “not better than chance”, publish that. The negative result is the contribution
Build: ~2 months

Every verse ending, extracted and released

data unclaimed

Occupied by: Klar’s catalogue of sajʿ techniques (Oxford); Fourier-based rhythm analysis; a visualisation finding that ~70% of verse endings move from short to long vowels
The opening: Those are arguments and summaries. A complete, reproducible extraction of all 6,236 verse endings, tagged by rhyme class, mapped to phase and released as data, was not found
Build: ~1 month

Order of writing

#	Paper	Build	Venue
1	Twenty-one defects	written	JDIR
2	Counting: root, lemma, sense	2 weeks	JDIR / DSH
3	Iltifāt detected computationally	~1 month	JDIR / JQS
4	The oaths catalogued	~2 weeks	JQS / JAL
5	Verse endings released as data	~1 month	JOHD / JAL
6	Which words the tradition found difficult	~1 month	JQS / JDIR
7	Ring composition vs a null model	~2 months	JAL / JQS
8	Speaker attribution	~2 months	JDIR
9	The sense layer	~1 year	LREC / JOHD

Three standing conditions.

Re-search immediately before starting each one. These were verified on 17 August 2026. Fields move, and topics 4, 5 and 6 sit close to active work.

Read the neighbouring literature before writing, not after. For iltifāt that means Abdel Haleem; for ring composition, Cuypers and Sinai; for verse endings, Klar. Citing is not enough — a reviewer can tell.

Expect one or two of these nine to close as well. On the checked sample the attrition rate has been high, and there is no reason to think this batch is exempt.