The residue 108

Every divergence, examined.

The 108 sites where the engine and the reference datasets disagree, in six groups. For each group: the sites, both readings, the evidence consulted, and the verdict with its basis.

Method note. Each group was researched against the published tajweed literature twice, in two independent passes (2026-07-11 and 2026-07-12), and checked against both datasets and, where decisive, against page photographs of the printed Dar al-Maarifah tajweed muṣḥaf. The engine is never changed merely to match a dataset; it is changed when the evidence shows the engine wrong, which happened once (R-006, documented below). The canonical record is RESIDUE.md in the repository.

R-001. Idghām mutajānisayn ṭāʾ into tāʾ: 4 sites

Sites. 5:28 بَسَطتَ, 12:80 فَرَّطتُمْ, 27:22 أَحَطتُ, 39:56 فَرَّطتُ.

The engine derives. A ṭāʾ sākinah before tāʾ: same articulation point, different attributes, therefore idghām mutajānisayn (Tuḥfah 32 and 33), and of the incomplete (nāqiṣ) kind: the ṭāʾ's iṭbāq and istiʿlāʾ are retained through the assimilation.

The datasets say. Nothing, at all four sites. Both datasets key complete idghām on the written shaddah of the following letter; the muṣḥaf's rasm convention for partial idghām writes the ṭāʾ without sukūn and the tāʾ without shaddah, so a shaddah-keyed transcriber sees nothing to transcribe. Notably, the Quran.com dataset possesses an idgham_mutajanisayn class and fires it at قَد تَّبَيَّنَ (2:256), where the shaddah is written; its silence at the nāqiṣ sites is therefore a gap, not a position.

The literature. The nāqiṣ doctrine at these exact words is standard. Al-Mīzān fī Aḥkām Tajwīd al-Qurʾān names them in its chapter on the small mutajānis (p. 141); al-Nuwayrī's Sharḥ Ṭayyibat al-Nashr carries the verse line "وبين الإطباق من أحطت مع بسطت", which names two of the four sites in the poem itself; the teaching literature (Ayman Suwayd's lessons among others) uses بسطت and أحطت as the canonical examples of al-idghām al-nāqiṣ.

The print. Page 112 of the Dar al-Maarifah muṣḥaf prints the ṭāʾ of بَسَطتَ in blue, its retained-tafkhīm class: the idghām nāqiṣ doctrine rendered in color. The datasets missed it because neither ever digitized the blue class.

Verdict. The print annotates it, the literature names it, both datasets dropped it. The engine's derivation stands.

R-002. The lām of ٱللَّهُمَّ: 6 sites

Sites. 3:26, 5:114, 8:32, 10:10, 39:46, 62:11, every occurrence of ٱللَّهُمَّ.

The engine derives. ٱللَّهُمَّ is lafẓ al-jalālah with a suffixed mīm; its doubled lām takes the jalālah-lām rule (al-Jazariyyah 43): tafkhīm after a fatḥah or ḍammah, tarqīq after a kasrah, exactly as in the unsuffixed lafẓ al-jalālah.

The datasets say. Both annotate the first lām of ٱللَّهُمَّ as lām shamsiyyah, while carrying no lām annotation on the unsuffixed lafẓ al-jalālah anywhere. The asymmetry is the tell: the same lām in the same word is treated differently depending only on whether a mīm follows.

The literature. Current references state the jalālah-lām rule for the unsuffixed and suffixed forms of lafẓ al-jalālah together, and one gives وَإِذْ قَالُوا۟ ٱللَّهُمَّ, which is 8:32, one of these six sites, as its textbook example of the tafkhīm case.

The print. The page images resolve the asymmetry completely. At 3:26 (a tarqīq context) the first lām of ٱللَّهُمَّ is printed gray, the print's assimilated-letter class, with the rest black. At 10:10 (tafkhīm contexts) the jalālah lām of both ٱللَّهُمَّ and وَإِلَى ٱللَّهِ is printed blue, the tafkhīm class. The print gives ٱللَّهُمَّ exactly the treatment it gives the unsuffixed lafẓ al-jalālah. The datasets digitized the gray as lām shamsiyyah and dropped the blue class entirely; their asymmetry is a digitization artifact, not a scholarly position.

Verdict. The engine's jalālah tafkhīm and tarqīq output matches the print's actual color distinction better than either dataset does.

R-003. Munfaṣil ḥukmī sites the reference calls muttaṣil: 45 sites

Sites. يَٰٓـَٔادَمُ at 2:33, 2:35, 20:117 and 20:120, and the first madd of هَٰٓؤُلَآءِ at its 41 occurrences.

The engine derives. The vocative يَا and the deictic هَا are written joined to the following hamzah-initial word but are separate words in ruling. A madd letter at the end of one word with a hamzah opening the next is madd munfaṣil (Tuḥfah 44, "كُلٌّ بِكِلْمَةٍ"); when the separation is in ruling rather than in writing, the category is munfaṣil ḥukmī.

The reference says. Muttaṣil, at exactly these sites, while classifying يَٰٓأَيُّهَا and هَٰٓأَنتُمْ (hamzah seated on alif) correctly as munfaṣil. The criterion producing this split is visible in its published decision trees: it branches on the written form of the hamzah, not on the word boundary. See the root-cause analysis.

The literature. The munfaṣil ḥukmī category is stated by name across the literature, with these words as its standard examples. Ghāyat al-Murīd's chapter on the munfaṣil states that the madd is ḥukmī in the vocative yāʾ and the deictic hāʾ even when joined in the rasm; on هؤلاء specifically the analysis is explicit: the first madd is jāʾiz munfaṣil in ruling rather than in writing, the word's origin being ها أولاء, while the second madd, in لَآء, is wājib muttaṣil. The re-verification pass confirmed the same doctrine independently in Hidāyat al-Qārī's treatment of the connected-in-writing cases and in the general teaching literature.

The decisive test. The two categories differ in law: muttaṣil is obligatory lengthening by consensus, while munfaṣil admits shortening in the transmissions that shorten it. The literature confirms the first madd of هؤلاء may be read at two counts where qaṣr al-munfaṣil is transmitted (for Ḥafṣ, by ṭuruq of the Ṭayyibah). That reading would be impermissible for a muttaṣil. The recitation tradition therefore settles the classification independently of any dataset.

Verdict. Reference-dataset defect; the engine's derivation stands. For completeness: the engine originally matched the reference at هؤلاء under a one-word reading, and was corrected to follow the literature on 2026-07-11. The correction was made because of the books, not because of a dataset.

R-004. تَبُوٓأَ, تَنُوٓأُ, ٱلسُّوٓأَىٰ called munfaṣil: 3 sites

Sites. 5:29 تَبُوٓأَ, 28:76 تَنُوٓأُ, 30:10 ٱلسُّوٓأَىٰ.

The engine derives. A madd letter followed by a hamzah inside one word is muttaṣil by definition (Tuḥfah 43). These are single words, from the roots بوء, نوء and سوء; there is no particle and no morpheme boundary at the madd.

The reference says. Munfaṣil at all three: the same seat-of-the-hamzah criterion as R-003, here firing in the opposite direction, because the hamzah happens to be seated on an alif.

The literature. The muttaṣil definition (madd and hamzah in one word) is uncontested, and سوء-family words appear among the standard muttaṣil examples in the references. No source treats any of these three words as munfaṣil.

Verdict. Reference-dataset defect; the engine's derivation stands.

R-005. 17:7 لِيَسُـۥٓـُٔوا۟: 1 site

The engine derives. The small wāw restores a rasm-omitted letter of the word لِيَسُوءُوا۟ itself. A madd letter followed by a voweled hamzah in the same word is muttaṣil: four to five counts. It is not a ṣila; there is no pronoun hāʾ anywhere in the word.

The datasets say. Both give it two counts. The reference's tree apparently treats every small wāw as a two-count ṣila; the Quran.com dataset does the same here, and contradicts itself in doing so, since at 30:10 it marks the identical pattern (wāw with maddah followed by hamzah, in ٱلسُّوٓأَ) as obligatory length. Only the glyph differs between the two sites.

The print. Decisive here. Page 282 prints the small wāw and its maddah in red, the legend's obligatory madd of four or five counts. The muṣḥaf also prints the maddah sign itself on this wāw, which is the printed hint for a long madd; both datasets' two-count classification contradicts the very page they claim to encode.

Verdict. Print-confirmed; the engine's derivation stands, and both datasets carry a glyph-keyed undercount at this site.

R-006. 2:72 فَٱدَّٰرَٰٔتُمْ: resolved against the engine

This entry is retained because a verification record that only ever finds the other side wrong is not credible.

The puzzle. The engine's early letter model read the superscript-alif and hamzah cluster in فَٱدَّٰرَٰٔتُمْ as a dagger-alif madd letter followed by a sākin hamzah, forcing an invented policy for a madd before a sākin hamzah, and produced a madd annotation where the reference had none.

The resolution. The tafsīr and iʿrāb literature vocalizes the word iddāraʾtum (from تدارأتم, with the tāʾ assimilated into the dāl and a connective hamzah prefixed; so al-Naḥḥās in Iʿrāb al-Qurʾān, and al-Bayḍāwī ad loc.). The hamzah is sākinah directly after the rāʾ; there is no madd letter at that position. The Tanzil cluster is a hamzah seated on a superscript alif, parallel to the tatweel-seated hamzah. The tokenizer was corrected to treat the cluster as a single hamzah letter; the spurious annotation disappeared; the reference's silence at this site had been correct all along.

Verdict. Engine letter-model defect, found by this comparison and fixed, with regression tests. This is the verification process working as intended.

R-007. Assimilation across the fawātiḥ letter names: 6 sites

Sites. طسٓمٓ at 26:1 and 28:1 (the nūn ending the name of sīn assimilates into mīm: idghām with ghunnah); كٓهيعٓصٓ at 19:1 (the nūn of ʿayn before ṣād: ikhfāʾ); حمٓ عٓسٓقٓ at 42:1 and 42:2 (the nūn of ʿayn before sīn, and of sīn before qāf: ikhfāʾ); and طسٓ تِلْكَ at 27:1 (the nūn of sīn before tāʾ, across the verse's first word boundary: ikhfāʾ).

The engine derives. The opening letters are recited by their names, and several of those names end in a nūn sākinah. The nūn sākinah family then applies between the names exactly as it would between words. This requires a model in which rules operate on the recited letter names rather than the written letters, which the engine implements for the fawātiḥ.

The datasets say. Both annotate only the madd of the letter names; neither carries any assimilation annotation inside a fawātiḥ verse.

The literature. The recitation is not in dispute. The sources state that in the hijāʾ of طسم the nūn is assimilated into the mīm by all readers except Ḥamzah and Abū Jaʿfar (Ḥafṣ reads full idghām with ghunnah), and they note it explicitly as the exception to the same-word restriction on idghām; the ikhfāʾ across كهيعص and عسق is standard.

The print. Page 367 shows طسٓمٓ carrying only the lāzim-madd coloring, and necessarily so: the assimilating nūn exists in the letter name سين, which has no glyph on the page to color. Print-style annotation is structurally unable to express this rule. Only an engine that models letter names can annotate it.

Verdict. An annotation-scope gap in both datasets, and an unavoidable one. These six sites are classified as a scope difference rather than as dataset errors; they are recorded because the underlying phenomenon, rules applying to letter names rather than letters, is a genuine formalization point that distinguishes derivation from transcription.

Summary table

The group counts above are word sites. The figure of 108 counts disagreement spans as the diff harness sees them, where one misclassified word can contribute two spans: a class wrongly asserted and a class missing. Under that accounting the seat-glyph defect of R-003 and R-004 accounts for 90 of the 108 spans, and the remaining groups for the rest. Both accountings cover the same set of disagreements; nothing is outside the catalog.