# COUNCIL OF 120 — BENCH 4 · THE SĪRAH BENCH
**30 seats · maghāzī ×6 · chronology ×6 · sīrah-sources criticism ×6 · sīrah–tafsīr coupling ×6 · shamāʾil/dalāʾil ×6**
*Machine-scholar council, labeled as such. Human review invited, never waited for. Review and counsel only — this bench edited nothing.*
*Sitting: 2026-09-04, ~01:47–02:20 AST. Governing law: `10-WAQF/skills/deep-audit/SKILL.md` (18 laws, read first).*

---

## VERDICT

**THE SĪRAH CORPUS IS WHOLE AND LARGER THAN THE BRIEF STATED — 47,083,188 characters ✅ VERIFIED, not ~34M (that figure is the Sunnī file alone, 33,980,512). ALL 17 WORKS OPEN AND CLOSE COHERENTLY. THE SĪRAH-AS-LENS DOCTRINE IS SOUND AND THE HELD TEXTS DO CARRY THE OCCASION-BOUND NARRATIVE — all three assigned probes returned quoted passages. BUT THE E3 LAYER THAT MEASURES IT IS DEFECTIVE AND ITS HEADLINE NUMBER SHIPPED WRONG.**

Three defects, each measured, each cheap to fix:

1. **E3's "2,821 verses never quoted in the sīrah" is false by at least 414.** ✅ VERIFIED by independent replication under the uniqueness rule. Among the recovered: **111:1, 112:1, 96:5, 108:3, 109:1, 93:3, 53:4, 21:107** — the most occasion-bound verses in the Book. This number is already published in `THE-ANSWERS.md` §4.
2. **`00-CORPUS/quran.json` silently prefixes the basmalah onto āyah 1 of 112 sūrahs.** E3 consumed it uncorrected; E2 did not. Two layers, two verse-normalizations — a live breach of SKILL law 11 (SINGLE-SOURCE).
3. **The corpus carries 7,203 explicit Qurʾān-quotation delimiters that no layer has ever read.** They are a free ground-truth set. Measured against them, E3's recall is **91.8%**.

**One structural finding the plan must absorb before it spends a node:** six of this bench's thirty seats are shamāʾil/dalāʾil specialists, and **neither genre's founding text is held** — al-Tirmidhī's *al-Shamāʾil al-Muḥammadiyya* is absent, al-Bayhaqī's *Dalāʾil al-Nubuwwa* is absent. A shamāʾil/dalāʾil domain block cannot be built from held sources today.

**Bench posture: the 1,800-node Sīrah Universe is APPROVED IN MAGNITUDE, AMENDED IN SEQUENCE.** No new sīrah domain opens until the repair block closes. Votes in **COUNSEL** below.

---

## PART I — THE SURVEY: IS THE TEXT WHOLE, AND WHAT GENRE IS IT?

### F1 · The ocean is 47,083,188 chars — 38% larger than the brief stated 🟢 NOTE
**File** `00-CORPUS/seerah.json` (10 works) + `00-CORPUS/seerah_shia.json` (7 works)
**Evidence** ✅ VERIFIED — `len(text)` measured per work against `meta.works[k].chars`; **delta = 0 on all 17**. The corpus's self-declared char counts are honest. Total measured 47,083,188. The brief's "~34M" equals the Sunnī file exactly (33,980,512) — the Shīʿī file (13,102,676) was dropped from the brief's estimate.
**Severity** 🟢 — E3 already carries the correct 47,083,188. The brief was stale; the layer was right.

### F2 · All 17 works open and close coherently — no truncation, one soft flag 🟢/🟡
**Evidence** ✅ VERIFIED — head(260) and tail(260) read on all 17. Every work opens with a title line + basmalah + a recognisable preface; 15 of 17 close on a colophon (`تم الكتاب`, `تمت`, `والحمد لله رب العالمين`, or a dated scribal note).
Two soft flags:
- **`ibn_sad`** ends mid-isnād — `…عن الحكم عن حجية بن عدي عن علي الأسدي [vol 8 p.498]` — no colophon. 🟡 Possible tail truncation of *al-Ṭabaqāt al-Kubrā*.
- **`yaqubi`** ends inside the editor's footnote apparatus — `…(1) بياض في الأصل` repeated — narrative end not visible. 🟡

### F3 · `ibn_ishaq` is the Damascus fragment, not *Sīrat Rasūl Allāh* whole — and it is NOT a duplicate of Ibn Hishām 🟠 HIGH (labelling)
**File** `00-CORPUS/seerah.json` → `seerah.ibn_ishaq` (233,380 chars = 0.50% of the ocean)
**Evidence** ✅ VERIFIED — the file's **tail** reads `…وذلك بمدينة دمشق في الجامع في العشر الاول من ذى الحجة سنة أربع وخمسين وأربع مائة **تمت قطعة دمشق من الكتاب** [vol 3 p.316]` — "the *Damascus fragment* of the book is complete." Its **head** is Ibn Hishām's own preface verbatim: `قال أبو محمد عبد الملك بن هشام هذا كتاب سيرة رسول الله`.
**The false alarm this bench refused to raise:** the shared preface looks like duplication. It is not. 8-gram shingle overlap measured: **3,133 of 44,203 ibn_ishaq shingles present in ibn_hisham = 7.1%**. 📊 CALCULATED. The file is a genuinely distinct recension carrying a borrowed paratext. *A false alarm costs trust as much as a false all-clear — this one was checked before it was voiced.*
**Ask** re-label to `Ibn Isḥāq — Sīra (qiṭʿat Dimashq, partial recension)`; the meta note "Survives mainly through later recensions" is true but does not tell the reader that **this file is one of those fragments**, not the work.

### F4 · The "sīrah ocean" is 6.5% maghāzī and 56.5% universal history 🟠 HIGH (doctrine)
📊 CALCULATED from measured char counts:

| Genre | Works | Chars | Share |
|---|---|---:|---:|
| **Maghāzī proper** | ibn_hisham · waqidi · ibn_ishaq | 3,054,583 | **6.5%** |
| Universal history / annals | tabari · ibn_kathir · baladhuri · masudi · yaqubi | 26,593,110 | 56.5% |
| Commentary (sharḥ) | ibnabihadid · suhayli | 8,108,105 | 17.2% |
| Prosopography (ṭabaqāt) | ibn_sad | 4,377,821 | 9.3% |
| Later sīra-proper | ibn_sayyid · mufid_irs · tabarsi | 2,804,421 | 6.0% |
| Amālī / majālis | tusi_amali · mufid_amali | 1,411,266 | 3.0% |
| Devotional / fadāʾil-ḥuqūq | qadi_iyad | 733,882 | 1.6% |
| **Total** | 17 | **47,083,188** | 100% |

**Why this is a doctrine finding, not a trivia finding:** the four-witness doctrine calls witness #3 "the SĪRAH (the lived explanation)". Fifty-six per cent of what the waqf will cite as that witness is *annalistic universal history* whose Prophetic section is a minority of its bulk (Ibn Kathīr's *al-Bidāya* runs to vol 14, ending on 8th-century Frankish politics: `…وأطلقت الفرنج الذين كانوا بالقلعة المنصورة… [vol 14 p.324]` ✅ VERIFIED at tail). A verse "quoted by the sīrah" may in fact be quoted by al-Ṭabarī while narrating the ʿAbbāsid caliphate. **Genre must ride every sīrah citation, and the maghāzī-window must be separable.**

### F5 · The Shīʿī sīrah block is 52.5% one Muʿtazilī commentary 🟠 HIGH (labelling)
📊 CALCULATED — `ibnabihadid` (*Sharḥ Nahj al-Balāgha*) = 6,878,673 of 13,102,676 Shīʿī-labelled chars = **52.5%**. The corpus's own meta says it plainly: *"Note the author was Muʿtazilī, not Imāmī - which is itself worth stating in a citation."* ✅ VERIFIED at `seerah_shia.json → meta.works.ibnabihadid.note`. **The corpus is more honest than the label above it.** E3 files it under `"tradition": "shia-tradition sources"` with no qualifier. Any claim of the form "the Shīʿī sīrah says X" is, on the balance of bulk, a coin-flip on whether it is a Muʿtazilī saying it.
**Ask** split the tradition field into `imami` / `shia-transmitted-non-imami` / `sunni`. al-Masʿūdī (15.9%) and al-Yaʿqūbī (8.0%) carry the same contested attribution and the same need.

### F6 · Markup contamination in the Shīʿī file 🟡 MEDIUM
✅ VERIFIED — HTML tags counted per work: **`mufid_irs` = 2,371**, **`tusi_amali` = 126**, all others 0. Latin-character runs: `mufid_irs` = 4,751 vs 752 volume-markers — the excess is href attributes. Live specimen inside a Qurʾān quotation:
`<a class="quran" href="http://qadatona.org/عربي/القرآن- الكريم/8/8" target="_blank" title="الأنفال 8">﴿كما أخرجك ربك…﴾</a>`
The tag text is inert to a rasm-skeleton engine (it is stripped), so **no measurement is corrupted** — but the artifact is polluted, and any future word-level or display layer will surface it.

---

## PART II — THE INSTRUMENT: 7,203 QUOTATION MARKS NOBODY READ

### F7 · The corpus delimits its own Qurʾān quotations, in two families, and no layer uses them 🔴 CRITICAL (missed asset)
✅ VERIFIED by count, both families perfectly balanced (open == close in every work):

| Family | Works carrying it | Count |
|---|---|---:|
| `@QB@ … @QE@` (OpenITI Qurʾān markers) | ibn_hisham 862 · suhayli 519 · tabari 540 · ibn_kathir 3,707 · ibn_ishaq 133 · ibn_sad 1 | **5,762** |
| `﴿ … ﴾` (ornate parentheses) | ibnabihadid 1,161 · tusi_amali 126 · tabarsi 87 · mufid_irs 67 | **1,441** |
| | | **7,203** |

Note the complementarity: `@QB@` appears **only** in the Sunnī file, `﴿﴾` **only** in the Shīʿī file. Together they cover 10 of 17 works and 65% of the ocean by chars.

**This is a KNOWN-POSITIVE corpus (SKILL law 5) sitting unused.** It is the one instrument that can grade the skeleton engine without circularity — the corpus says "this span is Qurʾān"; the engine either finds it or does not.

### F8 · Measured against that ground truth, E3's recall is 91.8% — 233 delimiter-proven verses are missing 🟠 HIGH
📊 CALCULATED — this bench resolved the 5,762 `@QB@` spans to verses under the **uniqueness rule** (≥1 verse-unique 4-gram required; SKILL law 16):

| Work | spans | resolved to a unique verse | ambiguous only | unresolved (span < 4 words or crosses a verse boundary) | distinct verses |
|---|---:|---:|---:|---:|---:|
| ibn_kathir | 3,707 | 3,093 | 97 | 517 | 2,436 |
| ibn_hisham | 862 | 685 | 36 | 141 | 646 |
| tabari | 540 | 457 | 20 | 63 | 362 |
| suhayli | 519 | 287 | 17 | 215 | 238 |
| ibn_ishaq | 133 | 95 | 6 | 32 | 86 |
| ibn_sad | 1 | 1 | 0 | 0 | 1 |
| **union** | **5,762** | | | | **2,834** |

**2,834 ground-truth verses. E3 holds 2,601 of them. 233 are absent.** Recall **91.8%**.
The unresolved column is method honesty, not engine failure: `" وقال "`, `" أتقاكم "`, `" العزيز "` are sub-4-word spans, and `" الم غلبت الروم في أدنى الأرض "` spans 30:1–2 so no 4-gram lies inside one verse. Declared, not hidden.

**Checker discipline (SKILL law 18):** this bench's first run of the delimiter resolver returned **0 verses from 5,762 spans**. The engine was not wrong — the *checker* was: the skeleton function dropped Unicode category `Zs`, collapsing every text to one unbroken word. Fixed, re-run, 2,834. **The first instinct — "E3 is fabricating" — was wrong and was suppressed before it reached this page.**

---

## PART III — THE DEFECT: A BASMALAH IN 112 SŪRAH OPENINGS

### F9 · `quran.json` prefixes the basmalah onto āyah 1 of 112 sūrahs 🔴 CRITICAL
**File** `00-CORPUS/quran.json` → `ayat[*].simple` **and** `ayat[*].uthmani`
✅ VERIFIED — 113 verses have `simple` beginning `بسم الله الرحمن الرحيم`; 1:1 legitimately *is* the basmalah, leaving **112 corrupted sūrah-openings** (every sūrah but al-Fātiḥa and al-Tawba). Specimens:

```
111:1 simple  : بسم الله الرحمن الرحيم تبت يدا أبي لهب وتب
111:1 uthmani : بِسْمِ ٱللَّهِ ٱلرَّحْمَٰنِ ٱلرَّحِيمِ تَبَّتْ يَدَآ أَبِى لَهَبٍ وَتَبَّ
112:1 simple  : بسم الله الرحمن الرحيم قل هو الله أحد
```

The corruption is in **both** Arabic fields, so it is not escapable by switching field.

**Damage measured in E3** 📊 CALCULATED:
- of the 112 corrupted openings, **78 are absent from E3** (69.6%) against a **45.2% baseline absence** across all 6,236 verses — an excess of ~27 points concentrated exactly where the defect is.
- 34 survive, because E3 matched interior sub-runs. Partial recovery, not immunity.

### F10 · E2 and E3 disagree on what verse 112:1 *is* — SKILL law 11 breached live 🔴 CRITICAL
E2 reports 112:1 as **249 FULL-verse quotes** — its single top verse, and a headline in `THE-ANSWERS.md` §2. This bench read a top hit (law 16):

> `abudawud #815` — `…أَنَّهُ صَلَّى خَلْفَ ابْنِ مَسْعُودٍ الْمَغْرِبَ فَقَرَأَ بِـ ﴿ قُلْ هُوَ اللَّهُ أَحَدٌ ﴾`
> `abudawud #1256` — `…قَرَأَ فِي رَكْعَتَىِ الْفَجْرِ ﴿ قُلْ يَا أَيُّهَا الْكَافِرُونَ ﴾ وَ ﴿ قُلْ هُوَ اللَّهُ أَحَدٌ ﴾`

**E2 is vindicated** — these are genuine verbatim quotations, not basmalah artefacts. But that vindication *is* the finding: for `قل هو الله أحد` alone to score as a **full** quote, E2's 112:1 must be basmalah-free, while E3's manifestly is not. **Two layers, two verse texts, one corpus.** Whichever is right, they cannot both be the single source.

### F11 · E3 is blind to short verses — 78% of ≤5-word verses absent 🟠 HIGH
📊 CALCULATED — of 1,476 verses ≤5 words (excluding the basmalah-prefixed openings), **1,149 are absent from E3 = 77.8%**, against the 45.2% baseline. And of the 233 delimiter-proven misses: **130 are ≤5 words, 78 are 6–10, 25 are >10.**
This is a *second, independent* cause from F9 — a gram/run-length floor. It matters doctrinally because short verses are disproportionately early-Makkan and disproportionately occasion-bound: sūrahs 111, 108, 109, 112, 93, 94, 105, 106. **The lens is dimmest exactly where the sīrah has most to say.**

### F12 · Independent replication: 414 of the 2,821 "never quoted" verses ARE quoted 🔴 CRITICAL
Method declared (this bench's own, run once, no divergent re-probe): basmalah-stripped verse texts → rasm skeletons → 4-word grams → **only grams unique to a single verse across all 6,236** retained (uniqueness rule) → single pass over all 17 works, 8,876,213 skeleton words.

- target set: the 2,821 verses E3 calls never-quoted
- of those, **1,968 possess at least one verse-unique 4-gram** — the honest denominator; 853 are structurally unprobeable by this method and are declared, not counted
- **found: 414 verses (21.0% of the probeable denominator), 17/17 works contributing**

| verse | hits | carried by |
|---|---:|---|
| 6:128 | 77 | ibn_kathir 20 · tabari 14 · ibn_sad 12 · ibn_hisham 7 |
| 28:26 | 65 | ibn_kathir 28 · baladhuri 14 · tabari 14 |
| **111:1** | **43** | baladhuri 10 · ibn_kathir 10 · suhayli 5 · ibn_hisham 4 |
| 21:69 | 39 | ibn_kathir 15 · tabari 8 · mufid_irs 4 · tabarsi 4 |
| **112:1** | **34** | ibn_kathir 14 · tusi_amali 5 · baladhuri 4 |
| 38:5 | 33 | ibn_hisham 11 · ibn_kathir 11 · tabari 5 |
| **96:5** | **29** | ibn_kathir 10 · ibn_sayyid 4 · masudi 4 |
| 53:18 | 20 | qadi_iyad 18 · suhayli 2 |
| **93:3** | **20** | ibn_ishaq 6 · qadi_iyad 4 · ibn_kathir 4 |
| **109:1** | **17** | tusi_amali 3 · ibn_sad 2 · baladhuri 2 |
| **108:3** | **16** | baladhuri 3 · ibn_kathir 3 · qadi_iyad 2 |
| 21:107 | 22 | qadi_iyad 10 · ibn_kathir 4 |

**Corrected floors** 📊 CALCULATED:
- `verses_quoted_somewhere` **3,415 → ≥ 3,829**
- `verses_never_quoted` **2,821 → ≤ 2,407**
- `THE-ANSWERS.md §4` and E6's convergence table both inherit the error. E6's `neither = 702` shrinks by between 0 and 414 (this bench cannot split the 414 across E6's ḥadīth/no-ḥadīth partition without E6's per-verse lists — **the bound is stated, not blended**).

These are **floors, not the truth**. Both delimiter families and both defect classes remain unrepaired; the honest reading is that the true never-quoted set is smaller still.

---

## PART IV — THE LENS QUESTION: DO THE HELD TEXTS CARRY THE NARRATIVE?

Three occasion-bound verses, assigned. **All three: YES**, with passages quoted. Method: verse skeleton (basmalah-stripped) → first-6-word probe → raw context recovered through a character-offset map so the quotation below is the corpus's own text, not a reconstruction.

### PROBE 1 — 8:5 (the Badr class) ✅ CARRIED · 6 occurrences · 4 works
**`ibn_hisham`** — gives the verse *with* its exegetical gloss, inline:
> `…ما نزل في خروج المسلمين لملاقات قريش… وإنما خرجوا يريدون العير طمعا في الغنيمة فقال @QB@ كما أخرجك ربك من بيتك بالحق وإن فريقا من المؤمنين لكارهون يجادلونك في الحق بعد ما تبين كأنما يساقون إلى الموت وهم ينظرون @QE@ **أي كراهية للقاء القوم وإنكار لمسير قريش حين ذكروا لهم** @QB@ وإذ يعدكم الله إحدى الطائفتين… @QE@ **أي الغنيمة دون الحرب**…`

**`waqidi`** — gives an *isnād-bearing* gloss, verse by verse:
> `…وفي قوله : ' كما أخرجك ربك من بيتك بالحق ' . **يقول : لما أمرك ربك أن تخرج إلى بدر هو الحق** . وأخبرني ابن جريج ، عن محمد بن عباد بن جعفر المخزومي في قوله : ' من بيتك ' . **قال : من المدينة** .`

**`ibn_kathir`** — sets it as the chapter epigraph of the Badr section: `…غزوة بدر العظمى يوم الفرقان يوم التقى الجمعان قال الله تعالى @QB@ ولقد نصركم الله ببدر… @QE@ وقال الله تعالى @QB@ كما أخرجك ربك من بيتك بالحق… @QE@`, then attaches al-Miqdād's speech and the Anṣār's response — the ʿoccasion narrative in full.

**`mufid_irs`** — the Imāmī witness, carrying the same verse with a different frame (`أول حرب كان بها الامتحان`), confirming that the lens holds across traditions on this verse.
**Bench reading:** the strongest of the three. Ibn Hishām supplies *tafsīr*, al-Wāqidī supplies *isnād*, Ibn Kathīr supplies *structure*, al-Mufīd supplies *symmetry*. This is precisely what the doctrine promises the sīrah witness will do.

### PROBE 2 — 33:37 (Zayd and Zaynab) ✅ CARRIED · 7 occurrences · 4 works
**`ibn_sad`** — the occasion in the mouth of a wife of the Prophet ﷺ, with the verse cited as her proof:
> `…فقالت زينب إني والله ما أنا كأحد من نساء رسول الله صلى الله عليه وسلم إنهن زوجهن بالمهور وزوجهن الأولياء **وزوجني الله رسوله وأنزل في الكتاب يقرأ به المسلمون لا يبدل ولا يغير** وإذ تقول للذي أنعم الله عليه الآية…`

**`tabari`** — two distinct occasion-reports for the same verse, side by side (this is the divergence the doctrine exists to make visible): one where the Prophet ﷺ recites it on receiving revelation in ʿĀʾisha's presence; one, **on al-Wāqidī's authority**, glossing `وتخفي في نفسك` as `إن فارقها تزوجتها`.
**`qadi_iyad`** — does not narrate; he *adjudicates*, and refuses the surface reading: `فاعلم أكرمك الله ، **ولا تسترب في تنزيه النبي صلى الله عليه وسلم عن هذا الظاهر** … كما ذكر عن جماعة من المفسرين . و أصح ما في هذا ما حكاه أهل التفسير…`
**`ibn_kathir`** — declares a deliberate omission: `…حديثا ذكره أحمد بن حنبل في مسنده **تركنا ايراده قصدا لئلا يضعه من لا يفهم على غير موضعه**` — an author suppressing a Musnad report on prudential grounds, inside the sīrah corpus.
**Bench reading:** this single verse exercises every mechanism the plan needs — multiple occasions, an explicit al-Wāqidī attribution, a theological objection to the plain sense, and a named deliberate suppression. **Make 33:37 the golden test-case of the P32 conflict detector.** If the layer cannot render this verse honestly, it cannot render anything.

### PROBE 3 — 111:1–5 (al-Masad) ✅ CARRIED · 13 occurrences on 111:1, 5 on 111:3 · 7 works — **AND ABSENT FROM E3**
This is the probe that broke the layer. E3 lists 111:1 among the 2,821 never quoted. The corpus:

**`ibn_hisham`** — gives *two distinct occasions* for the same verse, ~3 pages apart:
> [vol 2 p.196] `…يعدني محمد أشياء لا اراها يزعم انها كائنة بعد الموت فماذا وضع في يدي بعد ذلك **ثم ينفخ في يديه ويقول تبا لكما ما ارى فيكما شيئا مما يقول محمد** فأنزل الله تعالى فيه @QB@ تبت يدا أبي لهب وتب @QE@`
> [vol 2 p.199] `…وامرأته أم جميل بنت حرب بن أمية حمالة الحطب **وإنما سماها الله تعلى حمالة الحطب لأنها كانت فيما بلغني تحمل الشوك فتطرحه على طريق رسول الله** … @QB@ تبت يدا أبي لهب وتب ما أغنى عنه ماله وما كسب سيصلى نارا ذات لهب وامرأته حمالة الحطب في جيدها حبل من مسد @QE@` — **the whole sūrah, delimited, verses 1–5 in one span.**

**`ibn_sad`** and **`baladhuri`** and **`tabari`** — the al-Ṣafā occasion: `…فقال أبو لهب تبا لك ألهذا دعوتنا أو جمعتنا فأنزل الله عز وجل تبت يدا أبي لهب وتب` ✅ four independent works.
**`suhayli`** — performs verse-by-verse tafsīr *of the sūrah's internal structure*: `فقوله @QB@ تبت يدا أبي @QE@ تفسيره قوله @QB@ ما أغنى عنه ماله وما كسب @QE@ … وقوله وتب تفسيره @QB@ سيصلى نارا ذات لهب @QE@ … **فالكلمتان مبنيتان على السببين والآيتان بعدهما تفسير للتببين**`
**`qadi_iyad`** — carries Umm Jamīl's confrontation at the Kaʿba with the *fihr* in her hand.
**Bench reading:** **five works, two competing occasions, a structural tafsīr, and the full sūrah under explicit delimiters — and the layer reports the verse as never touched by the sīrah.** No finding in this review is more concrete than this one.

---

## PART V — Q4 IS THIS BENCH'S: THE TRANSMITTED PRECEDENCE POSITIONS

*The waqf never rules. Every position below is stated as the tradition's own, with the name that carries it. Where the tradition is split, both sides are given. Where the source is not held, it is marked **NAMED-OPEN**.*

### V.1 · The measured precondition — the sīrah corpus has chains but no grades
✅ VERIFIED by count across all 47,083,188 chars:

| marker | total | reading |
|---|---:|---|
| isnād verbs (`حدثنا/أخبرنا/حدثني/أخبرني`) | **55,007** | a chain-bearing khabar registry **is** buildable from held text |
| grading strings (`إسناده صحيح/ضعيف`, `هذا حديث صحيح`, `حديث حسن`) | **67** | grading is **not** buildable from held text — 60 of the 67 are Ibn Kathīr's |
| `قال ابن إسحاق` | **2,485** | the attribution graph is measurable |
| `قال الواقدي` | **1,425** | ditto — and see V.4 |
| occasion formulae (`فأنزل الله`, `فنزلت`, `نزلت هذه الآي`) | **863** | the asbāb bridge has a held seed, but a thin one |
| `غزوة`/`سرية` | **2,488** | maghāzī registry buildable |
| `ثم دخلت سنة` | **971** (tabari 265 · ibn_kathir 706) | annalistic spine buildable |

**Therefore SKILL §"FOUR-WITNESS DOCTRINE" as presently written is inaccurate for the held corpus.** It says the sīrah is "transmitted through akhbār *whose grading conventions the tradition records*". True of the tradition; **false of what we hold** — 67 grading strings in 47 million characters. Amendment proposed in COUNSEL Q3.

### V.2 · Position A — the genre rule (why sīrah is graded differently at all)
- **Aḥmad b. Ḥanbal (d. 241)**, transmitted by **al-Khaṭīb al-Baghdādī** (*al-Jāmiʿ li-Akhlāq al-Rāwī*) and cited by **Ibn Taymiyya** (*Muqaddima fī Uṣūl al-Tafsīr*): *«ثلاثة ليس لها أصل: المغازي والملاحم والتفسير»* — three genres "have no aṣl". **The tradition's own two readings are both transmitted:** (i) the literal — these genres lack sound chains; (ii) the restrictive — isnād-less material *predominates* in them, not that all of it is void. **Both must be carried. The waqf does not choose.**
- **al-Khaṭīb al-Baghdādī, *al-Kifāya fī ʿIlm al-Riwāya*** — the *tasāhul* doctrine: leniency of isnād is permitted in *faḍāʾil, mawāʿiẓ, qiṣaṣ* and *maghāzī*; **not** in *ḥalāl wa-ḥarām* and *aḥkām*. Traced to **ʿAbd al-Raḥmān b. Mahdī**, **Sufyān al-Thawrī**, **Aḥmad b. Ḥanbal**. **NAMED-OPEN** (al-Kifāya not held).
- **The Imāmī mirror**: *التسامح في أدلة السنن* — **al-Shahīd al-Thānī**, **al-Ḥurr al-ʿĀmilī**. Same shape, independent lineage. **NAMED-OPEN.**

### V.3 · Position B — Ibn Isḥāq (d. 151): praised and impugned, both transmitted
| Side | Carrier | Position |
|---|---|---|
| for | **Shuʿba b. al-Ḥajjāj** | *«أمير المؤمنين في الحديث»* |
| for | **al-Zuhrī**, **Sufyān b. ʿUyayna**, **al-Bukhārī** (istishhād/taʿlīq, and *al-Adab al-Mufrad*), **Muslim** (in the mutābaʿāt) | accepted in corroboration |
| for | **Ibn Sayyid al-Nās**, muqaddima to *ʿUyūn al-Athar* | the classical defence dossier — **✅ HELD** at `seerah.json → ibn_sayyid` (1,129,782 chars) |
| against | **Mālik b. Anas** | *«دجال من الدجاجلة»* — and the tradition transmits the *cause* of the quarrel with it, which is itself part of the evidence |
| against | **Hishām b. ʿUrwa** | objected to his narrating from Fāṭima bint al-Mundhir |
| against | **Aḥmad b. Ḥanbal** | accepted in *maghāzī*, refused him alone in *sunan/aḥkām* — the domain rule applied to a person |
| against | **Ibn Ḥajar**, *Ṭabaqāt al-Mudallisīn* | placed among the *mudallisūn*; **NAMED-OPEN** |

**The mechanical point that decides how the layer must render him:** Ibn Isḥāq's own material, as preserved, is repeatedly introduced *without* a chain. This bench read it in the corpus in this sitting — `qadi_iyad`: `و ذكر ابن اسحاق عنها…`; `ibn_hisham`: `قال ابن اسحاق **وحدثت** أنه كان يقول…` (passive, source unnamed) and `**فيما بلغني**` (in the 111 Umm Jamīl passage above). ✅ VERIFIED in quoted text. **So a sīrah node carrying Ibn Isḥāq must display the chain-state — `musnad` / `mursal` / `balāghan` / `ḥuddithtu` / `ʿan man lā attahim` — as a first-class field, because the corpus states it and hiding it would be the waqf's own editorial act.**

### V.4 · Position C — al-Wāqidī (d. 207): matrūk in ḥadīth, cited in maghāzī
| Side | Carrier | Position |
|---|---|---|
| against | **al-Shāfiʿī** | *«كتب الواقدي كذب»* |
| against | **Aḥmad b. Ḥanbal** | *«كذاب»* / *«يقلّب الأحاديث»* |
| against | **Yaḥyā b. Maʿīn** | *«ليس بثقة»* |
| against | **al-Bukhārī**, **al-Nasāʾī**, **al-Dāraquṭnī** | *matrūk* / *yaḍaʿ al-ḥadīth* |
| against | **Ibn Ḥajar**, *Taqrīb al-Tahdhīb* | *«متروك مع سعة علمه»* — the verdict and its qualification in four words. **NAMED-OPEN** |
| for (in domain) | **al-Dhahabī**, *Siyar* / *Mīzān al-Iʿtidāl* | the dual verdict: weak in ḥadīth, an authority in maghāzī detail. **NAMED-OPEN** |
| for (in practice) | the corpus itself | **1,425 explicit `قال الواقدي` citations** ✅ VERIFIED — baladhuri 460 · ibnabihadid 470 · ibn_kathir 255 · tabari 164 · ibn_sayyid 20 · ibn_sad 7 · suhayli 7 · masudi 1 · qadi_iyad 1 · yaqubi 2 |

**Commendation, entered in the record:** `00-CORPUS/seerah.json → meta.works.waqidi.note` already reads *"CONTESTED RELIABILITY - graded weak by several ḥadīth critics, used by historians for detail. **Always cite with this stated.**"* ✅ VERIFIED. **The corpus builder got this right before any council sat.** The 1,425 measured citations are the hook that turns that note into a mechanical field: every one of them is a place where a later author knowingly relied on a matrūk transmitter, and the layer can mark it automatically.

### V.5 · Position D — the Imāmī apparatus, carried symmetrically
- The four-grade scheme (*ṣaḥīḥ / ḥasan / muwaththaq / ḍaʿīf*) is attributed to **al-ʿAllāma al-Ḥillī** and **Ibn Ṭāwūs** (7th/13th c.) — a *later* systematisation, so early akhbār are graded retrospectively. **This asymmetry with the Sunnī timeline must be stated, not smoothed.**
- **Akhbārī** (**al-Astarābādī**, **al-Fayḍ al-Kāshānī**, **al-Ḥurr al-ʿĀmilī**): the akhbār of the Four Books are *qaṭʿī al-ṣudūr* — chain-grading is not the operative question.
- **Uṣūlī** (**al-Ḥillī** onward; **al-Khūʾī**, *Muʿjam Rijāl al-Ḥadīth*): every chain graded individually. **NAMED-OPEN.**
- **The rule that inverts the direction — and the most important one for a Qurʾān waqf:** *العرض على الكتاب* — akhbār are presented against the Book and discarded where they contradict it. Carried in **al-Kāfī's own bāb** (**✅ HELD** — `00-CORPUS/hadith_shia.json`, al-Kāfī vols 1 & 8) and in the Sunnī corpus as the contested *ḥadīth al-ʿarḍ*. **Under this rule the Qurʾān is not one witness among four — it is the judge of witness #2 and #3.** The four-witness doctrine as written does not say this. It must, as a *transmitted* position, attributed, not adopted.

### V.6 · The six precedence rules the layer must CARRY (never apply)
1. **Domain rule** — aḥkām → graded ḥadīth prevails; qiṣaṣ/maghāzī/faḍāʾil → tasāhul permitted. *(al-Khaṭīb, al-Kifāya; Ibn Mahdī; Aḥmad. Imāmī mirror: tasāmuḥ fī adillat al-sunan.)*
2. **Aṣl rule** — a marfūʿ occasion-report with a ṣaḥīḥ chain in the Ṣaḥīḥayn outranks a mursal sīrah khabar. *(muṣṭalaḥ: musnad over mursal.)*
3. **Corroborated-mursal rule** — a mursal reinforced by an independent mursal of a different ṭabaqa rises to ḥasan li-ghayrih. *(al-Shāfiʿī, al-Risāla, with his four conditions; Ibn Ḥajar, Nuzhat al-Naẓar.)* **NAMED-OPEN.**
4. **Specialist rule** — on dates, routes, numbers and troop-lists the maghāzī specialists are cited *even when weak in ḥadīth*, with the weakness stated. *(al-Dhahabī's dual verdict on al-Wāqidī.)*
5. **Earliest-layer rule** — **Mālik**: *«مغازي موسى بن عقبة… أصحّ المغازي»*. **Mūsā b. ʿUqba's *Maghāzī* is NOT HELD** — only fragments survive; a named acquisition. *This rule currently points at a book the waqf does not have.*
6. **The waqf's own rule (the only one that is ours, and it is a rule of restraint)** — where a graded ḥadīth and a sīrah khabar diverge **in content**, the layer renders: both texts · both carriers · both chain-states · the grade where one exists and the word ABSENT where none does · **and the name of the transmitted precedence rule that the tradition would apply** — and then stops. It does not compute a winner. It does not rank. **A layer that outputs a verdict has left the waqf.**

---

## PART VI — MECHANICAL TABLE

| # | Finding | File / layer | Evidence | Label | Sev |
|---|---|---|---|---|---|
| F1 | Ocean = 47,083,188 chars, 17 works; brief's "~34M" = Sunnī file only (33,980,512) | `00-CORPUS/seerah*.json` | `len(text)` per work; delta vs meta = 0 on all 17 | ✅ VERIFIED | 🟢 |
| F2 | All 17 open/close coherently; `ibn_sad` ends mid-isnād, `yaqubi` ends in footnote apparatus | both files | head/tail 260 chars read on all 17 | ✅ VERIFIED | 🟡 |
| F3 | `ibn_ishaq` = Damascus fragment (`تمت قطعة دمشق من الكتاب`), 0.50% of ocean; **not** a Hishām duplicate | `seerah.json` | tail quoted; 8-gram overlap 3,133/44,203 = **7.1%** | ✅ VERIFIED · 📊 CALCULATED | 🟠 |
| F4 | Genre: maghāzī proper **6.5%**; universal history **56.5%**; commentary 17.2% | E3 / corpus | char shares, table in Part I | 📊 CALCULATED | 🟠 |
| F5 | Shīʿī block is **52.5%** Ibn Abī al-Ḥadīd, a Muʿtazilī — corpus meta says so, E3's label does not | `seerah_shia.json` | 6,878,673 / 13,102,676 | ✅ VERIFIED | 🟠 |
| F6 | HTML contamination: `mufid_irs` 2,371 tags, `tusi_amali` 126 | `seerah_shia.json` | regex count; specimen quoted | ✅ VERIFIED | 🟡 |
| F7 | **7,203 unused Qurʾān-quote delimiters** — 5,762 `@QB@` (Sunnī only) + 1,441 `﴿﴾` (Shīʿī only), open==close in every work | both files | per-work counts, Part II | ✅ VERIFIED | 🔴 |
| F8 | E3 recall vs that ground truth = **91.8%**; **233 delimiter-proven verses absent** | E3 | 2,834 resolved verses; 2,601 ∩ E3 | 📊 CALCULATED | 🟠 |
| F9 | `quran.json` prefixes the basmalah to āyah 1 of **112 sūrahs**, in `simple` **and** `uthmani`; 78 of them absent from E3 (69.6% vs 45.2% baseline) | `00-CORPUS/quran.json` | 3 specimens quoted | ✅ VERIFIED · 📊 CALCULATED | 🔴 |
| F10 | E2 and E3 use **different verse texts** — E2's 249 "full" 112:1 quotes are real (Abū Dāwūd #815/#1256 read), which proves E2 is basmalah-free and E3 is not | E2 vs E3 | both hadith quoted | ✅ VERIFIED | 🔴 |
| F11 | E3 blind to short verses: **77.8%** of ≤5-word verses absent vs 45.2% baseline; 130 of the 233 misses are ≤5 words | E3 | word-length profile | 📊 CALCULATED | 🟠 |
| F12 | **414 of the 2,821 "never quoted" verses are quoted** (denominator 1,968 probeable). Quoted ⩾3,829 · never-quoted ⩽2,407 | E3 · `THE-ANSWERS.md` §4 · E6 | uniqueness-rule replication over 8,876,213 words | 📊 CALCULATED | 🔴 |
| F13 | 8:5 · 33:37 · 111:1–5 — **all three carried**, passages quoted from 4/4/7 works | corpus | Part IV | ✅ VERIFIED | 🟢 |
| F14 | Sīrah has **55,007** isnād markers but **67** grading strings — the sīrah witness has chains and no grades in the held corpus | both files | Part V.1 | ✅ VERIFIED | 🟠 |
| F15 | **Neither shamāʾil nor dalāʾil founding text is held** — al-Tirmidhī's *Shamāʾil* absent, al-Bayhaqī's *Dalāʾil* absent | corpus enumeration | 17 works listed; neither present | ✅ VERIFIED | 🔴 |
| F16 | 1,425 `قال الواقدي` + 2,485 `قال ابن إسحاق` — the precedence register has a mechanical hook | both files | Part V.1 | ✅ VERIFIED | 🟢 |
| F17 | Chronology spine buildable from held text: 971 `ثم دخلت سنة` + ~7,747 dated expressions + 2,488 غزوة/سرية | both files | Part V.1 | ✅ VERIFIED | 🟢 |
| F18 | E2's top verses (112:1 · 87:1 · 1:3) are **recitation-reports**, not exegetical engagement — no class distinguishes them | E2 / E7 | Abū Dāwūd #815/#828 quoted | ✅ VERIFIED | 🟡 |

---

## PART VII — GAP-HIGHLIGHT DUTY: WHAT THE COMMISSION ITSELF MISSED

*SKILL law 13. Named plainly, including where this bench is the one that missed it.*

1. **7,203 quotation delimiters, unread by every layer.** The corpus has been telling us where the Qurʾān is since 2026-08-16 and no engine listened. (F7)
2. **The basmalah defect is upstream of everything.** It sits in `quran.json`, the most-read file in the waqf. E3 consumed it; E2 did not; **no layer declared which text it used.** Every count keyed to a sūrah-opening verse across all 26,400 nodes is now suspect until re-derived. (F9, F10)
3. **A wrong number is already published.** `THE-ANSWERS.md` §4 states "2,821 verses never quoted in the held sīrah". It is wrong by ≥414 and the correction should travel with the same prominence as the claim (I-3).
4. **The plan names a shamāʾil/dalāʾil bench for a genre the waqf does not hold.** Six of this bench's thirty seats found their primary sources absent. (F15)
5. **The 120,000-ḥadīth target is gameable.** Acquiring *Wasāʾil al-Shīʿa* (~35,850) would leap the counter while adding little matn — it is a topical recompilation of books already held. **Restate the target as 120,000 matn-distinct rows after skeleton deduplication**, or the number measures acquisition effort rather than knowledge.
6. **E6 conflates two silences.** `neither = 702` mixes "the witness has nothing to say" with "we do not hold the witness". These are opposite facts. A fifth state is needed. (COUNSEL Q3)
7. **Nobody has measured the sīrah's divergence *from itself*.** The doctrine keeps promising to locate divergence-of-content; every layer built so far measures only divergence-of-presence. Ibn Hishām alone gives **two different occasions for 111:1** — this bench found it by reading one passage. That class has never been counted. **This is the gap this bench most regrets not closing in its own sitting.**
8. **This bench did not verify the tail-completeness of `ibn_sad` or `baladhuri` against a reference edition.** `ibn_sad` ends mid-isnād at [vol 8 p.498] and `baladhuri` at [vol 4 p.332] where the standard Zakkār edition runs to 13 volumes. **Flagged as unresolved, not as a defect** — no reference edition was consulted, so no claim is made.
9. **No bench measured licence provenance on the sīrah files.** The meta asserts "Public-domain premodern works only" and names OpenITI version IDs, but `00-CORPUS/MANIFEST.sha256` was not verified against the live files in this sitting.

---

# COUNSEL — THE BENCH'S VOTES

*Thirty seats. Majority carries; dissents preserved verbatim.*

## Q1 — the split of the 3,600 free nodes (26,400 → 30,000)

### VOTE: **(b) — an amended split. 24 for · 6 dissent.**

| Allocation | Brief's (a) | **Bench 4 counsel** | Δ |
|---|---:|---:|---:|
| P28 Sīrah Universe 300 → 1,800 | +1,500 | **+1,500** | — |
| P32 new pillar | +1,800 | **+1,650** | −150 |
| Acquisitions & rijāl reserve | +300 | **+450** | +150 |
| **Total** | 3,600 | **3,600** | ✓ |

**Reasons, one line each:**
- **The magnitude of 1,500 for the sīrah is right and this bench does not contest it** — 47.08M chars carrying 3,829+ verse-engagements is under-served at 300 nodes by any measure.
- **But the sequence is wrong: the first 200 of those 1,500 must be a REPAIR block, not new domains** — E3 has a 414-verse hole, a corrupted verse text, and a 91.8% recall ceiling; 1,500 new nodes built on it would be 1,500 nodes carrying the defect.
- **+150 moved from P32 to acquisitions** because P32's most valuable domains (precedence register, grading gate, divergence bench) are **blocked on unheld books**, not on node count — Mūsā b. ʿUqba, al-Wāhidī, al-Dhahabī's *Mīzān*, Ibn Ḥajar's *Tahdhīb*, al-Khūʾī's *Muʿjam*, al-Tirmidhī's *Shamāʾil*, al-Bayhaqī's *Dalāʾil*. **Nodes cannot substitute for a book you do not hold.**
- **The reserve is not padding**; it is the only line item that can buy witness #3 a grading apparatus, which it presently has to the extent of 67 strings in 47 million characters.

**DISSENT (6 seats — 3 maghāzī, 3 sources-criticism):** *"Approve (a) unamended and take the repair block out of the standing engineering budget, not the expansion. A repair is a defect fix; charging it to the sīrah's growth allocation makes the sīrah pay for a bug in `quran.json` that it did not cause and that damages every pillar equally."* **This bench records that as the stronger argument on principle and the weaker one on delivery — recorded, not flattened.**

### The 18 domains of the 1,500-node Sīrah Universe
*Every domain marked **HELD** (buildable today) or **NAMED-OPEN** (buildable only after a named acquisition). No domain is proposed that is neither.*

**BLOCK 0 · REPAIR — 200 nodes — must close before any other block opens**
| # | Domain | Source state | Why |
|---|---|---|---|
| D1 | **Verse-normalization repair** — strip the basmalah from the 112 corrupted openings; publish ONE canonical verse-skeleton table; rebuild E3 from it; assert E2 and E3 consume the same table | **HELD** | recovers ≥414 verses at zero acquisition cost (F9, F12); closes the law-11 breach (F10) |
| D2 | **Delimiter harvest** — the 5,762 `@QB@` + 1,441 `﴿﴾` spans as a standing quotation registry, with a permanent precision/recall gate on the skeleton engine | **HELD** | turns F7 into the engine's KNOWN-POSITIVE selftest (SKILL law 5) |

**BLOCK 1 · CHRONOLOGY & EVENT — 300 nodes**
| # | Domain | Source state | Hook measured |
|---|---|---|---|
| D3 | **The annalistic spine** — 1 AH–11 AH expanded, each work's own dating carried side by side (they disagree; the disagreement is the node) | **HELD** | 971 `ثم دخلت سنة` (tabari 265 · ibn_kathir 706); ~7,747 dated expressions |
| D4 | **Maghāzī registry** — one node per expedition; participants, dates, numbers from each work in parallel columns | **HELD** | 2,488 `غزوة`/`سرية`; ~1,526 named expeditions |
| D5 | **The Makkan register** — the 13 years: the boycott, Ṭāʾif, al-Isrāʾ, the two Ḥabasha emigrations | **HELD** | Ibn Hishām, Ibn Saʿd, Balādhurī all carry it |
| D6 | **Documents & delegations** — the Ṣaḥīfa, Ḥudaybiyya, the wufūd, the letters to kings, as transmitted texts | **HELD** | Ibn Hishām and Ibn Saʿd carry the document bodies verbatim |

**BLOCK 2 · VERSE-SYNC (the lens itself) — 300 nodes**
| # | Domain | Source state | Hook measured |
|---|---|---|---|
| D7 | **Sīrah↔verse concordance v2** — one node per (verse × work), built on the repaired E3, genre-tagged per F4 | **HELD** | ≥3,829 verses; 47.08M chars |
| D8 | **The occasions bridge** — held seed now, full layer on acquisition | **HELD seed / NAMED-OPEN full** | 863 occasion formulae measured; needs **al-Wāḥidī, *Asbāb al-Nuzūl*** + **al-Suyūṭī, *Lubāb al-Nuqūl*** |
| D9 | **Revelation-order overlay** — cross D3's year spine against `rev_order`, already present per verse | **HELD** | `quran.json → ayat[*].rev_order` ✅ field verified present |
| D10 | **The occasion-conflict register** — where two works assign one verse to different events (Ibn Hishām gives 111:1 two occasions three pages apart) | **HELD** | derivable from D7; the gap named in Part VII §7 |

**BLOCK 3 · SOURCES-CRITICISM — 300 nodes**
| # | Domain | Source state | Hook measured |
|---|---|---|---|
| D11 | **The attribution graph** — who carries whom, work by work | **HELD** | 2,485 `قال ابن إسحاق` + 1,425 `قال الواقدي` |
| D12 | **Chain-state registry** — every khabar tagged `musnad`/`mursal`/`balāghan`/`ḥuddithtu`/`ʿan man lā attahim` | **HELD** | 55,007 isnād markers |
| D13 | **The precedence dossier** — Part V of this review, one node per transmitted position, each with its carrier | **HELD in part** (Ibn Sayyid al-Nās's muqaddima, 1,129,782 chars) **/ NAMED-OPEN** (al-Dhahabī *Mīzān*+*Siyar*, Ibn Ḥajar *Tahdhīb*, al-Khaṭīb *Kifāya*) |
| D14 | **Recension map** — Ibn Isḥāq through al-Bakkāʾī→Ibn Hishām vs the Damascus fragment (7.1% verbatim overlap measured) vs Yūnus b. Bukayr | **HELD in part / NAMED-OPEN** (Yūnus recension) |

**BLOCK 4 · SHAMĀʾIL & DALĀʾIL — 200 nodes — 🔴 DO NOT OPEN UNTIL ACQUIRED**
| # | Domain | Source state | Blocker |
|---|---|---|---|
| D15 | **Shamāʾil** — the described person: appearance, manner, habit | **HELD seed only** (Qāḍī ʿIyāḍ's *Shifāʾ* 733,882 chars + Ibn Saʿd's descriptive material) **/ founding text NAMED-OPEN** | **al-Tirmidhī, *al-Shamāʾil al-Muḥammadiyya* — NOT HELD** (F15) |
| D16 | **Dalāʾil** — the sign-reports, carried as transmitted with chain-state, never as argument | **NOT HELD AT ALL** | **al-Bayhaqī, *Dalāʾil al-Nubuwwa*; Abū Nuʿaym, *Dalāʾil* — NOT HELD** (F15) |

**BLOCK 5 · TRADITION-SYMMETRY — 200 nodes**
| # | Domain | Source state | Hook measured |
|---|---|---|---|
| D17 | **The Imāmī sīrah register** — *Irshād*, *Iʿlām al-Warā*, the two *Amālī* — **with the honest re-labelling of F5 applied** | **HELD** | 13,102,676 chars, of which Ibn Abī al-Ḥadīd is 52.5% and Muʿtazilī |
| D18 | **The divergence bench** — Sunnī/Shīʿī sīrah divergence-of-content on one event, both carried, neither ruled | **HELD** | 8:5 already shows it: Ibn Hishām's frame vs al-Mufīd's `أول حرب كان بها الامتحان` |

`200 + 300 + 300 + 300 + 200 + 200 = 1,500` ✓

---

## Q2 — P32: new pillar or an expansion of the Hadith Universe? And its name.

### VOTE: **NEW PILLAR — 30 for · 0 against. Unanimous.**
### NAME: **`P32 · THE CONCORDANCE OF WITNESSES` (al-Shahādāt al-Arbaʿ) — 19 for. Dissent: `THE FOUR-WITNESS PILLAR` — 11.**

**Reasons:**
- **Its object is not ḥadīth.** P32's object is the *relation between witnesses* — language↔ḥadīth↔sīrah↔cross-verse. Folding it into the Hadith Universe would make three witnesses subordinate features of the second one, and the pillar's name would then contradict its doctrine.
- **The witnesses are not co-equal in kind and the pillar must be free to say so** (see Q3): two are internal to the sealed text, two are transmitted. That distinction cannot live inside a pillar named for one of the transmitted ones.
- **Governance:** the ʿarḍ ʿalā al-Kitāb position (V.5) says the Qurʾān judges the khabar. A pillar housed inside the Hadith Universe cannot host a doctrine that subordinates ḥadīth without an obvious conflict of interest in the file tree.
- **On the name** — the majority holds that "four-witness" names the *method* and will age badly the moment a fifth witness is admitted (qirāʾāt variants and the tafsīr tradition are both plausible fifths); "Concordance of Witnesses" names the *object* and survives that. **The 11-seat dissent holds that "THE FOUR-WITNESS PILLAR" is what the owner already says out loud, that recognisability beats future-proofing, and that a doctrine named for four witnesses should be housed in a pillar named for four witnesses.** Recorded.
- **Bench 4 does not contest P32's internal domain list** — it lies outside our scope. We ask only for one addition: **a `sīrah-genre` field on every sīrah join** (F4), without which P32 will silently cite Ibn Kathīr on ʿAbbāsid politics as the "lived explanation" of a Makkan verse.

---

## Q3 — the four-witness doctrine wording

### VOTE: **APPROVE IN SUBSTANCE · AMEND IN FOUR PLACES — 27 for the amendments · 3 for approving as written.**

The doctrine's core — *"the waqf never adjudicates between witnesses; it makes every conflict VISIBLE with carriers named"* — this bench affirms without reservation. Four amendments, each forced by a measurement made in this sitting:

**AMENDMENT 1 — one clause in the present text is false of the held corpus.**
Present: *"(3) the SĪRAH (the lived explanation; itself transmitted through akhbār **whose grading conventions the tradition records**)"*.
The tradition does record them. **We do not hold them.** ✅ VERIFIED: 55,007 isnād markers, **67** grading strings, across 47,083,188 characters.
Proposed: *"(3) the SĪRAH — the lived explanation, transmitted through akhbār which in this held corpus carry **chains (55,007 isnād markers measured) but almost no grades (67)**. Grading for the sīrah witness is a NAMED ACQUISITION, not a held field. Until it is acquired, a sīrah node states its chain-state and the word ABSENT where a grade would go."*

**AMENDMENT 2 — type the witnesses. Two are internal, two are transmitted.**
Witnesses 1 (language) and 4 (cross-verse) are checkable *inside the sealed text* at no transmission cost. Witnesses 2 (ḥadīth) and 3 (sīrah) arrive through chains and carry a grading question. **A conflict between an internal and a transmitted witness is a different object from a conflict between two transmitted witnesses**, and the doctrine currently gives the reader no way to tell which one they are looking at.

**AMENDMENT 3 — type the silence. Five states, not four.**
A witness on a verse is in exactly one of: **(a) speaks and agrees · (b) speaks and differs · (c) silent — the witness genuinely has nothing · (d) silent — the held corpus lacks the witness · (e) unprobeable — the method cannot address this verse.**
E6 has (a) and a merged (c)+(d) called `neither = 702`. State (e) is not represented anywhere, and this bench measured it: **853 of the 2,821 target verses possess no verse-unique 4-gram at all** — they are invisible to the uniqueness rule by construction, and calling them "not quoted" is a category error. **Merging (c), (d) and (e) is the single most misleading thing the current layer does**, because (c) is a fact about revelation and (d) is a fact about our purchasing.

**AMENDMENT 4 — extend the visibility requirement, and name the doctrine that inverts the frame.**
Present: *"makes every conflict VISIBLE with carriers named."*
Proposed: *"…with **carriers named, chain-state named, grader named where one exists and ABSENT where none does, genre named, and the transmitted precedence rule that the tradition would apply named** — and then stops."*
And add, as a transmitted position and not as the waqf's adoption: *"The tradition also carries **al-ʿarḍ ʿalā al-Kitāb** (al-Kāfī's own bāb, ✅ held; and the contested Sunnī ḥadīth al-ʿarḍ) — under which the Qurʾān is not one witness among four but the judge of witnesses 2 and 3. The waqf transmits this position with attribution; it does not adopt it."*

---

## Q4 — the precedence doctrine when sīrah and graded ḥadīth diverge

### VOTE: **The layer carries the SIX RULES OF PART V.6 and the FOUR POSITION-DOSSIERS OF PART V.2–V.5, each with its named carrier, both sides where the tradition is split — and the layer issues NO ruling of its own. 30 for · 0 against. Unanimous.**

Restated in one screen, as the layer must render it:

| # | Transmitted rule | Carried by | Held? |
|---|---|---|---|
| 1 | **Domain rule** — aḥkām → graded ḥadīth prevails; qiṣaṣ/maghāzī/faḍāʾil → tasāhul permitted | al-Khaṭīb (*Kifāya*), Ibn Mahdī, al-Thawrī, Aḥmad · Imāmī mirror: al-Shahīd al-Thānī, al-Ḥurr al-ʿĀmilī | NAMED-OPEN |
| 2 | **Aṣl rule** — musnad ṣaḥīḥ occasion-report outranks mursal sīrah khabar | muṣṭalaḥ al-ḥadīth, standard | NAMED-OPEN |
| 3 | **Corroborated-mursal rule** — mursal + independent mursal of another ṭabaqa → ḥasan li-ghayrih | al-Shāfiʿī (*al-Risāla*, four conditions), Ibn Ḥajar (*Nuzha*) | NAMED-OPEN |
| 4 | **Specialist rule** — maghāzī specialists cited on dates/routes/numbers even when weak, weakness stated | al-Dhahabī's dual verdict on al-Wāqidī | NAMED-OPEN (practice ✅ held: 1,425 citations) |
| 5 | **Earliest-layer rule** — *«مغازي موسى بن عقبة أصحّ المغازي»* | Mālik | **points at a book NOT HELD** |
| 6 | **Genre caution** — *«ثلاثة ليس لها أصل: المغازي والملاحم والتفسير»*, **with both of the tradition's own readings carried** | Aḥmad, via al-Khaṭīb; Ibn Taymiyya | NAMED-OPEN |
| — | **The waqf's own rule, and the only one** | show both · name both carriers · name both chain-states · grade or ABSENT · name the applicable rule above · **stop** | ✅ enforceable today |

**Person-dossiers the layer must carry, both sides, never resolved:** Ibn Isḥāq (V.3 — Shuʿba's *amīr al-muʾminīn fī l-ḥadīth* beside Mālik's *dajjāl min al-dajājila*, with the quarrel's cause attached); al-Wāqidī (V.4 — al-Shāfiʿī/Aḥmad/Ibn Maʿīn/al-Bukhārī/al-Nasāʾī against, Ibn Ḥajar's *matrūk maʿa siʿat ʿilmih*, al-Dhahabī's dual verdict, and the 1,425 measured citations that show the tradition using him anyway); the Imāmī apparatus (V.5 — the retrospective four-grade scheme, and the Akhbārī/Uṣūlī split carried whole).

**Three operational consequences this bench asks be written into the layer, not left to convention:**
1. **`grade: ABSENT` is a value, never an empty field.** The sīrah witness will carry it 99.9% of the time and that must read as a *stated fact* about the held corpus, not as a rendering gap.
2. **`chain_state` is mandatory on every sīrah node** — the corpus supplies it (`فيما بلغني`, `حدثت`, `عن من لا أتهم` are Ibn Isḥāq's own words, verified in this sitting). Dropping it would be the waqf silently upgrading an anonymous report to a report.
3. **`precedence_rule_applicable` is a pointer, never a verdict** — it names which of the six rules a scholar would reach for. **If the field ever contains a winner, the pillar has broken its own charter.**

---

## Q5 — growing the ocean from the measured 68,921 toward 120,000

### VOTE: **Repair first at zero cost, then five acquisition waves in the order below, and the target restated as matn-distinct. 26 for · 4 dissent.**

**WAVE 0 — costs nothing, must precede everything.** Fix `quran.json`'s basmalah (F9); harvest the 7,203 delimiters (F7); re-derive E3 (F12). **This does not grow the ocean by one row and it is still the highest-value item on the list**, because every acquisition measured against a corrupted verse table inherits the corruption at scale.

| Wave | Acquisition | Approx. rows | Running | Why in this position |
|---|---|---:|---:|---|
| **1** | **Musnad Aḥmad b. Ḥanbal** | ~27,000 | ~95,900 | the single largest open jump; one book takes the ocean 39% of the way. **Take the matn only — Shuʿayb al-Arnaʾūṭ's gradings are IN COPYRIGHT; declare the grade field ABSENT rather than importing them** (grade-verbatim law) |
| **2** | **Tahdhīb al-Aḥkām (~13,590) + al-Istibṣār (~5,511)** | ~19,100 | ~115,000 | **the symmetry acquisition.** Today the Sunnī side holds all six + the Muwaṭṭaʾ while the Shīʿī side is missing 2 of its 4 canonical books. That is a fairness defect in the ocean itself, not a counting shortfall — and it is worth more than 19,100 rows |
| **3** | **Muṣannaf Ibn Abī Shayba (~37,000) + Muṣannaf ʿAbd al-Razzāq (~21,000) + Sunan al-Dārimī (~3,500)** | ~61,500 | ~176,500 raw | the muṣannaf layer is where *fiqh-of-the-verse* material lives — highest yield per row for the lens, and where deduplication first bites hard |
| **4** | **Ṣaḥīḥ Ibn Ḥibbān (~7,491) + Ṣaḥīḥ Ibn Khuzayma (~3,079) + al-Mustadrak (~8,800)** | ~19,400 | — | matters for the **CROWN tier**, not the counter: these extend the ṣaḥīḥ-graded stratum, which is the only stratum the strongest gate can use |
| **5** | **al-Muʿjam al-Kabīr (~18,000) · Wasāʾil al-Shīʿa (~35,850)** | ~53,900 | — | **large and duplicative.** Wasāʾil is a topical recompilation of books largely already held. **Dedupe by matn skeleton before counting or this wave will manufacture a false 120,000** |
| **S** | **The sīrah-side acquisitions — parallel track, funded from the +450 reserve** | — | — | **Mūsā b. ʿUqba, *Maghāzī*** (unblocks precedence rule 5) · **al-Wāḥidī, *Asbāb al-Nuzūl*** + **al-Suyūṭī, *Lubāb al-Nuqūl*** (unblocks D8) · **al-Tirmidhī, *al-Shamāʾil*** (unblocks D15) · **al-Bayhaqī + Abū Nuʿaym, *Dalāʾil*** (unblocks D16) · **al-Dhahabī, *Mīzān*+*Siyar*** and **Ibn Ḥajar, *Tahdhīb*+*Iṣāba*** and **al-Khūʾī, *Muʿjam Rijāl***, which together are the **only** things that can give witness #3 a grading apparatus (F14) |

**Reasons, one line each:**
- **Wave 1 before all else** because it is one acquisition, one licence check, one sha256, and 39% of the gap.
- **Wave 2 before Wave 3** even though Wave 3 is three times larger, **because an ocean that holds 7 Sunnī canonical books and 2 of 4 Shīʿī ones is not a symmetric witness**, and the four-witness doctrine's credibility rests on symmetry more than on size.
- **Wave 4 is ranked by tier value, not row count** — 19,400 rows that can enter the CROWN gate outrank 53,900 that cannot.
- **Wave 5 last and conditional** — it is the wave most likely to be acquired *because it moves the number*, which is exactly the reason to hold it until deduplication is enforced.
- **120,000 must be restated as 120,000 matn-distinct rows after skeleton deduplication.** As a raw row count the target is reachable by acquiring recompilations of books already held, and would then measure procurement rather than knowledge. 📊 On raw counts, Waves 1–4 alone reach ~176,500 — **the target is not the binding constraint; deduplication is.**

**DISSENT (4 seats — 2 chronology, 2 sīrah-tafsīr coupling):** *"Put Wave S first, ahead of Musnad Aḥmad. The ocean does not need to be bigger to be more useful; witness #3 currently has 67 grading strings in 47 million characters, and no quantity of new ḥadīth repairs that. Al-Wāḥidī and al-Dhahabī would raise the quality of every existing node; Musnad Aḥmad raises a counter."* **This bench records that dissent as the most serious one in this document** — the majority preferred Wave 1 first only because it is a single clean acquisition that can run in parallel with Wave S, not because it is more valuable per node.

---

## CLOSING — WHAT THIS BENCH WOULD ASK IF IT COULD ASK ONE THING

Not a number. This: **`THE-ANSWERS.md` §4 currently tells the owner that 2,821 verses of his Book are untouched by the sīrah. At least 414 of them are not — including the sūrah revealed about Abū Lahab, the fifth verse of the first revelation, and `قل هو الله أحد`. The correction should travel with the same prominence the claim did.** Everything else in this review can wait for a node budget. That cannot.

*Bench 4 rose at 02:20 AST, 2026-09-04. Thirty seats, four dissents preserved. Nothing in this document adjudicates between witnesses, condemns a transmitter, or rules where the tradition is split.*