# THE COUNCIL OF 900 — BENCH 1

**LABEL: MACHINE-SCHOLAR.** Every seat on this bench is a chartered machine-scholar seat. No seat
here adjudicates between schools, issues a ruling, or acts as a mufti. This bench **documents**.

| | |
|---|---|
| **Bench** | 1 of the Council of 900 |
| **Families** | **A · Quranic sciences** (28 areas, 126 seats) · **B · Hadith sciences** (20 areas, 89 seats) |
| **Seats** | 215 machine-scholar seats, doctorate-depth, each chartered as if carrying 1,400 years of its discipline's record |
| **Sat** | 2026-09-05 |
| **Read** | `build/P41-COUNCIL/council_900_charter.json` · `coverage_matrix_v1.json` · `preliminary_audit.json` · `framework/_FRAMEWORK-40000.json` (40,000 points) · `framework/p41_council.py` · P17/P18/P28/P30/P31/P32/P33/P34/P35/P36/P37/LENS-AUDIT/C3/C4/P8 artifacts · `00-CORPUS/acquired/*` |
| **Human verdicts on record** | **0** — `preliminary_audit.json:109`. This bench is machine; it does not pretend otherwise. |

---

## 0 · HOW THIS BENCH DECIDED A CELL

v1's law: *PRESENT = probe hits the canonical ledger's own domain text; TOUCHED = hits built-layer
text only; ABSENT = no hit.* This bench keeps that ladder and adds **three rules v1 lacked**, because
without them the ladder measures the probe, not the waqf:

1. **THE SELF-REFERENCE EXCLUSION.** `council_900_charter.json` and `coverage_matrix_v1.json` name
   all 313 areas. A hit inside those two files is the council reading its own handwriting and counts
   for nothing. (Measured: for 8 of this bench's 48 areas, the charter/matrix files are the *only*
   build-layer hits in the entire repository — `probe3`, files=3–4 each.)
2. **NO GENERIC-TOKEN CREDIT.** A multi-word discipline is not PRESENT because one generic token of
   its name (`hadith`, `tafsir`, `usul`, `quran`, `methodology`, `genre`, `transmission`) appears
   somewhere in 1,219,320 characters of domain text.
3. **THE ARTIFACT MUST CARRY THE DISCIPLINE'S OBJECT.** A layer named for a science but whose object
   is something else (a name-count called a narrator graph; a prefix-dedup called matn criticism; a
   word-count called manuscripts) is TOUCHED, not PRESENT. This is the waqf's own I-2 law turned on
   the waqf.

**This bench corrected itself once, in public, mid-sitting.** Its first count of cross-tradition
tafsir coverage read `editions_in_shard` and returned "2 of 146". That field is a dict in 108 files,
absent in 36, and a bare list in 2 (Finding 6). The recount, recovering the edition list from all
three fields, returned the same number — 2 of 146 — but the first number had been produced by an
unreliable instrument and was not yet evidence. The build's own law: *the audit that cannot find its
own errors cannot be trusted to find anyone else's.*

---

## 1 · MATRIX CORRECTIONS — 48 areas re-probed

**23 of 48 cells (48%) are corrected.** All ledger counts are points in `_FRAMEWORK-40000.json`
whose `pillar|domain|block|title|purpose` text matches a discipline-specific regex.

### FAMILY A · QURANIC SCIENCES (28 areas — 14 corrected)

| # | Area | v1 | **CORRECTED** | Evidence |
|---|---|---|---|---|
| A1 | Ulum al-Quran | PRESENT | **TOUCHED** ⬇ | v1 fired on `ulum` → matched **"curriculum** of understanding" (P30 D5w) and on `quran` → the single occurrence, **"QuranEnc"** (P35 E19). Zero ledger points name the umbrella science; no Itqān/Burhān held. Children built, parent not. |
| A2 | Usul al-tafsir | PRESENT | **TOUCHED** ⬇ | v1 fired on `usul` → "the **usul**i seat audits method" (C1). No usūl al-tafsīr register; the operative method is the waqf's own 4-pass consensus engine (`consensus_103_1.json:_meta`), not the discipline's. |
| A3 | Tafsir bil-mathur | PRESENT | **PRESENT** ✓ | Right cell, wrong reason: v1 fired on the bare token `tafsir`; `ma'thur` = 0 ledger. But P30 D-B1 marfūʿ + D-B2 mawqūf + B7 era-stratification *are* tafsīr bi-l-maʾthūr, named in their own terms. Thin — see F9. |
| A4 | Tafsir bil-ray (documented) | PRESENT | **TOUCHED** ⬇ | v1 fired on `tafsir` + `documented`. `ra'y` / `dirayah` = **0 hits in build/ + reviews/ + framework/ combined** (probe3). 8+ bi-l-raʾy works are read cell-by-cell; none is ever classified as such. |
| A5 | Ilm al-qiraat | PRESENT | **PRESENT** ✓ | 431 ledger points; `W1-L9/qiraat_layer.json` real. See F10. |
| A6 | Ilm al-tajwid | PRESENT | **PRESENT** ✓ | 400 points; W1-L1/L2/L3 madd, nūn-sākinah (46,129 madd positions), qalqalah — all measured from the script. |
| A7 | Ilm al-rasm | PRESENT | **PRESENT** ✓ | 631 points (`rasm` 331 + `hadhf` 300); rasm census live in W1-L9. |
| A8 | Add al-ay | PRESENT | **PRESENT** ✓ | 520 points; `W1-L10/counting_layer.json` real, three schemes. See F13. |
| A9 | Asbab al-nuzul | PRESENT | **PRESENT** ✓ | 100 points (P30 B3); `D-B3-OCCASIONS/occasions_layer.json` real. See F3. |
| A10 | Nasikh wa-mansukh | PRESENT | **PRESENT** ✓ | P31 W6-R6 named in the ledger; 515 marker hits / 371 verses. See F16. |
| A11 | Muhkam wa-mutashabih | PRESENT | **PRESENT** ✓ | P31 W6-R7 named; 327 hits / 264 verses, 3:7 anchored. |
| A12 | Makki-Madani | PRESENT | **PRESENT** ✓ | 442 points; P30 B4 sīrah-synchronisation places verses in the life ﷺ. |
| A13 | Ijaz al-Quran | PRESENT | **TOUCHED** ⬇ | `i'jaz` = **0** in the ledger. v1 fired on `quran` → **"QuranEnc"**. Build hits are 45, every one of them the mufassirūn's own remarks quoted inside P17 shards (18 in `consensus_36_1.json`). No iʿjāz layer exists. |
| A14 | Amthal al-Quran | PRESENT | **PRESENT** ✓ | P37 G16 similitudes register named; `graph_layers.json: mathal_similitude_verses = 70`. Thin — see F17. |
| A15 | Aqsam al-Quran (oaths) | PRESENT | **TOUCHED** ⬇ | v1 fired on `quran` → "QuranEnc". The 3 real ledger domains (`W4-S10 Shart & qasam`, `W5-T3 waw types`, `W-H qasam sub-census`) are **naḥw** — oath *syntax*. The science of aqsām (muqsam bihi / muqsam ʿalayhi / the oath's wisdom; al-Farrāʾ, Ibn al-Qayyim's *al-Tibyān*) has no artifact. |
| A16 | Gharib al-Quran | PRESENT | **PRESENT** ✓ | P31 W3-X1 `gharib_register.json`; rarity measured by lemma. See F18. |
| A17 | Wujuh wa-nazair | PRESENT | **TOUCHED** ⬇ | Ledger names W3-X2, but the artifact is `wujuh_seed.json` whose own law reads *"the alignment program … REGISTERED, not faked"*. No classical wujūh list (al-Dāmaghānī, Ibn al-Jawzī) is held. A declared seed is TOUCHED. |
| A18 | Ilm al-munasabat | ABSENT | **ABSENT** ✓ | Confirmed both ways: `munasab`/`nazm` = 0 ledger; build hits = charter/matrix + 12 `nazm` occurrences inside P17 shard prose. A **CROWN** area (5 seats) with no artifact. See F2. |
| A19 | Waqf wa-ibtida | PRESENT | **PRESENT** ✓ | `W1-L4/waqf_signs_layer.json` — 4,573 marks in 8 classes, two editions reconciled three ways, differences declared. The strongest layer this bench measured. See F14. |
| A20 | Fadail al-Quran | PRESENT | **PRESENT** ✓ | P31 W6-R8 named; `fadail_layer.json` real, grades verbatim. See F15. |
| A21 | History of jam and tartib | TOUCHED | **TOUCHED** ✓ | `tartib` = 0 ledger; build hits are P17 shard prose + `al_ihsan_register.json`. The jamʿ narrative (Abū Bakr → ʿUthmān → Zayd; the ʿarḍa akhīra) has no register. |
| A22 | Mushaf codicology & manuscripts | PRESENT | **TOUCHED** ⬇ | The 100 `manuscript` points are one C3 domain **literally titled "The text & its manuscripts"** — whose artifact, `C3-QURAN/D1-TEXT/text_integrity_layer.json`, holds `texts_held: ["uthmani","simple-clean"]` and word/letter counts. **Zero manuscripts.** No folio, codex, dating, palimpsest. See F5. |
| A23 | Tartil pedagogy & orthoepy | PRESENT | **TOUCHED** ⬇ | v1 fired on `pedagogy` → "the **pedagogy** bench attacks clarity" (P30 A8). Real ledger hits: 13, all via `memoriz`. Build's 93 `tartil` hits: 59 are in `consensus_73_4.json` — the tafsīr of 73:4 *wa-rattil il-Qurʾāna tartīlā*. No tartīl pedagogy artifact. |
| A24 | Tadabbur methodology | PRESENT | **TOUCHED** ⬇ | v1 fired on `methodology` → "P8 wisdom & **methodology**". `tadabbur` = 0 ledger. A genuine adjacent artifact exists — `C1-ALLAH/D7-KNOWING-HIM/knowing_layer.json` (52 hits) — which is exactly TOUCHED. A **CROWN** area. |
| A25 | Thematic tafsir (mawdui) | PRESENT | **TOUCHED** ⬇ | v1 fired on the bare adjective `thematic` (564 points). The artifact: `P37-GRAPH/graph_layers.json → thematic_tier: {"status":"OPEN - scholars' intake", "rows": []}`. **Declared, zero rows.** See F8. |
| A26 | Comparative tafsir (cross-tradition) | PRESENT | **PRESENT** ✓ | 213 points on `cross-tradition`; 34 distinct editions across both traditions read cell-by-cell under a cluster+dedup law. Real. See F6. |
| A27 | History of Quran translation | PRESENT | **TOUCHED** ⬇ | 1,120 points on the bare token `translation`. `P35-EDITIONS/editions_inventory.json` = **74 held modern editions, 55 languages, one aggregator (QuranEnc)**. That is an inventory of the present, not a history. Ketton (1143) → Marracci → Sale → Pickthall → the Urdu line: absent. See F19. |
| A28 | Digital Quranic corpora studies | PRESENT | **PRESENT** ✓ | 1,384 points; Tanzil, KFGQPC, OpenITI, the ʿirāb treebank, all with sha-sealed `_META.json`. Real and strong. |

### FAMILY B · HADITH SCIENCES (20 areas — 9 corrected)

| # | Area | v1 | **CORRECTED** | Evidence |
|---|---|---|---|---|
| B1 | Mustalah al-hadith | TOUCHED | **PRESENT** ⬆ | v1's alias probe (`mustalah`, `hadith terminolog`) missed the layers built in the science's own vocabulary: 350 ledger points on `marfu'`/`mawquf`, realised as `D-B1-MARFU` and `D-B2-MAWQUF`, plus the grade-verbatim law and the T1/T2/T5s tier ladder. **Upgrade.** See F11. |
| B2 | Ilm al-rijal (Sunni) | PRESENT | **PRESENT** ✓ | 1,360 real `rijal` points; Ibn Ḥibbān's Thiqāt + Majrūḥīn acquired and parsed. **Warning: the 100 `tahdhib` points are *Tahdhīb al-Aḥkām* (al-Ṭūsī, Shīʿī fiqh-hadith), not *Tahdhīb al-Tahdhīb*** — a name-join, not an ID-join. See F4. |
| B3 | Ilm al-rijal (Shii) | PRESENT | **PRESENT** ✓ | Kashshī's Ikhtiyār (1,147 blocks, numbered), Ṭūsī's Rijāl + Fihrist, Ibn al-Ghaḍāʾirī's Ḍuʿafāʾ (226). Real. Note the 100 `istibsar` points are the *hadith* book, not rijāl. See F4, F12. |
| B4 | Ilm al-tabaqat | PRESENT | **TOUCHED** ⬇ | Ledger names P33 R9 "the tabaqat scaffold", but the artifact is a *token count* — `"min ashab (tabaqah note)": 79 / 51` inside `rijal_registries.json`. No ṭabaqāt register, no Ibn Saʿd. |
| B5 | Jarh wa-tadil | PRESENT | **PRESENT** ✓ | P33 R7 term census, verbatim, per work (thiqah 40/173, kadhdhāb 38, majhūl 45 …). Real — but never applied to a person. See F7. |
| B6 | **Ilal al-hadith** | PRESENT | **ABSENT** ⬇⬇ | v1 fired on `ilal` → **"H*ilal*i-Khan's brackets"** (P30 C4w — the *translator*) and on the bare token `hadith`. Ledger: 0. Build: **3 files, all charter/matrix self-reference.** No ʿIlal al-Dāraquṭnī / Ibn Abī Ḥātim / al-Tirmidhī. A **CROWN** area (5 seats) at zero. |
| B7 | **Gharib al-hadith** | PRESENT | **ABSENT** ⬇⬇ | v1 fired on `gharib` — which matched the *Qurʾānic* gharīb register (W3-X1) — plus `hadith`. Ledger: 0. Build: 3 files, all self-reference. No *al-Nihāya* (Ibn al-Athīr), no Abū ʿUbayd, no al-Khaṭṭābī. |
| B8 | Mukhtalif al-hadith | PRESENT | **TOUCHED** ⬇ | v1 fired on bare `hadith`; my own first probe's 170 hits were `reconcil`**iation of counts** — a false positive I withdraw. Real adjacency: P28-X5 `precedence_register.json` (schema + 3 seeded positions, **`rows: []`**, all `NEEDS-VERBATIM-RESOURCING`) and P36 B6. No Ibn Qutayba / al-Ṭaḥāwī. |
| B9 | **Nasikh in hadith** | PRESENT | **ABSENT** ⬇⬇ | v1 fired on `nasikh` — which matched the *Qurʾānic* W6-R6 — plus `hadith`. Build: **0 files, 0 hits**, not even self-reference. No al-Ḥāzimī's *Iʿtibār*, no Ibn Shāhīn. |
| B10 | Takhrij method | ABSENT | **TOUCHED** ⬆ | This bench upgrades v1. `takhrij` is nowhere, but `P34-OCEAN/matn_clusters.json` produces takhrīj *output* — 5,574 clusters locating one report across collections (e.g. `bukhari 1253 / abudawud 3142 / ibnmajah 1458 / malik 521`). The function exists unnamed. A **CROWN** area. |
| B11 | Isnad analysis | PRESENT | **TOUCHED** ⬇ | 410 real ledger points, but the object is never touched: `narrator_census_layer.json` parses **the first narrator only**, its own law reads *"the full isnād graph is AWAITED as its own pass"*, and P32 W-K stage 3 (`tabaqat/riwayah-paths`) is declared *"needs isnad parsing"* — undone. A **CROWN** area. See F1, F20. |
| B12 | Matn criticism (documented) | PRESENT | **TOUCHED** ⬇ | 700 real points, but `matn_dedup_layer.json` is a **first-80-character skeleton + length band** near-duplicate detector. Naqd al-matn — testing a text against Qurʾān, mutawātir sunna, history, and parallel wordings for idrāj/qalb/iḍṭirāb — is not built. A **CROWN** area. See F21. |
| B13 | Mawduat (fabrication registers) | PRESENT | **TOUCHED** ⬇ | v1 fired on `fabrication`, which matched exactly one domain: W6-R8's *"fabrication warnings surfaced (the Ubayy forgery pattern)"* — inside the faḍāʾil layer. `mawdu'at` = 0. No Ibn al-Jawzī, no al-Shawkānī, no al-Laʾāliʾ. There is no fabrication register. |
| B14 | Collection-genre studies | PRESENT | **PRESENT** ✓ | Genre-honest census is real and explicit: `THE-ANSWERS.md:§1` — 68,646 narrations + 226 rijāl entries + 49 treatise sections, Nahj al-Balāgha labelled as khuṭab under Imāmī usage. Exemplary. |
| B15 | **Arbainat genre** | PRESENT | **ABSENT** ⬇⬇ | v1 fired on the bare token **`genre`** → "the domain's own **genre** literature". `arba'in` = 0 ledger. Build: 4 files — charter, matrix, `p41_council.py`, and 4 incidental hits in `consensus_18_29.json`. No arbaʿīn work held. |
| B16 | **Shuruh al-hadith** | PRESENT | **ABSENT** ⬇⬇ | v1 fired on `hadith` + `commentary`. `shuruh`/`sharh` = 0 ledger. Build: 20 files, but 19 are single bibliographic mentions in `SOURCES.md` lists. **No Fatḥ al-Bārī, no Sharḥ al-Nawawī, no ʿUmdat al-Qārī, no Tuḥfat al-Aḥwadhī.** The commentary tradition on the Sunna is unheld. |
| B17 | Shamail studies | TOUCHED | **PRESENT** ⬆ | v1's probe (`shamail`) missed a 400-point ledger presence (`shama'il` 300 + `noble character` 100 across P28/C4) and three acquired works: `wave56/Shamail.txt`, `wave7/MukhtasarShamail.txt`, `wave7/IbnHibbanSira.txt`, plus 2,633 candidates in `shamail_candidates.json`. **Upgrade.** See F22. |
| B18 | Fiqh al-hadith | PRESENT | **TOUCHED** ⬇ | v1 fired on `fiqh` + `hadith`, both bare. Real adjacency: `P36-BENCH/verdict_intake.json` carries a `legal-inference` verdict class — and `"verdicts": []`. Infrastructure, no fiqh. See F23. |
| B19 | Transmission history (ijazah & sama) | PRESENT | **TOUCHED** ⬇ | v1 fired on `transmission` → "P12 **transmission** — the second link". `ijazah` = 0, `sama'` = 0 in the ledger. Real adjacency: Ṭūsī's **Fihrist** is held and parsed (1,338 blocks) — a fihrist *is* a transmission-licence register. No thabat / fahrasa / muʿjam al-shuyūkh; no ijāza or samāʿ concept anywhere in 40,000 points. |
| B20 | Digital hadith corpora studies | PRESENT | **PRESENT** ✓ | 422 real points; 119,261-row ocean, 39+ books, per-wave sha-sealed `_META.json`. Real and strong. |

### CORRECTIONS SUMMARY (this bench's 48 areas)

| | v1 | Bench 1 | Δ |
|---|---|---|---|
| PRESENT | 43 | **21** | **−22** |
| TOUCHED | 3 | **21** | +18 |
| ABSENT | 2 | **6** | +4 |
| Cells changed | — | **23 of 48 (48%)** | |

**Direction of the error is one-way: v1 over-reports.** 20 of 23 corrections are downgrades. The
cause is diagnosed in Finding 2, not guessed at.

---

## 2 · THE FINDINGS — 24, worst first

**CENTRE-CLAUSE key:** **(a)** the Qurʾān as centre of research · **(b)** everything returns to
Allah ﷻ · **(c)** meaning derived through sīrah / sunnah / hadith · **POLISH** = craft, not centre.

| # | Finding | Area(s) | Cell | Sev | Centre | THE FILL |
|---|---|---|---|---|---|---|
| **F1** | **A child-facing artifact states falsehoods about the Qurʾān.** `P18-HADITH/D1-NARRATOR-GRAPH/education_pack.json → children_layer` reads verbatim: *"Abu Hurairah - 1241 attribution lines appears in the Book 1 time(s)."* Abū Hurayra ؓ does not appear in the Book. The "1 time" is the layer's own anchor verse 49:6. The same pack lists him **twice as two people** ("Abu Hurairah - 1241", "Abu Huraira - 976"). `P8-SAHABA/D3-NARRATORS` repeats it for 12 Companions; `C4-MUHAMMAD/D3-SHAMAIL` says *"manner - 647 reports appears in the Book 647 time(s)"* — an English category word asserted to occur in the Qurʾān. **Measured: 10 education packs carry count-suffixed pseudo-names in `children_layer`.** | Isnad analysis · Ilm al-rijal (Sunni) · Shamail | TOUCHED | **HIGH** | **(a)** | **Engine.** A `children_layer` gate that refuses any item whose subject is not a Qurʾānic lexeme, and refuses "appears in the Book N times" unless N is a verified `quran.json` occurrence count. Then rebuild all 10 packs. Zero new nodes; this is a repair. |
| **F2** | **The coverage matrix's PRESENT column is produced by an instrument that cannot fail.** `framework/p41_council.py:181-189` splits an area name into tokens, keeps the first 3, and marks PRESENT if **any one** is a substring of the domain text. Consequences, each replayed and printed: `Ijaz al-Quran` → PRESENT on `quran`, whose **sole occurrence in 1,219,320 chars is "QuranEnc"**. `Ulum al-Quran` → PRESENT on `ulum` ⊂ "**curriculum**". `Ilal al-hadith` → PRESENT on `ilal` ⊂ "H**ilal**i-Khan". `Arbainat genre` → PRESENT on `genre`. Six B-family areas → PRESENT on the bare token `hadith`. | all 48 | — | **HIGH** | **POLISH** (but it hides (a)/(c) gaps) | **Engine.** matrix v2: word-boundary regex, per-area alias lists, **AND**-logic on multi-word names, the self-reference exclusion, and a **negative control** (probe 20 invented area names; any hit fails the build). Then re-run all 20 benches' corrections through it. |
| **F3** | **Asbāb al-nuzūl — the discipline that binds an occasion to a verse — carries 54 rows with no verse.** `P30-SIGHAH/D-B3-OCCASIONS/occasions_layer.json`: 414 rows, of which **54 (13%) have a bare surah number as the `verse` key** (`"73"`, `"96"`, `"1"`) instead of `surah:ayah`. A further **135 rows (33%) carry `verse_binding: "claimed-requires-check"`** — unverified. Distinct verse keys across the whole layer: **120 of 6,236 (1.9%)**. No classical asbāb work (al-Wāḥidī, al-Suyūṭī's *Lubāb al-Nuqūl*) is acquired; occasions are marker-detected from tafsīr text. | Asbab al-nuzul | PRESENT | **HIGH** | **(c)** | **Acquisition + engine + nodes.** Acquire al-Wāḥidī's *Asbāb al-Nuzūl* and al-Suyūṭī's *Lubāb al-Nuqūl* (public domain; OpenITI carries both). Then an engine that resolves every surah-only key or drops it declared, and clears the 135 `claimed-requires-check` rows. ~300 nodes under P30 wing B. |
| **F4** | **Two ledger tokens are name-joins, not ID-joins — the exact error I-1 forbids.** The 100 ledger points reading `Tahdhib` are **Tahdhīb al-Aḥkām** (al-Ṭūsī, a Shīʿī fiqh-hadith collection, `acquired/shia_wave1/TahdhibAhkam.txt`), yet the string is what makes "Ilm al-rijal (Sunni)" look furnished; the actual *Tahdhīb al-Tahdhīb* is not held. Likewise `Istibsar` (a hadith book) sits under "Ilm al-rijal (Shii)". Verified: `ibn hajar` = 0, `dhahabi` = 0, `mizzi` = 0, `mizan al-` = 0 across all 40,000 points. | Ilm al-rijal (Sunni/Shii) | PRESENT | **HIGH** | **POLISH** | **Acquisition.** *Taqrīb al-Tahdhīb* (Ibn Ḥajar) is already the named acquisition in `P32-SHAHADAT/W-K/rijal_program.json` stage 2 — fund it, or a public-domain equivalent (*al-Tārīkh al-Kabīr*, *al-Jarḥ wa-l-Taʿdīl*, *Mīzān al-Iʿtidāl*). Until then, stamp both tokens `NOT-A-RIJAL-WORK` in the ledger. |
| **F5** | **A domain titled "The text & its manuscripts" holds no manuscript.** `C3-QURAN/D1-TEXT/text_integrity_layer.json`: `texts_held: ["uthmani","simple-clean"]`, `translations_held: [5]`, and word/letter counts. 100 ledger points ride on the word `manuscript`. No folio, no codex, no dating, no palimpsest — no Ṣanʿāʾ, Birmingham, Topkapi, Samarqand. The waqf's own I-2 law: a variable name is not its meaning. | Mushaf codicology | TOUCHED ⬇ | **HIGH** | **(a)** | **Acquisition + nodes.** Open manuscript metadata is free (Corpus Coranicum, the Birmingham/Mingana catalogue, Cambridge & BnF IIIF records). ~200 nodes under C3: a codex register (shelfmark, script, folios, dating basis, digitisation licence) — **or** rename the domain to what it holds. Renaming is honest and costs nothing. |
| **F6** | **The consensus shelf — the waqf's meaning layer — has three incompatible schemas, and its cross-tradition axis is 1.4%.** Across 146 shards: `editions_in_shard` is a dict in **108**, absent in **36**, a bare list in **2**; `editions_read` is a list in **110**, an **integer** in **35**, absent in 1. **35 of 146 (24%) expose no recoverable edition list at all.** Recovering across all three fields: **2 of 146 shards (1.4%) name any Shīʿī tafsīr edition**, against a stated law of *"one ummah, both corpora"*. 34 editions exist in the shelf, including al-Mīzān, al-Tibyān, Majmaʿ al-Bayān, Nūr al-Thaqalayn — read in two shards only. | Comparative tafsir · Digital corpora | PRESENT | **HIGH** | **(c)** | **Engine + nodes.** One schema, one migration, one gate that refuses a shard whose edition list is not a `{sunni:[],shia:[]}` dict. Then ~400 nodes extending the shelf with the 5 held Shīʿī editions across the existing 146 verses first — coverage before expansion. |
| **F7** | **Zero narrators are identified anywhere in 40,000 points.** `P33-RIJAL/rijal_registries.json → cross_work_name_strings: {"count": 0}`, with its own note *"name-strings never merged into persons"*. The five seed works are parsed as text blocks; the jarḥ/taʿdīl vocabulary is censused (thiqah 40/173, kadhdhāb 38, majhūl 45…) but **never attached to a person**. Ilm al-rijāl is the science of identifying people; the waqf holds its vocabulary and none of its persons. | Ilm al-rijal ×2 · Jarh wa-tadil · Ilm al-tabaqat | PRESENT/TOUCHED | **HIGH** | **(c)** | **Engine + HUMAN seat.** Per-work entry-head parsers (declared as "the bench's refinement" and undone) → a person register with a stable ID space → the jarḥ verdicts joined by ID. The **identification** step needs a **human rijāl scholar** — the build says so itself and is right to. ~500 nodes P33. |
| **F8** | **Two flagship tiers are declared and empty.** `P37-GRAPH/graph_layers.json → thematic_tier: {"status":"OPEN - scholars' intake","rows":[]}` (thematic tafsīr). `P28-SIRAH-X/X5/precedence_register.json → rows: []`, 3 seeded positions all `NEEDS-VERBATIM-RESOURCING` (the Q4 ruling, carried 120–0). `P36-BENCH/verdict_intake.json → verdicts: []`. Declared-and-empty is honest; it is not built. | Thematic tafsir · Mukhtalif al-hadith · Fiqh al-hadith | TOUCHED ⬇ | **HIGH** | **(c)** | **HUMAN seats — the binding constraint.** These three tiers are *designed* to be filled by scholars, and the design is correct. No engine fills them. The fill is: 3 named human seats (a mufassir for the thematic tier; a muḥaddith for precedence; a faqīh for the verdict intake) and a first sitting. |
| **F9** | **The centre gap, restated from this bench's vantage.** The waqf's law is *"meaning derived through sīrah / sunnah / hadith."* The **evidence** layers reach 59.4–80% of the Book. The **meaning** layers built through that lens: consensus shelf **146 verses (2.3%)**, Prophetic-stratum dossiers **90 verses (1.4%)**, and the marfūʿ explanation layer — the tafsīr bi-l-maʾthūr core — is **48 entries** (`D-B1-MARFU/marfu_layer.json`), 5 of them deferred. `preliminary_audit.json:104` states this; this bench confirms it independently and adds the 48. | Tafsir bil-mathur · Usul al-tafsir · Mustalah | PRESENT | **HIGH** | **(c)** | **Nodes at scale + human seats.** This is the largest single fill in the waqf and cannot be engineered away: the D-B1 method is *"verbatim-verified AND integrator content-read"* — a human reads every row. Realistic path: ~2,000 nodes carrying a staged dossier programme (90 → 500 → 1,500 verses), each stage gated on human review. |
| **F10** | **ʿIlm al-qirāʾāt holds 3 of 20 canonical riwāyāt.** `W1-L9/qiraat_layer.json` holds Ḥafṣ (ʿĀṣim), Warsh and Qālūn (Nāfiʿ) — **2 of the 10 imāms**. Missing: Ibn Kathīr, Abū ʿAmr, Ibn ʿĀmir, Ḥamza, al-Kisāʾī, Abū Jaʿfar, Yaʿqūb, Khalaf. The comparison is a **whole-surah rasm skeleton** (14 surahs identical, 100 differ); the layer's own law concedes *"vowel-level riwāyah variation (the majority) awaits a finer engine"*. **Farsh al-ḥurūf — the substance of the science — is not built**, and no qirāʾāt source text is held (*al-Shāṭibiyya*, *al-Taysīr*, *al-Nashr*, *al-Durra*: none). A **CROWN** area. | Ilm al-qiraat | PRESENT | **HIGH** | **(a)** | **Acquisition + engine + nodes.** KFGQPC publishes further riwāyāt; *al-Nashr* and *al-Taysīr* are public domain. Then a **word-level farsh engine** (the layer already names it as the next block). ~600 nodes under P31 W1. |
| **F11** | **The isnād-defect half of muṣṭalaḥ is at absolute zero.** Swept across build/ + framework/ + reviews/: **`mudallas` 0 · `tadlis` 0 · `munqati'` 0 · `mu'dal` 0 · `mudtarib` 0 · `maqlub` 0 · `mudraj` 0 · `mu'allaq` 0.** Only the attribution-level labels are carried (`marfu` 30,709 · `mawquf` 31,118 · `maqtu` 129 · `mursal` 104), and those come from the era-stratification engine, not a muṣṭalaḥ register. No founding text: `ibn al-salah` 0, `nukhbat` 0, `tadrib` 0. The waqf grades reports without holding the grammar of grading. | Mustalah al-hadith | PRESENT ⬆ | **HIGH** | **(c)** | **Acquisition + nodes.** *Muqaddimat Ibn al-Ṣalāḥ* and *Nukhbat al-Fikar* + its *Nuzha* are public domain and short. ~250 nodes: the full muṣṭalaḥ taxonomy as a register, each term with its transmitted definition and carriers, joined to the tier ladder that already exists. |
| **F12** | **Al-Ṭūsī's *Rijāl* — ~6,400 entries — is parsed as 69 text blocks.** `rijal_registries.json`: `Tusi Rijal: text_blocks_parsed 69, sections 313`. Ibn Ḥibbān's *Thiqāt* (13,000+ entries) → 973 blocks; the build's own law states the proxy runs *"~17× UNDER"* for Thiqāt and *"~2× OVER"* for Majrūḥīn — **the error inverts between works**, so no count is comparable. Separately: the narrator census covers only `hadith.json` (7 Sunnī collections, 36,390 reports). **There is no narrator census over the 32,531 Shīʿī reports at all.** | Ilm al-rijal ×2 | PRESENT | **HIGH** | **POLISH** | **Engine.** Per-work roll parsers (Ṭūsī's rolls are line-coded `(ع)` entries — the build names this and did not do it). Plus a Shīʿī-side narrator census mirroring the Sunnī one. ~300 nodes P33. |
| **F13** | **ʿAdd al-āy: 3 schemes, and the "Madanī" column is a printing house.** `W1-L10/counting_layer.json` carries Kūfī 6,236, schemeB 6,348, and `madani_measured (KFGQPC Warsh)` 6,214 — a count read off **one Warsh edition's verse numbering**, not from the ʿadd literature. It lands on 6,214 = Madanī **al-akhīr**, but the tradition has **two** Madanī schools (awwal 6,217 / akhīr 6,214) and the column distinguishes neither. Makkī (6,210), Baṣrī (6,204), Shāmī (6,225) are absent. The layer declares it: *"the classical six-school table awaits its documented source (AWAITED)"*. A **CROWN** area. | Add al-ay | PRESENT | **MED** | **(a)** | **Acquisition.** Al-Dānī's *al-Bayān fī ʿadd āy al-Qurʾān* (public domain) supplies the six-school table directly. ~150 nodes: one register, six columns, per surah, each sourced. |
| **F14** | **Waqf wa-ibtidāʾ is built without the ibtidāʾ.** `W1-L4/waqf_signs_layer.json` is the best-engineered layer this bench opened — 4,573 marks in 8 classes, two pause-marking editions reconciled three ways, every difference declared, none merged. But the classes held are a **printed-muṣḥaf sign set** (ṣlī, qlī, jāʾiz, lā, lāzim, muʿānaqa), not the discipline's own taxonomy (**tāmm / kāfī / ḥasan / qabīḥ**), and **the ibtidāʾ half — where one may resume — has no artifact**. No al-Sajāwandī's *ʿIlal al-Wuqūf*, no al-Ushmūnī, no al-Dānī's *al-Muktafā*. | Waqf wa-ibtida | PRESENT | **MED** | **(a)** | **Acquisition + nodes.** *ʿIlal al-Wuqūf* and *al-Muktafā* are public domain. ~200 nodes under P31 W1: the tāmm/kāfī/ḥasan/qabīḥ register joined to the 4,573 marks already positioned, plus an ibtidāʾ register. |
| **F15** | **Faḍāʾil al-Qurʾān is a register of 44 reports found by one regex.** `W6-R8/fadail_layer.json`: 44 reports (16 Sunnī / 28 Shīʿī), 43 of them matched by the single formula `man qara'a + surah within 120 chars`. The waqf's own ocean holds far more: `THE-ANSWERS.md:§2` reports **249 quote-pairs on 112:1 alone, "mostly recitation reports — the faḍāʾil truth of al-Ikhlāṣ"**. The register captured 44 of what the corpus already carries. | Fadail al-Quran | PRESENT | **MED** | **(c)** | **Engine.** Widen the detector to the genre's real formulae (*man qaraʾa / fī faḍl / thawāb / man ṣallā bihā / khayr min*) and reconcile against the 249 already-found pairs as a known-positive. Zero acquisition needed. ~100 nodes. |
| **F16** | **Nāsikh wa-mansūkh is 515 candidates with a declared ~40% false-positive rate, none read.** `W6-R6/R6_naskh_candidates.json`: 515 marker hits over 371 verses; the layer's own law records *"PB1 panel note: ~40% sampled FP (mansūkha in grammatical-convention contexts)"* and *"every row is a CANDIDATE until read"*. Nobody has read them. No naskh work is held (al-Naḥḥās, Ibn Salāma, Ibn al-Jawzī, Hibat Allāh). Same shape at W6-R7 (327 muḥkam/mutashābih candidates, `al-muhkam`-as-attributive declared as a known FP class). | Nasikh wa-mansukh · Muhkam wa-mutashabih | PRESENT | **MED** | **(a)** | **Acquisition + HUMAN seat.** Al-Naḥḥās's *al-Nāsikh wa-l-Mansūkh* (public domain) turns candidate-detection into a transmitted list with named carriers. The label-verification pass needs a human reader. ~150 nodes. |
| **F17** | **Amthāl al-Qurʾān is a mathal-formula regex returning 70 verses.** `P37-GRAPH/graph_layers.json: mathal_similitude_verses = 70`. The discipline's classes — **amthāl muṣarraḥa / kāmina / mursala** — do not exist in the build; only the explicit *mathal* formula is caught, so every implicit and unstated similitude is invisible. No classical amthāl work (al-Māwardī, Ibn al-Qayyim, al-Ḥakīm al-Tirmidhī). | Amthal al-Quran | PRESENT | **MED** | **(a)** | **Acquisition + nodes.** Ibn al-Qayyim's *al-Amthāl fī l-Qurʾān* is public domain. ~120 nodes: the three amthāl classes as a register, the 70 mechanical hits kept as the muṣarraḥa floor with the other two classes sourced, never inferred. |
| **F18** | **Gharīb al-Qurʾān measures rarity and cannot measure gharāba.** `W3-X1/gharib_register.json` states its own limit exactly: *"rarity is MEASURED (hapax/near-hapax by lemma); difficulty (gharāba) is the tradition's verdict — al-Sijistānī (scan, OCR program registered) remains the AWAITED classical anchor."* A hapax is not a gharīb word; the two sets overlap and are not the same set. The discipline's verdict layer is entirely awaited. | Gharib al-Quran | PRESENT | **MED** | **(a)** | **Acquisition.** The OCR programme is already registered — run it, or substitute al-Rāghib's *Mufradāt* / Ibn Qutayba's *Tafsīr Gharīb al-Qurʾān* (both public domain, both in OpenITI). ~150 nodes joining verdict to measured rarity. |
| **F19** | **74 editions from one aggregator, and no history of translation.** `P35-EDITIONS/editions_inventory.json`: 74 editions, 55 languages, **all from the QuranEnc shelf**, all contemporary. Device censuses are **English-only, 3 editions of 74**, self-declared. For a waqf whose stated posture is permanence, a single upstream aggregator is a single point of failure; and "History of Quran translation" is a chartered area with **2 seats and no chronology**. | History of Quran translation · Digital corpora | TOUCHED ⬇ | **MED** | **POLISH** | **Nodes + acquisition.** ~250 nodes: a translation-history chronology (Ketton 1143 → Marracci 1698 → Sale 1734 → Rodwell → Pickthall 1930 → Yūsuf ʿAlī 1934 → Arberry 1955; the Urdu line from Shāh Walī Allāh; the Persian and Turkish lines), each entry with its edition, printer and permission record. Plus a second independent translation source to break the QuranEnc dependency. |
| **F20** | **The narrator census matches 48.9% of reports and splits Companions across spellings.** `narrator_census_layer.json`: 17,802 attribution lines matched of 36,390 reports scanned — **51.1% unmatched**, undeclared as a coverage figure. **3,288 "distinct first narrators"** is an artifact of spelling: "Abu Hurairah" (1,241) and "Abu Huraira" (976) are one Companion. Only an `orthographic_merge_view` exists, explicitly *"no rijal identification performed"*. And `graph_layer.json` holds **30 nodes and 30 edges, every edge pointing at `ayah:49:6`** — one anchor verse for every narrator. | Isnad analysis · Ilm al-tabaqat | TOUCHED ⬇ | **MED** | **(c)** | **Engine.** Declare the 51.1% miss as a coverage number (zero-is-a-finding); extend the attribution parser to the unmatched forms; replace the 30-node/one-verse graph with a real narrator↔report bipartite graph. ~200 nodes P18. |
| **F21** | **"Matn criticism" is prefix deduplication, and it under-detects.** `P34-OCEAN/matn_dedup_layer.json`: signatures are *"first-80 matn-skeleton + length band"*. On 118,845 rows it found **5,574 repeat clusters covering 12,311 rows (10.4%)** — implausibly low for a corpus containing both Musnad Aḥmad and the kutub al-sitta, which overlap heavily. **`cross_tradition_echo_clusters: 3`** across a 119k two-tradition ocean is a measurement of the method, not of the ocean. A first-80-character prefix is defeated by exactly what hadith transmission does: riwāya bi-l-maʿnā, varied opening formulae, inserted headings. | Matn criticism · Takhrij | TOUCHED ⬇ | **MED** | **POLISH** | **Engine.** Replace prefix-hash with shingle/MinHash similarity over the full peeled matn, with a **known-positive control set** (20 hadiths whose parallel loci are already documented in the shelf). The takhrīj function comes free with it. ~200 nodes P34. |
| **F22** | **The Shamāʾil landed and its chapters were never read.** `P28-SIRAH-X/X8/x8_layer.json` reports `sections: 423` for al-Tirmidhī's *Shamāʾil* — **415 of those 423 (98%) are bare numerals ("1-","2-","3-")**, i.e. the edition's hadith numbers, and 5 are the modern editor's front-matter (*taqdīm*, *tarjama*, *nasabuhu*). **Zero of the 56 abwāb headers were captured** (`bab` headers = 0). For al-Bayhaqī's *Dalāʾil*, `sections: 2,391` — of which **1,697 (71%) are `[ص: N]` page-number stamps**; true bāb/jimāʿ headers: 668. The founding text is acquired and its structure — the whole point of a shamāʾil register — is unparsed. **Not one trait is registered.** | Shamail · Collection-genre | PRESENT ⬆ | **MED** | **(c)** | **Engine.** `framework/p28_x8x9_unlock.py:50` — `sections()` takes every OpenITI `### |` header at any depth. Filter page-stamps and bare numerals; parse the abwāb; join the 56 chapters to the 2,633 candidates already in `C4-MUHAMMAD/D3-SHAMAIL`. ~200 nodes P28. |
| **F23** | **1,000 bench rows, 1,000 PENDING.** `P36-BENCH`: 12 files, **1,000 rows** (T1 352 / T2 341 / T5s 307; Sunnī 693 / Shīʿī 307), and `bench_verdict` = **"PENDING" in 1,000 of 1,000**. `verdict_intake.json → verdicts: []`. The pack's own law: *"verdicts are human scholars'; the pack starts work, never finishes it."* True, honest — and the work has not started. This is `preliminary_audit.json:109-111` measured from the rows. | Fiqh al-hadith · Mustalah | TOUCHED ⬇ | **MED** | **(c)** | **HUMAN seats.** One sitting of qualified readers against pack 001 (≈83 rows) would produce the waqf's first non-machine verdict. This is the decisive open layer by the waqf's own law, and no node count substitutes for it. |
| **F24** | **The adab sweep enforces one rule, and 15 honorifics are wrong because of it.** `framework/adab_scan.py` (56 lines) checks exactly one thing: `Allah` must be followed by ﷻ. It carries no rule for ﷺ, none for ʿalayhi al-salām, and **none for honorific gender agreement**. Measured across build/: **15 exact-adjacency occurrences of the masculine `رضي الله عنه` immediately following a female Companion's name** — ʿĀʾisha ؓ (9) and Fāṭima ؓ (6) — in 6 files, including the child-facing `P8-SAHABA/D3-NARRATORS/education_pack.json` and the tafsīr shelf (`consensus_33_40.json`, `consensus_96_1.json`). Correct: `رضي الله عنها`. These files are inside the swept first 13,800 points; the sweep could not see it. | (bench-wide, adab) | — | **MED** | **(b)** | **Engine.** Extend `adab_scan.py` with a gendered-honorific rule keyed to a named-persons gender table, plus ﷺ and ʿalayhi al-salām rules, then re-sweep all 40,000 — including the 26,401–40,000 range the prelim declares **UNSWEPT** (`preliminary_audit.json:107`). |

**Not raised as findings, and recorded as such:** the two-axis inheritance gap, the waqf-wide derived-layer
licence statement, and the 26,401–40,000 adab sweep gap are already declared in
`preliminary_audit.json:75, 107-108`. This bench confirms them and does not re-report them as its own.

---

## 3 · WHAT THIS BENCH FOUND GOOD, MEASURED

Named because a bench that reports only defects is measuring with half an instrument (the waqf's own
LAW P, applied to a build instead of a founder):

| Layer | Measure |
|---|---|
| `W1-L4/waqf_signs_layer.json` | 4,573 marks, 8 classes, **two pause-marking editions reconciled three ways, every difference declared and none merged**. The cleanest evidence discipline this bench opened. |
| `THE-ANSWERS.md §1` genre census | 68,646 narrations + 226 rijāl entries + 49 treatise sections separated; Nahj al-Balāgha labelled as khuṭab under Imāmī usage rather than counted as hadith. Textbook collection-genre work. |
| `LENS-AUDIT/CH1_full_census.json` | The build declares **3,662 of 22,188 links (16.5%) rest on isnād-shingle alone** and **87 verses are entirely class-D**. A build that publishes its own weakest number is auditable. |
| `qiraat_layer.json` honest_reading | The layer explains that its own similarity metric is *depressed by a printing convention* and refuses to let the number stand unqualified. |
| `_META.json` per acquisition wave | Every acquired text sha-sealed with its licence (CC BY-NC-SA / MIT / PD) recorded at landing. |
| P30 D-B1 method | *"verbatim-verified against its corpus record AND integrator content-read"*, with 5 deferrals declared rather than guessed. 48 entries is thin; the 48 are sound. |

---

## 4 · THE FILL, TOTALLED

| Fill type | Findings | Scale |
|---|---|---|
| **Engines** (no acquisition, no new nodes) | F1, F2, F12, F15, F20, F21, F22, F24 | 8 repairs. F1, F2 and F24 are corrections to things now published. |
| **Acquisitions** — all public domain / open-licensed, all in OpenITI or Corpus Coranicum | al-Wāḥidī *Asbāb* · al-Dānī *al-Bayān* + *al-Muktafā* · al-Sajāwandī *ʿIlal al-Wuqūf* · Ibn al-Ṣalāḥ *Muqaddima* · Ibn Ḥajar *Nukhba* + *Nuzha* · al-Naḥḥās *Nāsikh* · Ibn al-Qayyim *Amthāl* · al-Rāghib *Mufradāt* · Ibn Qutayba *Gharīb* · al-Nashr / al-Taysīr · manuscript metadata | ~11 acquisitions close 7 findings and would move **6 ABSENT cells off zero**. |
| **Nodes** | ~5,400 across P17/P18/P28/P30/P31/P33/P34/C3 | Sized per finding above; largest single block is F9 (~2,000, staged and human-gated). |
| **HUMAN seats — the binding constraint** | F7 (rijāl scholar, identification) · F8 (mufassir + muḥaddith + faqīh, three empty tiers) · F9 (dossier readers) · F16 (label verification) · F23 (the first bench sitting) | **5 findings cannot be closed by any machine.** `human_verdicts: 0` is not a gap in the build — it is the gap the build was built to expose. |

---

## 5 · THE CENTRE TEST, TALLIED

| Clause | Findings | Reading |
|---|---|---|
| **(a) The Qurʾān as the centre of research** | F1, F5, F10, F13, F14, F16, F17, F18 | 8 findings. The Book's own sciences — qirāʾāt, ʿadd, waqf, gharīb, amthāl, codicology — are built at their *measurable* edge and stop where the tradition's verdict begins. The measuring is real; the tradition's own reading of what was measured is mostly awaited. |
| **(b) Everything returns to Allah ﷻ** | F24 | 1 finding. The exaltation law is enforced by a live engine; the *rest* of the adab law — the Messenger ﷺ, the prophets, the Companions' honorifics and their gender — is not enforced at all, and 15 published errors prove it. |
| **(c) Meaning derived through sīrah / sunnah / hadith** | F3, F6, F7, F8, F9, F11, F15, F20, F22, F23 | **10 findings — the centre of gravity of this bench.** The evidence infrastructure spans 59–80% of the Book. The meaning derived through it spans 1.4–2.3%. Every one of the five human-seat findings sits in this clause. This is not a defect of the build; it is the build's own honest position, measured from two more vantages. |
| **POLISH** | F2, F4, F12, F19, F21 | 5 findings. Measurement instruments, name-vs-ID joins, and single-source dependencies. F2 is POLISH by category and HIGH by consequence: a matrix that over-reports PRESENT hides clause-(a) and clause-(c) gaps from the very council convened to find them. |

---

## 6 · WHAT THIS BENCH COULD NOT DO

Stated so no reader mistakes silence for coverage:

- **215 seats, one sitting.** This bench probed 48 areas and opened roughly 60 artifacts. It did not
  read the Arabic of any acquired work; every claim about a *text's* content is a claim about its
  parse, never its meaning.
- **No finding here is a ruling.** Where the tradition splits — the ʿadd schools, the naskh lists,
  the al-Wāqidī status debate, the ṣaḥīḥayn's standing without a grade string — this bench records
  that it splits and stops.
- **Every finding above is a machine's finding.** Under the waqf's own law, none of them is closed
  until a qualified human reads it. F1, F5, F22 and F24 are the four this bench would put in front of
  a human first, because they are already published and already wrong.

---

**BENCH 1 VERDICT: SEAL WITH GAPS** — the 40,000-point build is honestly constructed, honestly
labelled, and stronger than its own coverage matrix reports it to be, but four published artifacts
state falsehoods that must be repaired before seal (F1 children's layers, F5 a "manuscripts" domain
holding none, F22 an unparsed founding text, F24 fifteen wrong honorifics), the coverage matrix's
PRESENT column is produced by an instrument that cannot fail and must be rebuilt (F2), and six
chartered hadith sciences — ʿilal, gharīb al-ḥadīth, nāsikh in hadith, arbaʿīnāt, shurūḥ, and the
identification of any narrator at all — stand at zero while the meaning-through-sunnah layer that is
the waqf's whole reason stands at 1.4–2.3% of the Book and awaits its first human verdict.