# THE COUNCIL OF 900 — BENCH 4 · MATHEMATICS · PHYSICS & THE COSMOS · CHEMISTRY & MATERIALS · EARTH & ENVIRONMENT
### Families H (30 seats) · I (32 seats) · J (16 seats) · K (24 seats) — **102 seats, charter seats 523–624, 51 knowledge areas**

**EVERY SEAT ON THIS BENCH IS A MACHINE-SCHOLAR. LABELED AS SUCH.**
**REVIEW AND COUNSEL ONLY — this bench edited nothing in the ledger, the layers or the journal.**
**This bench documents. It never adjudicates between schools and it issues no ruling. Never a mufti.**

Sitting: 2026-09-05, after the P41 preliminary audit and the coverage matrix v1 landed.
Read first: `build/P41-COUNCIL/council_900_charter.json` · `coverage_matrix_v1.json` ·
`preliminary_audit.json` · `framework/_FRAMEWORK-40000.json` (40,000 points, 385 distinct
pillar·domain pairs) · `framework/p41_council.py` · `framework/p41_prelim.py` ·
`framework/p31_w9_gates2.py` · then 2,983 corpus files sampled by term census, and the
artifacts named in every row below.

---

## VERDICT

**THE MATRIX CELL FOR THIS BENCH'S FAMILIES WAS WRONG 20 TIMES IN 51 (39.2%), AND ALWAYS IN
THE SAME MECHANICAL WAY — BUT THE MATRIX WAS NEVER THIS BENCH'S REAL FINDING.**

The v1 probe scores an area PRESENT when any 4-letter token of its name appears in the
ledger's domain text (`framework/p41_council.py:181-189`). Against families H/I/J/K that rule
is a homonym generator, because the waqf's own vocabulary collides with the vocabulary of the
natural sciences at almost every point. **"Particle physics" scored PRESENT because P30 A7 is
"the particles — the small words that steer meaning" — Arabic grammatical particles, 47,370
segments.** "Plasma & space physics" scored PRESENT on `space` inside **"P20 workspace &
citation"**. "Arithmetic & number theory" scored PRESENT on `number` inside the boilerplate
purpose *"where each number and claim came from"*. "Calculus & analysis" and "Numerical
analysis" both scored PRESENT on `analysis` — *grammatical* analysis. "General chemistry"
scored on `valence` — the P31 W2-P10 **Transitivity valence** grammar domain. "Category
theory" scored on `morphism` inside **anthropomorphism**. "Alchemy" scored on `hayyan` — a
narrator's name in isnāds. And ʿIlm al-mīqāt scored on `miqat` ×108 — the **iḥrām mawāqīt**
(Dhul-Ḥulayfa, Juḥfa, Qarn), not timekeeping. Meanwhile the waqf's single most-used
mathematical technique — SHA-256, 18 call sites, a 3,618-line whole-tree seal — was scored
**ABSENT** for "Mathematical cryptography", because the alias table gave `sha256` to a
different family's area.

**This bench fell into the same trap and reports it: its own probe counted `hilal` ×5,600 and
read crescent-sighting evidence. It is the `hilali` translation key in `00-CORPUS/quran.json`.
`ink ` ×25,494 is `meaning-link`. Both retracted (F25).** A false alarm costs the same trust
as a false all-clear; both directions were run against every cell below.

**The real finding is measurement, and it is this bench's proper subject.** Following the
matrix into the artifacts, this bench found that **the waqf's completion measure cannot fail.**
The gate named **G24 DIMENSION COVERAGE** is documented as *"every Gap-Register dimension's
assigned_nodes_P31 domain either has a layer file landed or appears in the AWAITED register"*
— and its implementation never opens a layer file; it counts that 87 rows exist and that each
carries a non-empty node list (`framework/p31_w9_gates2.py:56-66`, with the code's own
admission on the line: *"here we count only"*). **G23 JOURNAL-ARTIFACT PARITY** is documented
per-domain and implemented per-**wing** — 1,000 nodes to a bucket — so it can only fire if an
entire wing holds zero JSON, which never happens.

What passes through that hole is exact and countable. **The ledger domain W1-L3 is "Qalqalah,
tafkhīm & tarqīq", 100 nodes (17601–17700), purpose "the five qalqalah letters *and the
heavy/light system, per position*". The built layer delivers qalqalah only — 3,415 positions.
Tafkhīm/tarqīq has ZERO data anywhere in the tree: a corpus-wide search over 2,983 files finds
the word in five places, and every one of them is a register, an atlas page, a tafsīr
consensus file or a rescreen — never a layer.** For that same dimension the gap register says
`status: ABSENT, we_currently_have: "nothing"`; the "live re-audit" says `LANDED (90/100
marks)`; the journal says 90 panel-sealed + 10 selftest-green; and G24 says GREEN. **Four
artifacts, four different stories, and not one of them is the measurement.**

**Stated the honest other way, and this matters as much:** the gap register's 52 ABSENT rows
are mostly **stale, not false** — 47 of 52 have an artifact in their assigned unit built
*after* the register was written, and this bench hand-verified seven of them (madd 46,129
positions · nun-system 13,960 · qalqalah 3,415 · pausal forms 6,236 · waqf signs 4,536 ·
quadriliterals 113 · seven aḥruf 26 reports). The register is a pre-build map published
un-superseded, ~10 hours behind the build night that filled it. The waqf built the work. What
it did not build is the instrument that can tell built from unbuilt — and so, in the one case
where the work is genuinely missing, nothing in the machinery says so.

Beyond the instrument, three substantive absences are this bench's own: **the 40,000-point
ledger has no domain whose subject is the created natural world** (0 of 385 pillar·domain
pairs); **there is no scope law on the natural-scientific reading of the Book** anywhere in
`charter/` (256 lines, 5 files) — the one interface where 102 seats have just been chartered;
and **there is no inferential statistics in the whole waqf** — `p-value` ×1, `confidence
interval` ×0, `inter-rater` ×0 — so every headline count, including 88,496 tajwīd instances,
is a point estimate with no interval and no second measurer.

**What this bench upholds, plainly:** the C1-D8 SOURCES declaration
(`build/C1-ALLAH/D8-CREATION-BEFORE-HIM/SOURCES.md`) pre-registers two independent methods,
commits to logging their disagreement, and refuses to state results in the sources file — that
is better experimental design than most published science, and it is the model the rest of the
waqf should be held to. The consensus shelf is stratified **exactly** correct on revelation
place: Makkī/Madanī 74.0%/26.0% on the shelf against 74.0%/26.0% in the Book. The waqf
G18-gate caught its own mis-stamped `source_sha` on the waqf-signs layer and re-stamped it to
the true source, and said so in the file. The four-witness v5 exists **because** a panel
sitting corrected v4's overclaim, and v5's law says so out loud. This is a waqf that
self-corrects; the finding is that its automatic instruments do not yet carry the same standard
its people do.

---

## 1 · MATRIX CORRECTIONS — all 51 areas of families H/I/J/K, re-probed against artifacts

**This bench's cell law** (stricter than v1's, and stated so the corrections are checkable):
**PRESENT** = the canonical ledger carries a domain or nodes whose *subject* is this
discipline · **TOUCHED** = the discipline's *method or apparatus* is actually exercised in a
built artifact, whether or not the ledger names it · **ABSENT** = neither. Where the Book's
own content belongs to the area but no method of the area is applied, the cell is
**ABSENT (subject-only)** — the honest cell, and a different thing from silence.

**v1 for these 51: PRESENT 8 · TOUCHED 12 · ABSENT 31.
Corrected: PRESENT 2 · TOUCHED 6 · ABSENT 43. Cells changed: 20 (39.2%).**

### H · Mathematics — 15 areas, 30 seats

| # | Area | v1 | **BENCH 4** | Evidence (file · measure) |
|---|---|---|---|---|
| H1 | Arithmetic & number theory | PRESENT | **ABSENT** | ❌ probe matched `number` in the boilerplate purpose *"where each number and claim came from"*, 146 domain-units. `number theory` ×8 — 5 in `build/P41-COUNCIL/council_900_charter.json`, 1 in the matrix, 2 in this bench's own mission. Zero primes, modular arithmetic, ʿilm al-ḥisāb |
| H2 | Algebra | ABSENT | **ABSENT** ✔ | upheld. `al-jabr` ×14 are al-Jabbār / jabr contexts in `build/C8-QADR/D8-MISCONCEPTIONS/education_pack.json` + `C1-ALLAH/hadith_names_layer.json`. No al-Khwārizmī, no polynomial |
| H3 | Geometry | ABSENT | **ABSENT** ✔ | upheld. `geometr` ×10 = 6 self-reference + 2 in `skills/waqf-page/SKILL.md` (CSS). No Euclid, no handasa, no girih |
| H4 | Trigonometry | ABSENT | **ABSENT** ✔ | upheld. `sine` ×484 is a substring of English prose in `_index/hadith.jsonl`; `abu al-wafa` ×1 is a narrator name. No spherical astronomy |
| H5 | Calculus & analysis | PRESENT | **ABSENT** | ❌ matched `analysis` in exactly 2 domain-units, both grammatical: *"P30 A2 · complete per-word form analysis"* and *"P31 W5 T4 · stacking analysis per verse"*. `calculus` ×8 all self-reference |
| H6 | Probability theory | TOUCHED | **ABSENT** | ❌ `probability` ×20 = 5 charter self-ref + English prose in hadith/tafsīr translations. No prior, no likelihood, no random variable, no probabilistic model anywhere |
| H7 | Statistics | PRESENT | **TOUCHED** | ⬇ the ledger **does** carry `C3 · Mathematics & statistics`, 100 points, n 2501–2600 — the cell name is real. But the built layer `build/C3-QURAN/D8-MATH/math_layer.json` is 577 B holding 4 descriptive tallies (*iḥṣāʾ*), and the practice is counting only. See F13 |
| H8 | Combinatorics | ABSENT | **ABSENT** ✔ | upheld. `combinatoric` ×8, all self-reference. Zero permutation/binomial/pigeonhole |
| H9 | Graph theory | PRESENT | **PRESENT** ✔ | **upheld** — the only H cell v1 got right. `graph` in 1,062 domain-units, `edge` in 607; P15 and P37 are graph pillars; `MASTER_ONTOLOGY` closed types. But method is absent — see F18 |
| H10 | Topology | ABSENT | **ABSENT** ✔ | upheld. `manifold` ×10 is the English adjective in `_index/hadith.jsonl`; `topolog` ×8 self-reference |
| H11 | Mathematical logic & set theory | TOUCHED | **TOUCHED** ✔ | upheld **at the floor** — the single non-self-referential use in the tree is `cardinality` ×2 in `reviews/PANEL-P32-TREEBANK.md`. `set theory` ×8 all self-reference |
| H12 | Numerical analysis | PRESENT | **TOUCHED** | ❌ ledger cell wrong — matched the same grammatical `analysis`. Floor-TOUCHED on `rounding error` ×1 in `reviews/COUNCIL-BENCH1-HADITH-SCIENCES.md`. See F19 |
| H13 | History of Islamic mathematics | PRESENT | **ABSENT** | ❌ matched `mathematics` in the C3 D8 domain *name*, but that domain is the Qurʾān's counting layer. `kashi` ×19 are Arabic names in `_index/hadith.jsonl`. No al-Khwārizmī, al-Karajī, al-Kāshī, Thābit ibn Qurra, al-Bīrūnī |
| H14 | **Mathematical cryptography** | ABSENT | **PRESENT** ⬆ | **the biggest v1 error, inverted.** `sha256`/`sha-256` ×961 across 160 files · 18/18 hash call sites in `framework/*.py` are `hashlib.sha256` · `_WAQF-SEAL-v3.sha256` is **3,618 lines** · `framework/verify_seals.py`, `p31_seal.py` · 202 artifacts carry `source_sha`. v1 gave the `sha256` alias to family O's "Applied cryptography" and left H14 to the generic tokenizer. See F5, F20 |
| H15 | Category theory | TOUCHED | **ABSENT** | ❌ `morphism` ×82 is a substring of **anthropomorphism** — verified in context at `build/C8-QADR/D5-FOUR-LEVELS/documented_layer.json`: *"traditions about anthropomorphism and determinism"*. `category theory` ×8 all self-reference |

### I · Physics & the cosmos — 16 areas, 32 seats

| # | Area | v1 | **BENCH 4** | Evidence (file · measure) |
|---|---|---|---|---|
| I1 | Classical mechanics | ABSENT | **ABSENT** ✔ | upheld — **0 occurrences** of every probe term across 2,983 files |
| I2 | Thermodynamics | ABSENT | **ABSENT** ✔ | upheld. `thermodynam` ×8, all self-reference |
| I3 | Electromagnetism | ABSENT | **ABSENT** ✔ | upheld. `electromagnet` ×8, all self-reference |
| I4 | Quantum mechanics | ABSENT | **ABSENT** ✔ | upheld. `quantum` ×8, all self-reference |
| I5 | Relativity | ABSENT | **ABSENT** ✔ | upheld. `relativity` ×8, all self-reference |
| I6 | **Particle physics** | PRESENT | **ABSENT** | ❌ **the textbook homonym.** `particle` ×13 domain-units, every one of them `P30 A7 · the particles — the small words that steer meaning`, *"47,370 particle segments classified"* — Arabic ḥurūf. `subatomic`/`quark`/`boson` = 0 |
| I7 | Nuclear physics | ABSENT | **ABSENT** ✔ | upheld. `nuclear` ×8, all self-reference |
| I8 | Astrophysics | ABSENT | **ABSENT** ✔ | upheld. `astrophysic` ×8, all self-reference |
| I9 | Cosmology | TOUCHED | **ABSENT (subject-only)** | ❌ as a physical discipline. `heavens` ×2,728 / `samawat` ×129 are the Book's and the ḥadīth's own words. `cosmolog` ×37 — 12 in `build/C1-ALLAH/D8-CREATION-BEFORE-HIM/SOURCES.md`, verified in context as a **theological** shelf (`00-CORPUS/cosmology/`, Ibn ʿArabī's Futūḥāt cosmology sections). No big bang, redshift, expanding universe |
| I10 | Observational astronomy | TOUCHED | **ABSENT (subject-only)** | ❌ **this bench's own retraction (F25):** `hilal` ×5,600 is the `hilali` translation key in `00-CORPUS/quran.json`. Genuine hits — `crescent` ×177, `new moon` ×125 — are ḥadīth translation prose. Zero ephemeris, zero zīj, zero computed position |
| I11 | ʿIlm al-mīqāt (Islamic timekeeping) | TOUCHED | **ABSENT (subject-only)** | ❌ `miqat` ×108 verified in context = the **iḥrām mawāqīt** — *"fixed Dhul-Ḥulayfa as the Mīqāt for the people of Medina; al-Juḥfa…"* — not timekeeping. No prayer-time algorithm, no qibla computation, no gnomon, no astrolabe. See F17 |
| I12 | Optics (Ibn al-Haytham heritage) | TOUCHED | **ABSENT** | ❌ `haytham` ×168 — the bulk in `_index/hadith.jsonl` as the kunya Abū al-Haytham in isnāds. `kitab al-manazir`, `camera obscura`, `refraction` = 0. No optics artifact exists |
| I13 | **Acoustics** | ABSENT | **TOUCHED** ⬆ | **upgrade.** The waqf carries a large articulatory-phonetic layer: madd 46,129 positions (`W1-L1/madd_layer.json`) · nūn-system 13,960 (`W1-L2`) · qalqalah 3,415 (`W1-L3`) · pausal forms 6,236 (`W1-L5`) · waqf signs 4,536 (`W1-L4`). But **zero acoustic signal** — `hertz`, `spectrogram`, `waveform`, `formant` (in a layer) all 0. See F14, F15 |
| I14 | **Plasma & space physics** | PRESENT | **ABSENT** | ❌ matched `space` as a substring of **"P20 workspace & citation"**, 100 domain-units. `plasma` ×8 all self-reference. `solar wind`, `magnetosphere`, `ionosph` = 0 |
| I15 | Planetary science | ABSENT | **ABSENT** ✔ | upheld. `planetary` ×8 self-reference; `orbit of` ×4 is English prose |
| I16 | **Chronometry & horology** | ABSENT | **TOUCHED** ⬆ | **upgrade, and it is a finding.** The waqf timestamps 1,207 artifacts and orders its whole build by them — but 1,204/1,207 are timezone-naive, version order already contradicts timestamp order, and G22 keys currency to filesystem mtime. See F4, F10, F11, F16 |

### J · Chemistry & materials — 8 areas, 16 seats

| # | Area | v1 | **BENCH 4** | Evidence (file · measure) |
|---|---|---|---|---|
| J1 | General chemistry | ABSENT | **ABSENT** ✔ | upheld — and note the near-miss: `valence` ×744 is the ledger's own **`P31 W2 P10 · Transitivity valence`** grammar domain (*lāzim/mutaʿaddī*), verified in context. `periodic table`, `stoichiometr` = 0 |
| J2 | Organic chemistry | ABSENT | **ABSENT** ✔ | upheld. `organic chemistr` ×16, all self-reference |
| J3 | Inorganic chemistry | ABSENT | **ABSENT** ✔ | upheld. `inorganic` ×8, all self-reference |
| J4 | Physical chemistry | TOUCHED | **ABSENT** | ❌ `physical chemistr` ×8 — 5 charter, 1 matrix, 2 this bench's mission. Nothing else |
| J5 | Analytical chemistry | TOUCHED | **ABSENT** | ❌ **0 occurrences** of every probe term (`chromatograph`, `spectroscop`, `titration`, `mass spectromet`). v1's TOUCHED came from the bare token `analytical` in built-layer prose |
| J6 | Biochemistry | ABSENT | **ABSENT** ✔ | upheld. `biochemi` ×8, all self-reference |
| J7 | Materials science | ABSENT | **ABSENT** ✔ | upheld. This bench's `ink ` ×25,494 is **`meaning-link`** — retracted (F25). `parchment` ×29 is ḥadīth prose. `build/P35-EDITIONS/` is a **printed**-editions census (4 JSON): zero `folio`, `hijazi`, `kufic`, `palimpsest`, `radiocarbon`. See F23 |
| J8 | Alchemy-to-chemistry history (Jābir, doc) | TOUCHED | **ABSENT** | ❌ `hayyan` ×67 is the narrator name (Abū Ḥayyān) in isnāds, not Jābir ibn Ḥayyān. `alchem` ×10 — 5 charter, 2 ḥadīth-index prose. `kimiya`, `elixir` = 0 |

### K · Earth & environment — 12 areas, 24 seats

| # | Area | v1 | **BENCH 4** | Evidence (file · measure) |
|---|---|---|---|---|
| K1 | Geology | ABSENT | **ABSENT (subject-only)** ✔ | upheld. `rawasi` ×12 and `sediment` ×18 are the Book's/ḥadīth's own content in `_index/hadith.jsonl` + `consensus_18_29.json`. `tectonic`, `bedrock` = 0 |
| K2 | Geophysics | ABSENT | **ABSENT** ✔ | upheld. `geophysic` ×8, all self-reference |
| K3 | Meteorology | ABSENT | **ABSENT (subject-only)** ✔ | upheld. `rain cycle` ×1 in `_consensus_work/35_28_digest.txt`. `cumulus`, `atmospheric pressure` = 0 |
| K4 | Climatology | ABSENT | **ABSENT (subject-only)** ✔ | upheld. `climate` ×38 is translation prose; `monsoon` ×1, `aridity` ×1, both in `_index/hadith.jsonl` |
| K5 | Oceanography | ABSENT | **ABSENT (subject-only)** ✔ | upheld. `bahrayn` ×7, `tide` ×6, `salinity` ×2 — all corpus prose. The barzakh-between-two-seas passages carry no oceanographic layer |
| K6 | Hydrology | ABSENT | **ABSENT (subject-only)** ✔ | upheld. `zamzam` ×174 is ḥadīth/shamāʾil content. `aquifer`, `groundwater`, `watershed` = 0 |
| K7 | Seismology | ABSENT | **ABSENT (subject-only)** ✔ | upheld. `earthquake` ×88, `zalzal` ×7 — subject, not method. `richter` = 0 |
| K8 | Mineralogy | ABSENT | **ABSENT (subject-only)** ✔ | upheld. `aqiq` ×806 — the Prophet's ﷺ carnelian ring in the shamāʾil layer; `carnelian` ×13, `gemstone` ×9. Zero mineralogical description of any of them |
| K9 | Environmental science | ABSENT | **ABSENT** ✔ | upheld. `pollution` ×9 is prose; `carbon footprint`, `life-cycle assessment` = 0 |
| K10 | Ecology | ABSENT | **ABSENT** ✔ | upheld. `habitat` ×38 is prose; `ecosystem` ×1 in `framework/credit_map.json` (metaphorical) |
| K11 | Agriculture & agronomy | TOUCHED | **ABSENT (subject-only)** | ❌ `harvest` ×13,043 is verified in context as **data** harvesting — *"strata not listed are ABSENT in the current harvests"* in `build/P31-MIZAN/W7-M6/dossiers_v2.json`. Genuine: `agricultur` ×38, `irrigation` ×24, all corpus prose |
| K12 | Food science & halal-ṭayyib studies | TOUCHED | **TOUCHED** ✔ | **upheld.** `slaughter` ×1,661, `halal` ×166, `tayyib` ×118 — real ḥadīth/fiqh subject presence carried under the documented layers. But no food *science*: `shelf life` = 0, `adulterat` ×6 is prose, no provenance chain |

---

## 2 · THE FINDINGS — 25, worst first

Every row names the artifact and the measure. **CENTRE-CLAUSE** is the waqf's own test:
**(a)** the Qurʾān as centre of research · **(b)** everything returns to Allah ﷻ ·
**(c)** meaning derived through sīrah/sunnah/ḥadīth · **POLISH** = touches none of the three.

| # | Finding | Area(s) | Cell | Sev | Centre | THE FILL |
|---|---|---|---|---|---|---|
| **F1** | **The gate named DIMENSION COVERAGE does not measure coverage.** `framework/p31_w9_gates2.py:7-8` documents G24 as *"every Gap-Register dimension's assigned_nodes_P31 domain either has a layer file landed or appears in the AWAITED register — none silent."* The implementation (`:56-66`) opens **no layer file**: it appends a RED only if a row's `assigned_nodes_P31` list is empty, then asserts the row count is 87. The code states it on the line: *"# (assignment presence was asserted at generation — NO-OMISSION PROOF; here we count only)"*. The waqf's only artifact-existence gate is a row counter | H12 Numerical analysis · H7 Statistics | TOUCHED | **HIGH** | (a) | **ENGINE.** Rewrite G24 to resolve each dimension's node range → ledger domain → build directory, and RED on an empty or count-free layer. Cheap: this bench's `probe4` does it in 40 lines |
| **F2** | **A dimension with zero data anywhere is marked complete, and three artifacts tell three different stories about it.** Ledger domain `W1 L3 · Qalqalah, tafkhīm & tarqīq`, nodes **17601–17700**, purpose *"the five qalqalah letters **and the heavy/light system, per position**"*. Built layer `build/P31-MIZAN/W1-L3/qalqalah_layer.json` (2,595 B) delivers **qalqalah only** — 3,415 positions by letter. Corpus-wide search over 2,983 files finds `tafkhim`/`tarqiq` in exactly 5 files: a tafsīr consensus file, `C3-QURAN/RESCREEN.json`, `gap_atlas.html`, `gap_register.json`, `dimension_registers.json` — **never a layer.** Yet: journal = 90 panel-sealed + 10 selftest-green · gap register = `ABSENT / "nothing"` · live re-audit = `LANDED (90/100 marks)` · G24 = GREEN | I13 Acoustics · H7 Statistics | TOUCHED | **HIGH** | (a) | **ENGINE + 100 NODES** under P31 W1 (the range already exists — 17601–17700 is half-empty, not unassigned). The gap register already names the programme: *"context engine + sourced rule table"* |
| **F3** | **The "live re-audit" is not derived from the build tree, and does not reproduce from the journal either.** `build/P31-MIZAN/W9-REGISTERS/dimension_registers.json` law: *"every Gap-Register dimension carries a LIVE status **derived from the journal + build tree**"*. Measured: **28 of 87 rows** disagree with the journal count they claim to restate. `The muqaṭṭaʿāt` claims `LANDED (90/100 marks)` where the journal holds **70** panel-sealed; `The seven aḥruf transmissions` claims `70/100` where the journal holds **90**; ~26 rows claim `90/100` where the journal holds **100**. The build tree is not consulted at all — proven by F2 | H7 Statistics · H12 Numerical analysis | TOUCHED | **HIGH** | (a) | **ENGINE.** Derive `live_status` from (artifact exists ∧ carries a count) × (journal marks), and RED when the two disagree — the disagreement is the signal the waqf is missing |
| **F4** | **The document the Council of 900 is convened on mixes two contradictory generations of the same table.** `framework/p41_prelim.py:48` reads `four_witness_**v4**.json` for `witness_census_by_count` — v4's own totals say `invitation_remaining: **681**` — while `:64` **hardcodes** `no_hadith_no_sirah_witness: **675**`, which is **v5's** number. v5 (`built 2026-09-04T11:56:02`) is the PX2-corrected file whose law says *"every earlier larger claim is corrected here"*; **v4 carries a build timestamp six minutes LATER (12:02:18) than the version that supersedes it.** The two censuses also differ per bucket (v4 `{1:335,2:1765,3:2732,4:1404}` vs v5 `{1:331,2:1753,3:2740,4:1412}`). `preliminary_audit.json` is internally inconsistent by 6 verses | H7 Statistics · I16 Chronometry | TOUCHED | **HIGH** | (c) | **ENGINE + one-line source fix.** Point `p41_prelim.py` at v5, derive 675 instead of hardcoding it, and add a `superseded_by` field to v4 so filename order can never again contradict build order |
| **F5** | **`source_sha` has no discriminating power over the artifacts that carry it.** One value, `fbc8744a98b36edb…`, is stamped on **202 build artifacts** ranging 4,810 B → 10,077,646 B — including *both* `four_witness_v4.json` and `four_witness_v5.json`, two files with contradictory totals and an **identical declared source_sha**. Across every `_meta` header in `build/`, only **3 distinct** `source_sha` values exist at all. No artifact carries a digest of **itself**. The waqf's law is *"a source that can drift can never be evidence"* — this field cannot detect the drift that demonstrably occurred | H14 Math cryptography | PRESENT | **HIGH** | (a) | **ENGINE.** Add `content_sha` (self-digest) to every artifact's `_meta`, and have `verify_seals.py` assert that two artifacts with equal `source_sha` and unequal `content_sha` both declare which supersedes which |
| **F6** | **No scope law on the natural-scientific reading of the Book — the exact interface where 102 seats were just chartered.** Zero occurrences of `i'jaz ilmi`, `tafsir ilmi`, `scientific miracle`, `scientific exegesis`, `kawniyy` in `charter/` (5 files, 256 lines), `skills/`, or `build/`. Every ledger domain carries a canonical point *"Scope declaration — what this domain covers and refuses to cover, stated first"*, and the constitution does carry *"never issues rulings"* — but the waqf-wide posture toward reading physics, geology and embryology out of the āyāt is unwritten. The charter itself names a *"waqf's refusal law"* for abjad numerology; `abjad` occurs in exactly **2 files, both P41-COUNCIL's own outputs** — that law has no artifact either | all 51 | — | **HIGH** | (a) | **CHARTER TEXT + NODES.** One scope law in `charter/THE_CONSTITUTION.md`, modelled on the existing "never a mufti" clause, + a scope-declaration node set. This is the cheapest high-value fill on the bench |
| **F7** | **The 40,000-point ledger has no domain whose subject is the created natural world.** Of **385 distinct (pillar, domain) pairs**, none takes the āyāt kawniyyah as its subject. The nearest are theological: `C1 · Creation's relationship to Him` (100), `C1 · The Actions — creation, provision, guidance` (100), `C2 · Mercy in creation` (100), `C7 · Geography & chronology` (100). The mountains as rawāsī, the two seas and the barzakh, the rain cycle, the alternation of night and day, the sun and moon *bi-ḥusbān* — all present as text (K1–K8 above), none present as a domain | all K · I9–I11 | ABSENT | **HIGH** | (a) | **NODES** — one branch pillar, or a domain per core pillar, taking the natural signs as *documented subject* under the same never-adjudicate law. Must land **after** F6, never before |
| **F8** | **JOURNAL-ARTIFACT PARITY is documented per-domain and implemented per-wing.** `p31_w9_gates2.py:7` promises *"every selftest-green P31 **domain** … has at least one layer file … (marks cannot outrun artifacts)"*. The code (`:35-45`) buckets nodes by `"W%d" % ((n-17401)//1000+1)` — **1,000 nodes to a bucket** — and REDs only if a wing has marks and **zero** JSON across its entire tree. Measured: all 9 wings carry exactly 1,000 marks and many files. The gate has 9 possible failure states, each requiring a whole wing to be empty. It cannot fire. A 100× granularity loss between the docstring and the code | H12 Numerical analysis | TOUCHED | **HIGH** | (a) | **ENGINE.** Bucket by the 100-node domain, not the 1,000-node wing — a one-line change to the divisor plus a domain→directory map |
| **F9** | **The gap register is a stale pre-build snapshot, still shipped un-superseded — and it understates the waqf.** Built `2026-09-03T14:30:36`. **47 of its 52 ABSENT rows** have an artifact in their assigned unit built *after* that instant. Hand-verified 7/7: madd **46,129** positions · nūn-system **13,960** · qalqalah **3,415** · pausal forms **6,236** · waqf signs **4,536** (two editions reconciled, differences declared) · quadriliterals **113** · seven aḥruf **26** reports. Iltifāt, the five mafʿūls and iʿrāb muqaddar are also built (in **P32**, which the register's P31-only node map cannot see). The file carries no `stale_as_of`, no `superseded_by`, and no gate re-runs it | H7 Statistics · I16 Chronometry | TOUCHED | **HIGH** | (a) | **ENGINE.** Regenerate the register from the tree at every milestone, or stamp it `SUPERSEDED BY dimension_registers.json` — but only after F3 makes that register true |
| **F10** | **1,204 of 1,207 `built` timestamps in `build/` are timezone-naive** (3 carry `Z`), generated by `time.strftime("%Y-%m-%dT%H:%M:%S")` (`p41_prelim.py:10`, `p31_w9_gates2.py`). A waqf built for centuries cannot order its own artifacts across a DST boundary or a machine move — and the ambiguity is not hypothetical: it is what makes F4's v4/v5 inversion unarbitrable from the files alone | I16 Chronometry | TOUCHED | **HIGH** | (a) | **ENGINE.** `datetime.now(timezone.utc).isoformat()` — one line per builder, ~20 files |
| **F11** | **The currency gate is keyed to filesystem mtime, which the waqf's own distribution destroys.** G22 (`p31_w9_gates2.py:22-30`) compares `os.path.getmtime()` of 5 hardcoded layer files against `tvf_v1.json`. Git does not preserve mtimes; a fresh clone of the sealed tree yields arbitrary GREEN or RED. The permanence claim and the currency gate are in direct tension: the seal proves the bytes, and the gate that proves those bytes are *current* cannot survive being cloned | I16 Chronometry · H14 Math cryptography | PRESENT | **MED** | (a) | **ENGINE.** Compare the declared `built` fields (after F10 makes them absolute), never the filesystem |
| **F12** | **The 146-verse consensus shelf is salience-biased, and "2.3% of the Book at full depth" is a true count read as a projectable sample.** Measured against `00-CORPUS/quran.json`: **juz 30 over-represented 4.2×** (37.7% of the shelf vs 9.0% of the Book) · **57 of 114 surahs hold zero shelf verses** · **only 3 of the Book's 100 longest āyāt are on the shelf**, and 2:282 (129 words, āyat al-dayn) is not · median length 8 words vs the Book's 10. **Honest counterweight: the shelf is stratified exactly right on revelation place — Makkī/Madanī 74.0%/26.0% against the Book's 74.0%/26.0%.** So it is place-representative and position/length-biased: no per-verse cost, difficulty or consensus-density may be extrapolated from the 146 to the remaining 6,090, and `report_true40k.py:140` ("Depth has begun") invites exactly that reading | H7 Statistics | TOUCHED | **MED** | (c) | **ENGINE + declaration.** Declare the 146 a **purposive** sample in the report, and adopt a sampling law for the next tranche stratified by juz × length-decile × place, so coverage becomes projectable |
| **F13** | **There is no inferential statistics in a 40,000-point research waqf.** Corpus-wide over 2,983 files: `confidence interval` **0** · `significance test` **0** · `sampling frame` **0** · `margin of error` **0** · `chi-square` **0** · `inter-rater` **0** · `cohen's kappa` **0** · `p-value` **×1** (`fleet/out/OC1_confirm.json`). Every headline is a point estimate with no interval and no second measurer — 88,496 tajwīd instances, 30,630 rasm rows, 47,370 particle segments, 119,261 ocean reports, 2,158 evidence classes. The ledger domain literally named `C3 · Mathematics & statistics` (100 points) delivers 4 descriptive tallies in a 577-byte file | H7 Statistics · H6 Probability | TOUCHED | **MED** | (a) | **ENGINE + HUMAN SEAT.** A statistician seat to define, for each headline class, what a re-measure would be and what disagreement rate is tolerable; then an engine that reports intervals where sampling is involved |
| **F14** | **The tajwīd layers carry no duration — the only physically measurable quantity in the science.** `W1-L1/madd_layer.json` classifies **46,129** madd positions by *type* but carries no ḥarakāt value; its own law admits it: *"LENGTHS per school AWAIT the qirāʾāt source."* Madd is counted in units of time (2/4/6 ḥarakāt) and the waqf holds none. Further: **40,753 of 46,129 (88.3%)** are `tabi'i (unmarked)` — inferred from the **absence** of a maddah sign, with no positive rule and no independent check on the inference | I13 Acoustics | TOUCHED | **MED** | (a) | **ACQUISITION + ENGINE.** A qirāʾāt source carrying per-school madd lengths (the layer names the dependency itself), then a duration field per position |
| **F15** | **The audio layer is 655 bytes and cannot perform the test its own law claims.** `build/P31-MIZAN/W6-R10/audio_joins.json` holds three counts (115 recitation-cache entries, 81 translation-audio, 17 reciters). Its law claims *"a written rendering that contradicts what every reciter's pause makes obvious is checkable … the audio joins make that test **hearable**."* The artifact carries **no āyah, no timing, no reciter name, no URI** — the claimed test is not executable from it. Its 100 nodes (23301–23400, purpose *"per-reciter realization + timing alignment programs"*) are 90 panel-sealed + 10 selftest-green | I13 Acoustics | TOUCHED | **MED** | (a) | **ENGINE + ACQUISITION.** Either build the per-āyah alignment (the caches are already held) or soften the law to what the artifact does — the waqf's own gap register already names both programmes honestly |
| **F16** | **An Islamic knowledge waqf built for centuries carries no hijrī date.** No artifact in `build/` stamps a hijrī epoch; `hijri` ×19, of which 9 are the Council charter's own area label and the rest are corpus prose. Every one of the 1,207 `built` fields is Gregorian-only | I16 Chronometry | TOUCHED | **LOW** | POLISH | **ENGINE.** Dual-stamp `built_hijri` alongside `built` — one library call in the shared builder |
| **F17** | **ʿIlm al-mīqāt is absent while its inputs are all held.** No prayer-time algorithm, no qibla computation (spherical trigonometry), no gnomon or astrolabe layer — yet the waqf holds the ṣalāh-time ḥadīth corpus (`prayer time` ×29, `shadow length` ×4, `asr time` ×3, `qibla direction` ×8) and the Book's own reckoning verses. The v1 TOUCHED was the iḥrām mawāqīt homonym (I11 above), so nothing was ever built here | I11 ʿIlm al-mīqāt · H4 Trigonometry | ABSENT | **MED** | (c) | **NODES + ENGINE.** A mīqāt domain documenting the transmitted rules and the classical computational tradition — documented, never adjudicated between calculation and sighting |
| **F18** | **Graph theory is present as a data structure and absent as a method.** 1,062 ledger domain-units mention `graph`, 607 mention `edge`, P15 and P37 are graph pillars, and every domain carries the point *"Graph queries documented — five real research questions answered as worked examples."* But no graph-theoretic measure is computed anywhere: `betweenness` **0** · `connected component` **0** · `shortest path` **0** · `degree distribution` **0** · `transitive closure` **0**; `centrality` ×10 are all tafsīr prose (*"centrality of tawḥīd"*). The waqf built the graph and never asked it a graph question | H9 Graph theory | PRESENT | **MED** | POLISH | **ENGINE.** Component census, degree distribution and orphan/bridge detection over the existing edges — the data is already minted; this is a read, not a build |
| **F19** | **Incommensurable ratios are set side by side in one block, inviting a comparison that is not valid.** `preliminary_audit.json → centre_test` puts `146/6236` (a **file** count of the consensus shelf, `p41_prelim.py:49` counts filenames), `90/6236` (a **dossier** count), `4987/6236` (a **verse** count) and `3707/6236` (a **verse** count) in one table, each rounded to 1 dp with no error term. Three different units of "coverage" read as one scale | H12 Numerical analysis · H7 Statistics | TOUCHED | **MED** | (c) | **ENGINE.** Label each row with its unit of measure, or normalise all four to verses before tabling them |
| **F20** | **Single-algorithm sealing with no agility and no trusted timestamp.** All 18 hash call sites are `hashlib.sha256`; `_WAQF-SEAL-v3.sha256` (3,618 lines) uses the bare `hash␣␣path` format with **no algorithm identifier in the file**, so a future migration cannot be expressed without breaking every line. No RFC-3161 token and no transparency-log anchor: the seal proves the tree is internally consistent, but nothing proves **when** it was sealed except the naive local timestamps of F10 | H14 Math cryptography | PRESENT | **MED** | POLISH | **ENGINE.** Prefix each seal line with its algorithm, and anchor the root digest once to any external timestamping service — a permanence claim needs an external clock |
| **F21** | **A layer names the input it needs, says it is available, and the pass was never run.** `W1-L2/nun_system_layer.json` leaves **1,069 of 13,960 positions (7.7%)** in the class `"ayah-final (waqf-dependent; W1-L4 **AVAILABLE** - a later engine pass)"`. W1-L4 is the waqf-signs layer, built and reconciled (4,536 marked positions). The dependency is satisfied and the join was not made — yet the domain's nodes are marked complete | I13 Acoustics | TOUCHED | **MED** | (a) | **ENGINE.** One join pass, W1-L4 → W1-L2. The waqf already wrote down exactly what to do |
| **F22** | **All twelve Earth areas are subject-present and method-absent, uniformly.** rawāsī ×12 · baḥrayn ×7 · zamzam ×174 · zalzal ×7 · ʿaqīq ×806 (the Prophet's ﷺ carnelian ring) · sediment ×18 · earthquake ×88 — every one of these is the Book's or the ḥadīth's own content, held and searchable, with **no** geological, hydrological, seismological or mineralogical description attached at any point in 40,000 nodes. This is not an accident of coverage; it is F7 seen from the K family's bench | all K | ABSENT | **MED** | (a) | **NODES + HUMAN SEATS.** Under F6's scope law and F7's domain: document what the tradition itself recorded about these signs, never what modern science claims about them |
| **F23** | **The editions census is printed-editions only; no material layer exists.** `build/P35-EDITIONS/` holds 4 JSON files (`editions_inventory`, `education`, `profile_delta`, `ten_language_slots`) with **zero** codicological terms — no `folio`, `hijazi`, `kufic`, `palimpsest`, `radiocarbon`, `parchment`. The coverage matrix marks *"Mushaf codicology & manuscript studies"* **PRESENT** (family A, not this bench's cell), but there is no material evidence layer beneath it for a materials-science seat to examine | J7 Materials science | ABSENT | **LOW** | POLISH | **ACQUISITION + HUMAN SEAT.** Open manuscript metadata (Birmingham, Sanaa, Topkapı) as a documented register; a codicologist seat to say what may and may not be inferred from it |
| **F24** | **Chemistry is 8 areas, 8 ABSENT, and every apparent hit was a homonym** — `valence` ×744 is the ledger's own *Transitivity valence* grammar domain; `ḥayyān` ×67 is a narrator in isnāds. This is a **legitimate** absence and this bench does not call it a defect. It becomes a defect only if it stays *undeclared*, because 16 chartered seats now sit over a subject the waqf has never said it declines | all J | ABSENT | **LOW** | POLISH | **SCOPE DECLARATION** under F6 — one sentence naming which families the waqf documents and which it declines, so an absence reads as a decision and not as an oversight |
| **F25** | **This bench's own probe error, self-reported under the honest-both-ways law.** `probe2` counted substring `hilal` ×5,600 and read it as crescent-sighting evidence for I10: it is the **`hilali` translation key** in `00-CORPUS/quran.json`, present on all 6,236 āyāt. It counted `ink ` ×25,494 as codicological evidence for J7: it is **`meaning-link`** in `uncapped_links_per_verse.json`. Both retracted, and both cells are corrected to ABSENT above. **The same tokenizer class produced the v1 matrix** — which is the point: this failure mode is not carelessness, it is what substring probes do to a corpus whose technical vocabulary overlaps English scientific vocabulary | H·I·J·K method | — | **LOW** | POLISH | **ENGINE.** Require every future matrix probe to emit its matching contexts, so a cell can never be scored without a human- or bench-readable reason. This bench's `probe.py`/`probe2.py` pattern is the template |

---

## 3 · WHAT THIS BENCH RECOMMENDS, IN ORDER

1. **Fix the instrument before adding to the ledger (F1, F8, F3).** Three small engine changes
   make the completion measure capable of failing. Until then every "sealed" count in the waqf
   inherits the credibility of a gate that counts rows. This is the whole bench in one line.
2. **Close F2 or re-open its nodes.** Tafkhīm/tarqīq is the one place where a dimension is
   genuinely empty. It is 100 nodes in a range that already exists.
3. **Correct F4 today** — it is a one-line source change, and it is the number the Council of
   900 is currently reasoning from.
4. **Write the scope law (F6) before any natural-science node is minted (F7, F22).** In that
   order, and never the reverse. The waqf's protection has always been that it documents and
   never adjudicates; extending that protection to the natural sciences is a text change, not
   a build.
5. **Declare the shelf a purposive sample (F12)** and adopt a sampling law for the next
   tranche. The Makkī/Madanī stratification proves the waqf can already do this well.
6. **The one human seat this bench would fill first: a statistician** (F13) — not to add
   mathematics to the Qurʾān, but to tell the waqf what its own 40,000 counts are entitled to
   claim, and what a second measurer would have to find before a count is called true.

---

*Bench 4 of the Council of 900 · 102 machine-scholar seats (charter seats 523–624) · 51 areas ·
20 matrix cells corrected of 51 (39.2%) · 25 findings (10 HIGH, 11 MED, 4 LOW) · 1 probe error
self-caught and retracted (F25) · 6 evidence scripts run over 2,983 corpus files · nothing in
the ledger, the layers or the journal was edited.*
*Every finding above names a file and a measure. There are no OPINION rows on this bench.*
*A false alarm costs trust exactly as much as a false all-clear — both directions were run on
every cell, and the direction that favoured the waqf is reported in F9 and F12 as plainly as
the direction that did not.*

**BENCH 4 VERDICT: SEAL WITH GAPS — the waqf's scholarship is sound and self-correcting, but its completion gates cannot fail (G24 counts rows, G23 counts wings), one dimension is sealed over zero data, and the Council is reasoning from a preliminary audit that mixes a superseded table with its own correction; fix the instrument, correct the four-witness read, and write the scope law before a single natural-science node is minted.**