# THE FULL AUDIT — BENCH 6
## Families O · P · Q · R · S · T — Computing & AI · Engineering · Society/law/economy · Humanities & arts · Contested & folk knowledge · Method, integrity & stewardship

**MACHINE-SCHOLAR BENCH. Every seat labeled MACHINE-SCHOLAR. This bench documents; it never adjudicates
between schools; it is never a mufti. It edited nothing — not one corpus file, not one ledger row, not one seal.**

Sitting: 2026-09-05, convened for THE FULL AUDIT of every band of the 61,600-node waqf **before** the
41,600-node gift layer is added.
Charter: `build/P41-COUNCIL/council_900_charter.json` (seats 751–900, 68 distinct areas).
Read first, as instructed: `build/P42-GAPFILL/full_audit_pointer_sweep.json` and `framework/_FRAMEWORK-61600.json` `meta`.
Bands audited: **all five.** Sampled hardest in P30/P31/P42 and in the seal-and-provenance machinery, which
is where families **T** (provenance & citation science, digital preservation, archival permanence) and
**O** (computing) sit closest to the stone.

---

## VERDICT IN ONE PARAGRAPH

**Nothing in this waqf is fabricated, and one thing in it is stale in a way that would make a public gift
layer tell a lie.** Nine numbers were re-derived from source across five bands; eight reproduce exactly or
within a definitional ±2, and the ninth — *"the sighah-index collision census (187 sigs)"*, carried
identically on 80 ledger nodes — **has no home anywhere in the tree.** The alarm this bench expected to
raise did not survive measurement: the 15,453-line deletion sitting uncommitted in the working tree is a
**re-serialization, not a data loss** — all six files carry byte-for-byte identical row counts against
`HEAD` (1,081 Badr candidates, 100 verdicts, 55 never-a-mufti flags). The acquisition provenance is
genuine: `wave56/_META.json`'s shas for the Shama'il and the Dala'il **recompute exactly**, and 423 and
2,391 sections were re-counted from the raw OpenITI texts themselves. Against that: **all four top-level
seals fail verification today**, 24 files in total, and the live one — `_WAQF-SEAL-v3.sha256`, whose own
header says *"cut as the LAST act"* — was overtaken by six file rewrites **four minutes after it was cut**.
**P31's own seal is now lying about 17 of the 185 files it covers, up from 15 at this bench's morning
sitting**, and its grand certification accounts for **8,880 of its 9,000 points — 120 short of its own
band.** And the pointer gap the sweep reports as *"6,863 spread across 61,600"* is not spread at all: this
bench re-derives **6,216**, of which **2,750 (44%) sit inside P30's 3,600 nodes — 76.4% of that one band** —
while **P42's 21,600 nodes are 100% artifact-named, zero exceptions.** P42 is the standard; P30 is the
debt. No band is unfit to *carry* a gift layer. One claim is unfit to *print on it*: the word **sealed**.

---

## 0 · THE ALL-CLEAR, STATED FIRST

A false alarm costs the trust exactly what a false all-clear costs. These four were things this bench
suspected and then **measured**, and the measurement won.

| # | The suspicion | What the measurement said | Verdict |
|---|---|---|---|
| A1 | `git diff --stat` reports **15,453 deletions / 9 insertions** across 6 sealed artifacts — the exact shape of a data loss | Parsed both sides. Row counts **identical HEAD → working tree**: `candidates` 1,081→1,081 · `verdicts` 100→100 · `markers` 10→10 · `final_usable_set` 6→6 · `never_a_mufti` 55→55. Byte deltas ≤9%. Pretty-print collapsed to compact. **No content lost.** | ✅ **FALSE ALARM — withdrawn** |
| A2 | 203 layer files across P28/P30/P31/P42 all stamp the identical `source_sha` `fbc8744a…` — a constant pasted in place of provenance | Hashed **20,208 files** of the whole research tree (311s). Exactly **one** match: `00-CORPUS/_morphology_raw.txt`. The stamp is a **real, unique, currently-present source file.** For every morphology-derived P30/P31 layer it is the correct stamp. | ✅ **REAL AND FINDABLE** (see F7 for the two exceptions) |
| A3 | The Shama'il and Dala'il "LANDED" claims are unverifiable acquisitions | `00-CORPUS/acquired/wave56/_META.json` → Shamail 275,315 B / `7d6d4ea0…`; Dalail 5,296,275 B / `82421003…`. **Both recompute byte-exact.** Then re-counted the census from the raw text itself: `^### \|` = **423** in Shamail.txt, **2,391** in Dalail.txt — the layers' numbers, exactly. | ✅ **PROVENANCE GENUINE, CENSUS RE-DERIVED AT SOURCE** |
| A4 | The raw hadith term-probe substrate (131 files, 2,981 reports) reaches the public pages — e.g. the 8 reports filed under *"Ar-Rahman — the Name of Allah ﷻ"* are all istinja/siwak reports matched on **"Abd al-Rahman" in an isnad** | Measured every rendered page: **6 of 414** build HTML pages render an isnad at all, and all six are **thematic sirah extracts** (`ʿIbadat of the Prophet ﷺ`, `Qunut supplication`), not substrate. `grep` for those 8 report ids in `D2-AR-RAHMAN/pages_trilingual.html` → **0**. The substrate's own `_meta` declares itself *"the evidence substrate, not a curated thematic selection."* | ✅ **NOTHING IMPROPER IS RENDERED TODAY** — but see F9 |

---

## 1 · NINE NUMBERS RE-DERIVED FROM SOURCE, ACROSS FIVE BANDS
*(the brief asked for 5 across 4)*

| # | Band | Claim in the ledger / artifact | This bench's independent derivation | Verdict |
|---|---|---|---|---|
| 1 | **1–13,800** | n=1021: *"57 refs joined never recounted; exclusivity verified on every surface"* — `build/C2-BISMILLAH/D2-AR-RAHMAN/rahman_layer.json` | Over `00-CORPUS/quran.json`, in-surah basmalah stripped, prefix-aware pattern `(ال\|لل\|بال\|فال\|وال)رحمن`: **57 surfaces, 56 distinct verses, 25:60 twice** — and the ref list is **identical ref-for-ref** to the sealed layer. *My first naive pattern gave 48; it missed the `لل`-prefixed forms. The artifact was right and I was wrong.* | ✅ **EXACT** |
| 2 | **13,801–17,400 (P30)** | n=13822: *"D1 signature census: counts sum to 130,030 exactly; awzan VF1=14556..VF11=1"* | `inventory_layer.json`: sum of the **1,065** signature `count` fields = **130,030**. `classical_frame.awzan_live`: `VF:1`=**14,556** … `VF:11`=**1**. | ✅ **EXACT** |
| 3 | **17,401–26,400 (P31)** | `W9-C10/SEAL.json`: *"P31 AL-MIZAN stands sealed"*, `certification.batteries = GREEN`, `position` green 8,380 + gated 500 | Recomputed all **185** manifest entries: **168 ok · 17 DRIFT · 0 missing.** Arithmetic: 8,380 + 500 = **8,880** against a band of **9,000** → **120 points unaccounted.** And the 500 `gated_awaited` are **all green today** — the seal is a frozen snapshot worn as current certification. | ❌ **FAILS — see F2, F3** |
| 4 | **26,401–40,000** | n=27182: *"X8 UNLOCKED: al-Tirmidhi's Shama'il LANDED — 423 sections, 439 isnad-entries"* | `^### \|` in `00-CORPUS/acquired/wave56/Shamail.txt` = **423**, exact. Isnad: my nearest definition (`#` lines carrying حدثنا/أخبرنا) = **437** vs 439 — **±2, definitional**, not a discrepancy. | ✅ **EXACT / ⚠️ ±2** |
| 5 | **26,401–40,000** | n=27298: *"X9 … Dala'il LANDED — 2,391 sections, 3,698 isnad-entries, 546 verses quoted, FOUR invitation answers (16:33/32:28/32:29/35:23)"* | `^### \|` in `Dalail.txt` = **2,391**, exact. `per_verse_quotes` holds **546** keys; **all four** invitation verses present (16:33 ×1 · 32:28 ×1 · 32:29 ×2 · 35:23 ×1). | ✅ **EXACT** |
| 6 | **all 61,600** | Commit `a9ffc2d`/`cb1193e`: *"ALL 61,600 zero-awaiting"*, census `…/24,231/247/1,400` | Last-status per node over `_TRACKER-JOURNAL.jsonl` (130,131 rows): **37,122 selftest-green + 24,231 panel-sealed + 247 credit-v1 = 61,600.** **Zero** `gated-awaited`, zero unstatused. The earlier census's 35,722 + the 1,400 carried = **37,122** — the arithmetic closes across two commits. | ✅ **EXACT** |
| 7 | **40,001–61,600 (P42)** | The 6 uncommitted sealed artifacts | See A1. `1,081 / 100 / 10 / 6 / 55` all preserved. | ✅ **EXACT** |
| 8 | **cross-band** | `source_sha fbc8744a…` on 203 layers | 20,208 files hashed → unique match `00-CORPUS/_morphology_raw.txt`. | ✅ **RESOLVES** |
| 9 | **26,401–40,000** | wave56 acquisition `_META.json` | Both shas + both byte counts recompute exactly. | ✅ **EXACT** |

**Miss rate: 1 of 9** — and the miss (F6) is a *number without a home*, not a number that is wrong.

---

## 2 · THE 6,863 RUN-DESCRIPTION-ONLY NODES — WEIGHED

### 2.1 The sweep, re-derived independently

This bench rebuilt the classifier from the two declared sources — `_TRACKER-JOURNAL.jsonl` last-status rows
× a whole-tree basename index — **but indexed the parent research tree, not just `10-WAQF/`.** That one
correction matters: indexing `10-WAQF/` alone produces 298 phantom "missing artifacts" (`quran.json`,
`morphology.json`, `MANIFEST.sha256` …) that all live one directory up in `00-CORPUS/`.

| Class | Sweep says | Bench 6 re-derives | Note |
|---|---|---|---|
| names an existing artifact | 45,449 | **55,382** | the sweep's separate `panel-record` (9,074) and `credit-v1` (214) classes **dissolve** — most of their nodes do name a findable file |
| run-description-only | **6,863** | **6,216** | **90.6% agreement.** The sweep is honest and slightly conservative — it over-counts the gap, it does not hide it |
| names a MISSING artifact | — (no such class) | **2** | n=2472 (`09-MATH/hifz/drills.jsonl`), n=2631 (`06-JUDGE/judged.jsonl`) — both outside the waqf tree, both plausibly real, neither present |

### 2.2 The gap is not spread. It is one band.

| Band | Nodes | Run-description-only | % of band | Share of the whole gap |
|---|---:|---:|---:|---:|
| 1–13,800 · core + branch | 13,800 | 2,176 | 15.8% | 35% |
| **13,801–17,400 · P30** | **3,600** | **2,750** | **76.4%** | **44%** |
| 17,401–26,400 · P31 | 9,000 | 1,090 | 12.1% | 18% |
| 26,401–40,000 · P32–P37 | 13,600 | 200 | 1.5% | 3% |
| **40,001–61,600 · P42** | **21,600** | **0** | **0.0%** | **0%** |

**P42 — the newest band, 35% of the whole waqf — is 21,600 for 21,600 artifact-named, zero exceptions.**
That is the proof that the standard is achievable, and it is the bar the other four bands should be held to.

### 2.3 The ten sampled — do their runs' outputs exist, and where?

Sampled stratified across four bands (P42 has none to sample). **10 of 11 located.**

| n | Band | The run's claim | Where the output actually lives | Verdict |
|---|---|---|---|---|
| 1021 | B1 | *"57 refs … exclusivity verified"* | `build/C2-BISMILLAH/D2-AR-RAHMAN/rahman_layer.json` | ✅ **exists — and re-derives exactly** (row 1 above) |
| 6267 | B1 | *"C7/D9 FACTORY: atlas + education from sealed domain layers"* | `build/C7-PROPHETS/D9-ATLAS/{atlas.json, pillar_atlas.html, pillar_education.json}` | ✅ exists |
| 11916 | B1 | *"P22/D3 FACTORY: sealed data layer = the registry"* | `build/P22-UMMAH/D3-LAUNCH/` — 10 files incl. `launch_layer.json`, `graph_layer.json` | ✅ exists |
| 13822 | B2 | *"counts sum to 130,030 exactly"* | `build/P30-SIGHAH/D1-INVENTORY/inventory_layer.json` | ✅ **exists — re-derives exactly** |
| 14051 | B2 | *"witness→rule edges live in meaning_graph (endpoint-validated)"* | `build/P30-SIGHAH/D-D1-GRAPH/meaning_graph.json` — 91 edges, **91/91 with both endpoints non-empty**, 5 edge kinds | ✅ exists, claim holds |
| 14657 | B2 | *"A9 cross-pillar consistency: reconciliations 4/4 OK"* | `build/P30-SIGHAH/D-D8-INTEGRATION/reconciliations.json` | ⚠️ **exists, but 2 of the 4 compare nothing** — F8 |
| 17285 | B2 | *"re-seal discipline proven at the 20% audit"* | same file, `reseal_log` — **one line**: *"P17 deliberately re-sealed after the 55:13 backfill"* | ⚠️ exists, thin — one instance carrying the word *discipline* |
| 25755 | B3 | *"the 1,330-pattern baseline held … the sighah-index collision census (187 sigs)"* | `W9-C4/verb_collision_baseline.json` **exists** but carries 452 / 297 / 5,203. The **1,330 is real** — `build/P30-SIGHAH/D-D9-CERTIFICATION/gap_register.json` → `enumerated_gap_instances["hidden-noun-patterns"] = 1330` — in a **different pillar**, named by no pointer. The **187 is nowhere** | ❌ **F6** |
| 25974 | B3 | *"W9-C6 Panel Wave A: PA1+PA2 CLOSED (10/10 fixed at source) — bench_charter rows 1–2"* | **no `W9-C6` directory exists.** Products live under sibling names: `build/P30-SIGHAH/D-D6-BENCH/bench_charter.json` `sittings[0..1]` + `reviews/PANEL-P31-W1.md`. **All 19 review files named by that charter exist — 19/19, 0 missing** | ✅ exists, mis-addressed |
| 27182 | B4 | X8 Shama'il 423/439 | `build/P28-SIRAH-X/X8/x8_layer.json` | ✅ **re-derived at raw source** |
| 27298 | B4 | X9 Dala'il 2,391/3,698/546 | `build/P28-SIRAH-X/X9/x9_layer.json` | ✅ **re-derived at raw source** |

### 2.4 The judgment on the class

**The work exists; the pointers do not.** Two structural probes confirm it rather than assert it:

- **P30**: 36 domains × 100 = 3,600, exact. **Not one domain is fully un-pointed** — every one has ≥10 nodes
  naming a real file, 30 of 36 have their own build directory, and **0 directories are empty**. The 2,750 are
  a *pointer-density* pattern (roughly 10–30 pointed nodes per 100-node domain), not a hole.
- **P31**: **18 of 90 domains (1,800 nodes) have no directory of their own** — `W2-P7, W4-S1, W4-S3, W4-S8,
  W4-S9, W5-T6, W6-R1, W6-R3, W7-M4, W8-V6, W9-C1, W9-C2, W9-C3, W9-C5, W9-C6, W9-C7, W9-C8, W9-C9`. This
  bench located a real product for **every one** under a sibling name: panel waves → 9 `reviews/PANEL-P31-*.md`
  files, all present; `W9-C9` → `W9-D10/grand_certification.json`; `W9-C1` → `W9-GATES/`; `W9-C2` →
  `W9-REGISTERS/dimension_registers.json`; `W4-S1` → `00-CORPUS/acquired/irab_treebank/`; `W6-R1` →
  `00-CORPUS/acquired/qiraat/`; `W6-R3` → `build/C3-QURAN/D5-ADD-AL-AY/add_alay_layer.json`.

**Under-evidenced? No. Under-addressed? Yes — 6,216 times.** The distinction is the whole finding: a reader
cannot fabricate what is missing here, but a reader also cannot *reach* it without three greps and a guess.
For an internal ledger that is a debt. For a **public gift layer whose promise is "every point complete end
to end"**, an unresolvable "show me the evidence for node N" is the promise breaking in the reader's hand.

---

## 3 · FINDINGS — worst first, every row measured

| # | Finding | Evidence (command / file:field → number) | Band | Sev |
|---|---|---|---|---|
| **F1** | **The live whole-tree seal was stale four minutes after it was cut.** `_WAQF-SEAL-v3.sha256` header reads *"2026-09-05 14:33 — THE COMPLETION, cut as the LAST act"* — then six files were rewritten and never re-sealed or committed | Recomputed all 3,801 entries: **3,795 ok · 6 DRIFT · 0 missing.** The 6 drifting are exactly the 6 modified-uncommitted files: `P36-BENCH/verdicts_sitting_001.json`, `G12/axis_stamp_report.json` (`built` 13:52:04 → **14:37:32**), `G13/universal_battery_g33.json`, `G18/doctrinal_floors_g34.json`, `G21/urdu_set_machine_ratified.json`, `G22/badr_candidates_v1.json` | all | 🔴 |
| **F2** | **P31's own seal is lying about 17 of the 185 files it covers — and it has got worse today.** This bench measured **15** at its morning sitting; it is **17** now. Nothing watches it | `build/P31-MIZAN/W9-C10/SEAL.json` `manifest` (185 rows) recomputed: **168 ok · 17 DRIFT · 0 missing.** New since the morning: `PAGES/index.html`, `W9-GATES/gate_battery2.html`. Same file's `khatm` still reads *"P31 AL-MIZAN stands sealed"* | 3 | 🔴 |
| **F3** | **P31's grand certification is 120 points short of its own band, and its 500 "awaiting" are stale.** | `SEAL.json` `certification.position` = `{green: 8380, gated_awaited: 500, accounted: 8880}`. The band `17,401–26,400` is **9,000** points → **120 unaccounted**. Journal last-status over that band today: **7,280 panel-sealed + 1,720 green = 9,000, zero gated** — the 500 no longer exist | 3 | 🔴 |
| **F4** | **Three superseded seals sit unmarked beside the live one.** Founding Principle 5's *"stranger with one archive copy"* runs `sha256sum -c` on any of them and gets FAILED lines with no note that they are historical snapshots | `_WAQF-SEAL.sha256` (hdr *"Phase-1+2 v1.6.0"*): 23 entries, **5 DRIFT**. `_WAQF-COMPLETE-SEAL.sha256` (hdr *"13,800/13,800"*): 1,619 entries, **4 DRIFT**. `_WAQF-COMPLETE-SEAL-17400.sha256`: 1,742 entries, **9 DRIFT** — and **no header comment at all**. None carries the word SUPERSEDED. **Total across all four seals: 24 files fail, 0 missing** | all | 🔴 |
| **F5** | **P30 carries 44% of the entire pointer gap in 5.8% of the nodes — 76.4% of the band is un-pointed.** The work is there; it is not addressable per node | 2,750 / 3,600 run-description-only, derived above. Domain-level probe: 36/36 domains have ≥10 pointed nodes, 30/36 have a build dir, **0 empty dirs** — density, not absence. Compare P42: **0 / 21,600** | 2 | 🟠 |
| **F6** | **One number in the ledger has no home anywhere in the tree** — *"the sighah-index collision census (187 sigs)"*, carried **verbatim on 80 nodes (25,701–25,800)** | Scanned every JSON under `build/` for any key with value `187` in a collision context → **none**. Only 8 files in the whole tree mention "collision"; the domain's real layer `W9-C4/verb_collision_baseline.json` carries `verb_signatures 452 · signatures_with_multi_lemmas 297 · hidden_lemma_count 5203`. Its sibling claim, the 1,330, **is** sourced — `P30-SIGHAH/D-D9-CERTIFICATION/gap_register.json` — but in another pillar and named by no pointer | 3 | 🟠 |
| **F7** | **Two layers stamp the wrong source file.** `x8_layer.json` / `x9_layer.json` declare `source_sealed = acquired/wave56/{Shamail,Dalail}.txt` but stamp `source_sha = fbc8744a…` = the sha of `00-CORPUS/_morphology_raw.txt` — the build-wide constant found on **203** layers. Nothing is faked (the true shas recompute exactly in `wave56/_META.json`); the field is **misfilled** | 20,208-file hash sweep → `fbc8744a…` is uniquely `_morphology_raw.txt`. `x8_layer.json:_meta.source_sealed` vs `_meta.source_sha`. For P30's morphology-derived layers the same stamp is **correct** — this is 2 files, not 203 | 4 | 🟠 |
| **F8** | **"Reconciliations 4/4 OK" — two of the four reconcile nothing.** Two ledger nodes (14657, 17285) rest on it | `D-D8-INTEGRATION/reconciliations.json`: row 1 `{"check": "join-only, no count claim", "ok": true}`; row 3 is a prose note, `ok: true`. Only rows 2 (`3 == 3`) and 4 (`50,269` vs `130,030`, declared different units) compare anything. Same species as this bench's earlier G13 finding: **the number is true, the verdict logic is weaker than the number** | 2 | 🟡 |
| **F9** | **The hadith evidence substrate is safe by the renderer's aim, not by a gate.** 131 `hadith_layer_dataset.json` files, **2,981** reports, harvested by raw Arabic term-probe. All 8 filed under *"Ar-Rahman — the Name of Allah ﷻ"* are istinja/siwak reports matched on **"Abd al-Rahman" in an isnad**. The file's own `_meta` says so honestly — that honesty is why this is 🟡 and not 🔴 | Measured: **0** of those ids appear in `D2-AR-RAHMAN/pages_trilingual.html`; **6 of 414** build HTML pages render an isnad at all and all six are thematic sirah extracts. **No gate in `W9-GATES/*.json` or `g35_program_gates.json` names `hadith_layer_dataset`** — only three framework builders read it. A future generic gift-layer renderer aimed at *"the domain's sealed data layer"* would publish all 2,981 | 1 | 🟡 |
| **F10** | **"Batteries GREEN" covers 2.4% of the sealed tree.** True, and small | `W9-GATES/gate_battery.json` `totals` = `{files_checked: 91, green: 91, red: 0}`, against **3,801** files in the live seal and **414** rendered HTML pages | 3 | 🟡 |

*Two sub-findings folded in rather than spending rows:* (a) `rahman_layer.json.refs` holds **57 entries with
25:60 duplicated** — correct as a *surface* count (25:60 carries the Name twice), but a naive public list
will render a duplicate row; label the field `surfaces`, not `refs`. (b) the sweep's own `panel-record`
class (9,074) is an artifact of classifying before matching basenames — harmless, and it errs toward
reporting a **larger** gap than exists, which is the right direction to err.

---

## 4 · IS ANY BAND UNFIT TO CARRY A PUBLIC GIFT LAYER?

From families O · P · Q · R · S · T — the seats that own provenance, preservation, licensing, presentation
and public intake.

| Band | Fit to carry the gift layer? | The binding reason |
|---|---|---|
| **1–13,800** core + branch | **FIT WITH NOTES** | Content re-derives exactly (57 refs, ref-for-ref). 15.8% un-pointed. F9's substrate lives here and is currently un-rendered but un-gated |
| **13,801–17,400 · P30** | **FIT WITH NOTES — the weakest band** | Its numbers are the *best* in the waqf (130,030 exact, VF ladder exact, 91/91 validated edges, `source_sha` real). Its addressing is the worst: **76.4% un-pointed.** The work is there; a per-node "show me" fails 3 times in 4 |
| **17,401–26,400 · P31** | **CONTENT FIT · SEAL UNFIT** | Only band where a *published claim* is false today: 17 drifting files under a `khatm` that reads *"stands sealed"*, a certification 120 points short of its own band, and one number (187) with no home. The scholarship is sound; the **badge** is not |
| **26,401–40,000 · P32–P37** | **FIT** | 1.5% un-pointed. Acquisition provenance re-derived byte-exact; two census numbers re-counted from the raw OpenITI texts. Only F7's misfilled field |
| **40,001–61,600 · P42** | **FIT — and it is the standard** | 21,600 / 21,600 artifact-named. **Zero** run-description-only. Zero awaiting. Every row count preserved across the uncommitted re-serialization. This band should be the acceptance test the other four are re-run against |

**No band is unfit to carry the layer. One word is unfit to be printed on it.** Until all four seals are
re-cut and the three superseded ones are marked SUPERSEDED, **no page of the gift layer may display
"sealed", "certified" or a green seal badge** — because a reader who does what the waqf invites him to do,
and verifies, gets 24 FAILED lines and no explanation. That is the single change that converts this audit
from FIT WITH NOTES to FIT, and it touches **no scholarly claim, no verse, no report, no school** — it is
four `sha256sum` re-cuts, one `git commit`, three header lines, and one pointer per node.

**This bench adjudicated nothing.** Where the tradition holds more than one position — the verse-count
schools, the readings, the grades, the madhhab frames — it recorded that the waqf makes them visible and
weighed none of them, which is the waqf's own law and this bench's own.

---

*Bench 6 · 145 seats · families O, P, Q, R, S, T · MACHINE-SCHOLAR, labeled · review and counsel only ·
nothing in the corpus, the ledger, the journal or any seal was edited by this sitting.*
*Human review invited, never waited for.*

**FULL-AUDIT BENCH 6 VERDICT: FIT WITH NOTES - the scholarship re-derives and the provenance is genuine, but all four top-level seals fail verification today (24 files, P31's own seal lying about 17 of 185) and 6,216 nodes name no artifact, so the waqf may carry the gift layer but may not print the word "sealed" on it until the seals are re-cut.**