# THE FULL AUDIT — BENCH 1 OF THE COUNCIL OF 900

**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** and
**measures**. Where two schools differ, both are recorded and neither is preferred.

| | |
|---|---|
| **Bench** | 1 of the Council of 900 — the FULL AUDIT sitting |
| **Families** | **A · Quranic sciences** (28 areas, 126 seats) · **B · Hadith sciences** (20 areas, 89 seats) — `build/P41-COUNCIL/council_900_charter.json` `totals.seats_per_family` |
| **Seats** | 215 machine-scholar seats |
| **Sat** | 2026-09-05 |
| **Convened for** | the full audit of all 61,600 waqf nodes before the 41,600-node gift layer is added |
| **Bands audited** | 1–13,800 · 13,801–17,400 · 17,401–26,400 · 26,401–40,000 · 40,001–61,600 (all five) |
| **Read first** | `build/P42-GAPFILL/full_audit_pointer_sweep.json` · `framework/_FRAMEWORK-61600.json` `meta` · `framework/_TRACKER-JOURNAL.jsonl` (130,131 rows) |
| **Human verdicts on record** | **0.** This bench is machine and says so. |

---

## 0 · THE INSTRUMENT WAS CHECKED BEFORE IT WAS USED

This bench does not accept the sweep's four class-counts on the sweep's word. It **rebuilt the
classifier from the two sources the sweep names** — the journal's last-status row per node and a
whole-tree basename index — and re-ran it over all 61,600.

| Class | Published sweep | **This bench's rebuild** | Δ |
|---|---|---|---|
| names-existing-artifact | 45,449 | 45,463 | +14 |
| run-description-only | 6,863 | 7,118 | +255 |
| panel-record | 9,074 | 8,805 | −269 |
| credit-v1 | 214 | 214 | 0 |
| **total** | **61,600** | **61,600** | 0 |

✅ **The sweep's priority chain is confirmed exactly.** Journal last-status over 61,600 nodes gives
`selftest-green 37,122 / panel-sealed 24,231 / credit-v1 247`, and the published classes decompose
against it without remainder: 30,259 + 15,157 + 33 = 45,449 named; 37,122 − 30,259 = 6,863 run-only;
24,231 − 15,157 = 9,074 panel; 247 − 33 = 214 credit. The class definition is therefore
*"selftest-green whose evidence names no basename that exists in the tree."*

📊 **The Δ is the regex, not the waqf.** My token pattern accepts 269 panel-sealed evidences the
sweep's did not, and misses 255 selftest-green ones it caught (269 − 255 = +14, the exact names
delta). **The residual class is robust at 6,863–7,118 (11.1–11.6%)** whichever instrument reads it.
Every count below uses the sweep's published figures where they exist; my rebuild is used only to
*select* nodes to sample. 0 of 61,600 nodes lack a status row.

---

## 1 · THE BAND MAP — WHERE THE UNDER-POINTED NODES ACTUALLY LIVE

The 6,863 are not spread evenly. They are one band's problem.

| Band | Nodes | run-description-only | % | panel-sealed | credit-v1 |
|---|---|---|---|---|---|
| **B1** 1–13,800 core+branch | 13,800 | 2,078 | 15.1% | 1,291 | 247 |
| **B2** 13,801–17,400 P30 | 3,600 | **2,750** | **76.4%** 🔴 | 360 | 0 |
| **B3** 17,401–26,400 P31 | 9,000 | 1,640 | 18.2% | 7,280 | 0 |
| **B4** 26,401–40,000 P32–P37+alloc | 13,600 | 650 | 4.8% | 12,900 | 0 |
| **B5** 40,001–61,600 P42 gap-fill | 21,600 | **0** | **0.0%** ✅ | 2,400 | 0 |

By pillar: P30 2,750 · P31 1,640 · RESERVE 450 · C1 337 · P28 235 · C8 172 · C5/C7 171 each ·
P18 151 · C2 144 · C3 113 · C4 97 · C6 86 · P8 60 · then 30 apiece across P9–P27.

✅ **B5 is the cleanest band in the waqf.** 19,200 of its 21,600 nodes carry an explicit
`CARRIED BY <path>` pointer, and **19,200 of 19,200 resolve to a file that exists on disk — zero
dangling.** The other 2,400 are the G01–G24 panel/seal blocks (100 per program × 24 programs,
n = 40,801–61,600). The newest band is the best-evidenced band. The pointer discipline arrived late;
it did not arrive at P30.

---

## 2 · RE-DERIVATION — 12 NUMBERS, 5 BANDS, FROM SOURCE

Requirement was 5 numbers across 4 bands. This bench re-derived **12 across all 5**, from the corpora
and raw files, not from the summaries that assert them.

| # | Band | Node(s) | Ledger claim | This bench's re-derivation | Verdict |
|---|---|---|---|---|---|
| 1 | **B2** | 13,821–13,830 | "D1 signature census: counts sum to **130,030** exactly; awzan **VF1=14556..VF11=1**" | `00-CORPUS/morphology.json` `segments` = **130,030**; `_morphology_raw.txt` = **130,030** non-empty lines; parsing `VF:n` out of column 4 gives **all 11 forms exact on segments AND on ayat** (VF:1 14,556/5,164 · VF:2 1,614/1,304 · VF:3 501/423 · VF:4 4,601/2,872 · VF:5 466/420 · VF:6 120/108 · VF:7 63/59 · VF:8 1,161/1,000 · VF:9 11/9 · VF:10 459/402 · VF:11 1/1) matching `build/P30-SIGHAH/D5-AWZAN/awzan_layer.json` field-for-field. Its stamped `source_sha` `fbc8744a98b3…c6ba199bb` **is** the SHA256 of `_morphology_raw.txt`. | ✅ **EXACT** — 22 of 22 |
| 2 | **B3** | 20,401–20,500 | "treebank LANDED — **139,376 tokens / 6,236 verses / 121 relations**" | `00-CORPUS/acquired/irab_treebank/noorbayan/Quranic.csv` (utf-16, tab) = **139,376** rows; distinct `chapter_id:verse_id` = **6,236**; distinct `rel_label_ar` = **121** | ✅ EXACT (see F7 on the 121) |
| 3 | **B3** | 20,401–20,500 | "**13,017 sentences** across all **6,236** verses (max **23**, median **2**)" | `Quran.csv` = **13,017** rows over **6,236** distinct verses; max sentences/verse = **23**, median = **2.0** | ✅ **EXACT** — 4 of 4 |
| 4 | **B3** | 21,101–21,200 | "**11,157** muqaddar rows on **4,538** verses" | `build/P32-SHAHADAT/W4-S8/irab_muqaddar_layer.json` `totals` = 11,157/4,538, `per_verse` has **4,538** keys whose row-lists **sum to 11,157** | ✅ EXACT + internally consistent |
| 5 | **B3** | 17,651–17,660 | "qalqalah **sughra 3,415 + kubra 415**" | `build/P31-MIZAN/W1-L3/qalqalah_layer.json` `totals.positions` = 3,415 (ب1134+د965+ج658+ق516+ط142 = **3,415** ✓); `kubra_verse_final.positions` = 415 (د196+ب160+ق40+ط10+ج9 = **415** ✓) | ✅ EXACT, both sums close |
| 6 | **B4** | 29,551–29,700 | RESERVE waves 1–2: "**Tahdhib (13,797) + Istibsar (5,458)** via OpenITI" | `00-CORPUS/hadith_shia_wave1.json` → TahdhibAhkam **13,797**, Istibsar **5,458**. Raw `TahdhibAhkam.txt` carries 13,800 `### $` markers, `Istibsar.txt` 5,475 — the deltas are the OpenITI header block | ✅ EXACT (see F9) |
| 7 | **B4** | P35 | profile-delta engine | `build/P35-EDITIONS/profile_delta.json` `marked_by_n_editions` 67+22+26 = **115** = `verses_marked_by_at_least_one` **115** | ✅ EXACT |
| 8 | **B1** | 41–50 | "C1/D1 CLASSICAL: consensus on 59:24 — **29 editions, 10 agreements**" | `build/C1-ALLAH/classical_layer_dataset.json` `editions_read` = **29**, `len(agreement_extracted)` = **10**; the named source `build/P17-TAFSIR/D1/consensus_59_24.json` **exists**, `editions_read` list length 29, `per_edition_claims` 29 | ✅ EXACT |
| 9 | **B1** | 331–340 | "C1/D4 HADITH (corrected): **24 reports across 3 terms**" | `build/C1-ALLAH/D4-RUBUBIYYAH/hadith_layer_dataset.json` `by_term` = 3 terms (`rabb`/`ar-rabb`/`rabbana`) × **8** = **24** | ✅ EXACT (see F3 — the 8 is a cap) |
| 10 | **B5** | G10 | "asbab harvest **1,895** / **1,084**-placed" | `build/P42-GAPFILL/G10/asbab_harvest_v1.json` — 1,895 rows; rows with a non-empty `placed_verse` = **1,084**; distinct placed verses = **339**; my recomputed densest-10 (3:77 37, 2:158 23, 25:68 19, 2:187 17, 2:223 17, 3:128 16, 2:222 15, 4:65 15, 5:3 15, 2:196 14) **matches `totals.densest` entry-for-entry**; PLACED 1,084 + UNPLACED-DECLARED 811 = 1,895 | ✅ **EXACT** — 14 of 14 |
| 11 | **B5** | G23 | "badi **1,436** device-located rows across 8 devices" | `badi_located_registry.json` `located` sums to **1,436** = 7 × 200 + saj 36; the 200/device cap **is declared** in `_meta.sources_true` | ✅ EXACT, cap honest |
| 12 | **B5** | G01 | dossiers **6,236 / 4,987 / 4,416** | `derivation_dossiers_v1.json` — **6,236** dossier keys; recomputed with-hadith-rows **4,987**; recomputed prophetic-stratum **4,416**; **the entire 10-class `strata_census` re-derives from the rows exactly** | ✅ **EXACT** — 13 of 13 |

**Score: 11 of 12 exact on every sub-number. One falsified — see Finding 1.**

---

## 3 · THE 6,863 WEIGHED — 10 SAMPLED, EVERY RUN'S OUTPUT LOCATED

The question put to this bench: *do these runs' outputs exist somewhere, or are the nodes
under-evidenced?* Ten nodes were drawn across the four bands that carry any (**B5 carries none**),
weighted toward families A and B. For each, the bench went looking for the artifact the run would
have written.

| # | Band | Node(s) | Evidence string (run description only) | Where the output actually lives | Verdict |
|---|---|---|---|---|---|
| 1 | B1 | 41–50 | "C1/D1 CLASSICAL: consensus on 59:24 — 29 editions, 10 agreements" | `build/C1-ALLAH/classical_layer_dataset.json` **+** `build/P17-TAFSIR/D1/consensus_59_24.json` | ✅ **EXISTS, re-derives exact** |
| 2 | B1 | 331–340 | "C1/D4 HADITH (corrected): 24 reports across 3 terms" | `build/C1-ALLAH/D4-RUBUBIYYAH/hadith_layer_dataset.json` | ✅ EXISTS, exact — **but F3** |
| 3 | B1 | 811–890 | "C1/D9 FACTORY: pillar atlas/education/graph/gates from the sealed PILLAR-COHERENT atlas" | `build/C1-ALLAH/D9-ATLAS/` — `atlas.json`, `pillar_education.json`, `gates_audit_record.json`, `pillar_atlas.html` | ✅ EXISTS (4 artifacts) |
| 4 | B1 | 10,641–10,750 | "P18 classical: consensus 16:44 in-session (26 editions)" | `build/P17-TAFSIR/D1/consensus_16_44.json` — `editions_read` **26** | ✅ EXISTS, exact — **but F2** |
| 5 | B2 | 13,821–13,830 | "D1 signature census: counts sum to 130,030 exactly; awzan VF1=14556..VF11=1" | `build/P30-SIGHAH/D5-AWZAN/awzan_layer.json` ← `00-CORPUS/_morphology_raw.txt` | ✅ EXISTS, **22/22 exact** |
| 6 | B2 | 14,801–14,840 | "B2 mawquf_layer: 305 companion (**19 marfu-via** restamped) + 494 successor rows" | `build/P30-SIGHAH/D-B2-MAWQUF/mawquf_layer.json` | ⚠️ EXISTS; 305 ✅ 494 ✅ **19 ✗ — actual 33. F1** |
| 7 | B3 | 20,401–20,500 | "W4-S1 UNLOCKED: 139,376 tokens / 6,236 verses / 121 relations" | `00-CORPUS/acquired/irab_treebank/noorbayan/Quranic.csv` **+** `build/P32-SHAHADAT/W4-S1/relation_census_full.json` | ✅ EXISTS, exact — **but F7** |
| 8 | B3 | 21,101–21,200 | "W4-S8 UNLOCKED AND RUN: 11,157 muqaddar rows on 4,538 verses" | `build/P32-SHAHADAT/W4-S8/irab_muqaddar_layer.json` (+ a rendered page) | ✅ EXISTS, exact + self-consistent |
| 9 | B3 | 17,651–17,660 | "W1-D3 qalqalah domain data complete (sughra 3,415 + kubra 415 + exemplars + page)" | `build/P31-MIZAN/W1-L3/qalqalah_layer.json` (+ `PAGES/W1-L3--qalqalah_layer.html`) | ✅ EXISTS, both sums close |
| 10 | B4 | 29,551–29,700 | "RESERVE waves 1-2 LANDED+SCANNED: Tahdhib (13,797) + Istibsar (5,458)" | `00-CORPUS/acquired/shia_wave1/*.txt` (5 sealed texts + `_META.json`) **+** `00-CORPUS/hadith_shia_wave1.json` | ✅ EXISTS, exact — **but F9** |

### THE BENCH'S JUDGMENT ON THE 6,863

**10 of 10 sampled runs wrote an output that is on disk today, and 9 of 10 re-derive their ledger
number exactly.** The class is therefore **a pointer-hygiene gap, not an evidence gap.** The waqf did
the work; it wrote the receipt in prose instead of in a path.

Two qualifications this bench will not soften:

1. **Prose evidence cannot go stale-safely.** Sample 6 is the proof: the run wrote a layer that was
   later re-stamped three times, the artifact updated, and the ledger's sentence did not
   (Finding 1). A `CARRIED BY <path>` pointer would have been re-read at the next sweep; a sentence
   is never re-read. **6,863 sentences are 6,863 numbers that cannot be re-checked mechanically.**
2. **The 15.1% in B1 is the more dangerous fraction than the 76.4% in B2**, because B1 is the core —
   C1 (337 nodes), C2–C8 (841), P8/P18 (211) — the pillars a gift layer will render first.

---

## 4 · FINDINGS — 10, WORST FIRST, EVERY ROW MEASURED

| # | Severity | Finding | Evidence |
|---|---|---|---|
| **F1** | 🔴 **BLOCKING for the gift layer** | **The marfūʿ-via-Companion count in the public ledger is stale by 14 reports, and the layer contradicts itself.** In family B this is not a rounding error: *marfūʿ* vs *mawqūf* is the mustalah distinction that decides whether a text is the Prophet's ﷺ speech or a Companion's. | `build/P30-SIGHAH/D-B2-MAWQUF/mawquf_layer.json`: `_meta.transmission_types` = {mawquf 194, mawquf_shia 92, **marfu_via_companion 7**, marfu_via_companion_shia 12} → 19. **Counting the 305 `companions` rows' own `transmission_type` gives {mawquf 185, mawquf_shia 87, marfu_via_companion 21, marfu_via_companion_shia 12} → 33.** Both sum to 305. The same file's `_meta.marfu_via` says **33**, contradicting its own `transmission_types`. The restamp arithmetic in `_meta.sp4_resolution` ("+6 marfu restamps") and `_meta.sp4_refinement` ("+8 marfu") gives 7+6+8 = **21** ✅ — **the rows are right; the header sub-block was never refreshed, and the header's number is the one that reached nodes 14,801–14,840.** |
| **F2** | 🔴 **HIGH — a repeat, unrepaired** | **The 36-shard consensus schema island this bench flagged on 2026-09-05 is still there, and it is worse than reported.** 36 shards write no per-edition substrate at all: they declare an integer `editions_read` and only clustered claims. **Across those 36, 1,016 edition-readings are declared while the clusters name 211 distinct edition-attestations — 20.8%.** A reader cannot audit "33 editions read" against 5 names. | `build/P17-TAFSIR/D1/consensus_*.json`, n=146. `editions_in_shard`: **dict 108 / MISSING 36 / list 2** — reproducing `reviews/COUNCIL-900-BENCH-1.md` Finding 6 **exactly**. `editions_read` type: **list 110 / int 35 / null 1**; the int-and-null set overlaps the MISSING-36 set on **35 of 36**. `consensus_68_4.json` carries `editions_read: null` with 15 attesting editions. Sampled node 4 above (16:44, "26 editions") sits in this island. |
| **F3** | 🟠 **HIGH — undeclared cap on a public number** | **Every "hadith layer" in the core band is a top-8 probe, and no `_meta` says so.** A public reader meets "24 reports across 3 terms" and reads a census. It is 3 × the cap. | **341 of 399 terms across 131 `build/**/hadith_layer_dataset.json` files return exactly 8 rows** (the remaining 58 return 1–7, i.e. the corpus held fewer). `_meta.method` declares the probe is "the evidence substrate, not a curated thematic selection" — **honest about kind, silent about the cap.** Contrast F10. |
| **F4** | 🟠 **HIGH — provenance that misreads as provenance** | **175 build artifacts stamp a `source_sha` that is not their source's hash, with no declaration.** `fbc8744a98b3…c6ba199bb` is the SHA256 of `00-CORPUS/_morphology_raw.txt` — the Quranic morphology file. It is stamped on hadith layers, sirah chronologies and a tafsīr consensus shard. | **203 JSON artifacts under `build/` carry that sha; only 28 carry a `source_note`/`sources_note` declaring it is an era stamp.** The undeclared 175 include `build/P30-SIGHAH/D-B1-MARFU/marfu_layer.json`, `build/P28-SIRAH-X/X1/chronology_spine.json`, `build/P17-TAFSIR/D1/consensus_55_13.json`, and all nine `build/LENS-AUDIT/E*.json`. `mawquf_layer.json` **does** declare it ("stamps the index era per the G12 convention") — so the convention is known and the declaration is simply missing from 86% of its uses. |
| **F5** | 🟠 **MEDIUM–HIGH — an unsourced headline** | **"cross-tradition upgrade 127/146 + parity" is not what the consensus shelf measures.** Measured across **both** schemas (per-edition keys ∪ cluster attesting-editions ∪ claim-text keys): **21 of 146 shards (14.4%) name at least one Shīʿī tafsīr edition.** Six Shīʿī slugs exist — `safi` 17, `tabrisi` 6, `mizan` 3, `tusi` 1, `jawami` 1, `nurthaq` 1 — against 29 Sunnī slugs appearing in 26–111 shards each (`tabari`/`jalalayn_en`/`mukhtasar`/`mukhtasar_en` 111). This **is** a genuine upgrade on the v1 bench's "2 of 146". It is not parity, and this bench found no artifact carrying 127/146 for this shelf. ⚠️ *The 21/146 is ✅ VERIFIED; the provenance of "127" is ❓ UNVERIFIED — I searched `build/P42-GAPFILL/`, `build/P35-EDITIONS/`, `build/P17-TAFSIR/D1/` and `reviews/` and did not locate it.* | `build/P17-TAFSIR/D1/consensus_*.json`, all 146, both schemas unioned. Named shards: `consensus_1_5.json` (all five: jawami/mizan/nurthaq/tabrisi/tusi), `consensus_2_255.json`, `consensus_33_33.json`, `consensus_3_103.json`, +17 more. |
| **F6** | 🟠 **MEDIUM — the law says verbatim; the field is truncated** | **428 of 1,895 asbab rows carry a hadith grade cut off mid-grader-name at exactly 120 characters.** G10's law reads "grades verbatim". The stored value is a *stringified Python list*, hard-truncated: `…{'name': 'Muhammad Muhyi Al-Din Abdul Hamid', 'grade': 'Sahih'}, {'name': 'Shu` — every one of the 428 is unparseable. | `build/P42-GAPFILL/G10/asbab_harvest_v1.json`. Length histogram of `grade_verbatim`: **428 rows at exactly 120 chars**, all 428 bracketed-and-unparseable; 207 bracketed and complete; 673 = `"NONE RECORDED IN SOURCE"` (honest); 587 = `"standing by collection (sahihayn)"` (honest). **The honest classes are the majority — the defect is confined to the 428 that carry real grader names.** |
| **F7** | 🟡 **MEDIUM — a census that claims exhaustiveness it does not have** | **The "FULL 121-relation census, published so no class can hide in a top-25 window" hides 8 classes.** The census is keyed on the treebank's **Arabic** labels (121 distinct); its **English** label inventory is **129**. **36 of the 121 Arabic labels carry more than one English label**, and two of the merges cross grammatical category: **فاعل absorbs 76 `Pass` tokens** and **مفعول به absorbs 90 `Subj` tokens**. This is upstream annotation, not a waqf edit — but the census's own law over-claims. | `build/P32-SHAHADAT/W4-S1/relation_census_full.json` — 121 keys, counts summing to **139,376** ✅ (= total tokens). Recount from `Quranic.csv`: `rel_label_ar` distinct **121**, `rel_label` distinct **129**. Per-label deltas confirm the merge: census فاعل 16,691 vs English `Subj` 16,624; مفعول به 10,960 vs `Obj` 10,852. |
| **F8** | 🟡 **LOW–MEDIUM** | **15 of the 110 well-formed consensus shards have fewer per-edition claim entries than editions declared read** — e.g. `consensus_21_107.json` 25 read / 20 with claims, `consensus_113_2.json` 25 / 21. Defensible (an edition may attest nothing) but undeclared, and it makes `editions_read` non-auditable at the row level. | `build/P17-TAFSIR/D1/`: `consensus_107_6.json` 25/24 · `113_2` 25/21 · `114_2` 22/20 · `21_107` 25/20 · `36_1` 25/23 · `36_3` 25/24, +9 more. |
| **F9** | 🟡 **LOW** | **6 OpenITI header lines were parsed as hadith rows in al-Istibṣār.** Rows 1, 3 and 12 carry `.AuthorNAME :: … #META`, `.AuthorSHORTNAMEtrans :: Shaykh Ṭūsī #META`, `.BookSHORTTITLEtrans :: al-Istibṣār #META`. 0 in Tahdhīb. 0.11% of the collection — trivial in aggregate, **not** trivial if row 1 of a Shīʿī canonical collection renders on a public page. | `00-CORPUS/hadith_shia_wave1.json` → `hadith.Istibsar`, 6 of 5,458 rows contain `#META`. |
| **F10** | 🟡 **LOW — a note, and a model** | **The band that needs the least fixing declares the most.** B5's G23 caps its badiʿ registry at 200 verses/device and **says so** in `_meta.sources_true` ("capped 200 verses/device"), so the 1,436 reads honestly. B1's hadith layers cap at 8 and say nothing (F3). **The correction for F3 already exists inside this waqf — it is one sentence, and G23 wrote it.** | `build/P42-GAPFILL/G23/badi_located_registry.json` `_meta` vs `build/C1-ALLAH/**/hadith_layer_dataset.json` `_meta`. |

---

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

From the vantage of 126 Quranic-sciences seats and 89 hadith-sciences seats:

| Band | Verdict | Why |
|---|---|---|
| **B1** 1–13,800 | **FIT WITH A DECLARATION PASS** | The core re-derives exactly (samples 1, 2, 3, 4, 8, 9). What it lacks is **declaration**, not work: the top-8 hadith cap (F3) and the 175 undeclared era stamps (F4). Both are `_meta` sentences, not rebuilds. A gift layer rendering "24 reports" as a census would misrepresent the waqf's own honesty. |
| **B2** 13,801–17,400 | **FIT WITH ONE BLOCKING FIX** | 76.4% run-only is *ugly, not unsound* — its flagship number re-derived 22-for-22 against the raw corpus and its own stamped SHA (sample 5). But **F1 lives here**, and F1 is a mustalah-grade error on the public ledger. Fix F1; the pointer debt can be paid after the gift lands. |
| **B3** 17,401–26,400 | **FIT** | Three independent re-derivations, all exact, one of them internally self-summing (11,157). F7 is a claim to trim, not a number to correct. |
| **B4** 26,401–40,000 | **FIT** | RESERVE's landings re-derive exactly against the raw OpenITI texts. F9 is 6 rows. The charter's law — *"RESERVE nodes are never green until an acquisition lands"* — holds: 450 green, 500 still panel-sealed, and the green ones name a landing that is on disk. |
| **B5** 40,001–61,600 | **FIT — and the standard the rest should be held to** | 0 run-only. **19,200 of 19,200 pointers resolve.** G01's dossiers re-derive 13-for-13 including a full 10-class strata census. G23 declares its own cap. F6 is a 428-row string bug in one field of one file. |

**No band is UNFIT.** But this bench records one thing it will not let pass in silence, per I-3:

⚠️ **The claim "ALL 61,600 zero-awaiting" is contradicted by the waqf's own ledger.** The status
vocabulary indeed contains no `awaiting` value — `selftest-green 37,122 / panel-sealed 24,231 /
credit-v1 247`. But the sealing evidence written onto **2,400 nodes (n = 40,801–61,600)** reads, in
its own words: *"the domain-true re-mark: 17,000 carried / **2,200 honestly frame-built-awaiting-wave**
… the 2,200 awaiting nodes stay awaiting."* **The word was removed from the status column; the
condition was not removed from the waqf, and the ledger says so honestly.** The honest statement is
"zero *undeclared* awaiting" — and that is a statement the waqf has earned. The absolute is not.

The same 2,400-node evidence records four BLOCKED bench verdicts from the P42 closing sitting, the
shared defect (81.9% positional pointers + false gap-id fragments), the fix, and the mechanical
re-verification. **A ledger that carries its own blocked verdicts forward instead of overwriting them
is the strongest single signal this bench measured.** That is what makes the bands fit.

---

## 6 · WHAT MUST HAPPEN BEFORE THE 41,600-NODE GIFT LAYER

| # | Action | Cost | Blocking? |
|---|---|---|---|
| 1 | Refresh `mawquf_layer.json` `_meta.transmission_types` from its own rows and re-stamp nodes 14,801–14,840 with **33**, not 19 | one recount | 🔴 **YES** |
| 2 | Add the cap sentence to the 131 `hadith_layer_dataset.json` `_meta` blocks — copy G23's wording | one sed | 🔴 **YES** — a public reader will read "24 reports" as a census |
| 3 | Add `source_note` to the 175 artifacts stamping the era sha, or replace it with each layer's real source hash | one pass | 🟠 before publish |
| 4 | Retire "127/146 + parity"; publish the measured **21/146** with its six named Shīʿī editions and the open invitation | one line | 🟠 before publish |
| 5 | Re-harvest `grade_verbatim` for the 428 truncated asbab rows as structured data, not `str(list)` | one re-run of G10-D2 | 🟠 before publish |
| 6 | Convert the 6,863 prose evidences to `CARRIED BY <path>` — B1's 2,078 first, B2's 2,750 second | mechanical, B5's format already exists | 🟡 after |
| 7 | Trim the relation-census law to "121 Arabic-labelled classes (129 English labels; 36 merges)" | one line | 🟡 after |
| 8 | Drop the 6 `#META` rows from `Istibsar` | one filter | 🟡 after |

---

## 7 · WHAT THIS BENCH WOULD HAVE ASKED, AND DID NOT

Recorded so the next sitting inherits the question, not the omission:

- **What is the denominator behind "127/146"?** Named in the seal but not locatable in the tree.
  Until it is, the honest public number is 21/146.
- **Are the 2,200 frame-built-awaiting nodes individually identifiable?** They are asserted in the
  2,400 seal-block evidences but carry no distinguishing status or field. A gift layer that renders
  them cannot currently tell them from the 19,200 carried ones. **This is the single largest
  unanswered question this bench leaves.**
- **The 269/255 sweep Δ** is inside the classifier's regex. Publishing the sweep's token pattern
  alongside the counts would make the sweep re-runnable by anyone, which is the waqf's own standard.

---

**FULL-AUDIT BENCH 1 VERDICT: FIT WITH NOTES - eleven of twelve re-derivations across all five bands were exact and 10 of 10 sampled run-description-only nodes have their run's output on disk, so the 6,863 are a pointer-hygiene debt and not an evidence gap; but the stale marfu-via count on nodes 14,801-14,840 and the undeclared top-8 cap on 131 hadith layers must be fixed before any of this is offered to the public as a gift.**