# PANEL REVIEW — C4 D3 SHAMA'IL PIPELINE
**Window:** OP31 · **Seat:** 9-seat panel (Muhaddith leads) · **Date:** 2026-09-02  
**Pipeline reviewed:** `10-WAQF/build/C4-MUHAMMAD/D3-SHAMAIL/`  
**Files audited:** `SOURCES.md` · `shamail_candidates.json` (_meta + sample) · `shamail_filtered.json` (_meta) · `shamail_layer.json` (full)  
**HS classification files:** `HS3–HS11_shamailclass.json` (9 files, 1,239 total verdicts)

---

## VERDICT: **FIX-FIRST**

The layer's 254 kept rows are substantively sound. The verbatim law is upheld. No composed descriptions exist. The describes-whom gate correctly identified "describes other" cases in its substance. However, the layer `_meta` contains three material accounting errors that make the pipeline non-reproducible, and 83 fleet-confirmed shama'il verdicts are silently absent from the layer with no documented justification. These must be corrected before the layer can be sealed.

---

## SEAT 1 — MUHADDITH (leads)
*Mission: sample 8 kept rows; look up each by id in candidates; read `ar`; verify (a) genuinely describes the Messenger ﷺ and (b) class is right.*

All 8 IDs found in candidates file ✅ VERIFIED. All `ar` fields are full corpus text with complete isnads. Evidence source: `shamail_candidates.json` — each candidate looked up by id.

| # | ID | Class | Describes ﷺ? | Class correct? | Notes |
|---|---|---|---|---|---|
| 1 | `sunni-abudawud-1173` | manner | ✅ YES — ʿĀʾishah رضي الله عنها: he ﷺ sat on the pulpit at sunrise, praised Allah ﷻ, addressed the congregation for the rain prayer | ✅ YES | Eyewitness Companion report; genuine shama'il |
| 2 | `sunni-tirmidhi-2807` | manner | ✅ YES — he ﷺ did not return the salām of a man in two red garments (Ibn ʿAmr رضي الله عنه) | ⚠️ BORDERLINE | Describes a specific behavioral act; defensible as "manner" but the ruling dimension is strong |
| 3 | `sunni-ibnmajah-3859` | speech | ✅ YES — ʿĀʾishah رضي الله عنها: "I heard the Messenger of Allah ﷺ say: Allāhumma innī asʾaluka bismikat-ṭāhir…" | ✅ YES | Habitual duʿāʾ; "kāna yaqūl" — pattern of speech, genuine shama'il |
| 4 | `shia-Al-Kafi-Volume-5-Kulayni-13` | speech | ⚠️ PARTIAL — the Messenger ﷺ is the speaker but the SUBJECT described is Abū Dujānah's proud walking | ⚠️ CONCERN | His ﷺ evaluative statement is documented; the trait being described is Abu Dujānah's behavior, not his ﷺ own manner. Borderline for shama'il proper. |
| 5 | `sunni-ibnmajah-3646` | dress | ✅ YES — the Messenger ﷺ wore a silver ring with an Ethiopian gemstone, stone worn toward his palm (Anas رضي الله عنه) | ✅ YES | Classic dress shama'il; unambiguous |
| 6 | `sunni-nasai-5290` | dress | ✅ YES — the Messenger ﷺ had a gold ring made, wore it stone-inward, then removed it publicly and abandoned it (Ibn ʿUmar رضي الله عنه) | ✅ YES | Dress history with explicit first-person conduct; valid shama'il |
| 7 | `shia-Al-Kafi-Volume-4-Kulayni-151` | appearance | ✅ YES — sweat appeared between his ﷺ eyes, complexion changed, he ﷺ looked at the ground when angered by an argumentative delegate | ✅ YES | Physical manifestation of anger; appearance class correct. Chain: Imam al-Ṣādiq from unnamed man — Shīʿī tradition, SHII-TRADITION label applies |
| 8 | `sunni-muslim-6066` | appearance | ✅ YES — Al-Barāʾ رضي الله عنه: "the Messenger of Allah ﷺ had the most handsome face among men and the best disposition (khulq), neither very tall nor short" | ✅ YES | Explicit Companion eyewitness testimony; dual physical+character attestation |

**Muhaddith finding:** 7 of 8 samples are clearly sound. Sample 4 is the one concern: the hadith records the Prophet's ﷺ spoken judgment about another person's behavior, not a description of his ﷺ own qualities. Its inclusion in the layer is defensible but the Muhaddith seat flags it as requiring a `describes_whom` note.

---

## SEAT 2 — ADAB (never-compose law)
*Mission: is ANY description in the layer composed rather than referenced?*

**Finding: CLEAN.** ✅ VERIFIED

The layer carries **zero Arabic text**. Every row contains only: `id` (reference to candidates file) · `class` · `reason` (English-language classification rationale, not a description) · `collection` · `num`. The actual Arabic descriptions are in `shamail_candidates.json`'s `ar` fields, which are programmatic verbatim copies from the corpus.

The candidates' selftest confirms:
```
"prove_fail": {
  "passed": true,
  "seeded_retype_method": "first-kaf-replaced-with-qaf (mark-stripped)",
  "note": "Verbatim gate must REJECT a seeded retype — gate returning False is the correct result"
}
```

The `reason` field in each layer row is English classification prose (e.g., "The Messenger ﷺ wore a silver ring…") — this is a one-line finding note, not a composed Arabic description of his ﷺ character. No description of the Messenger ﷺ was produced by this pipeline from any pen other than the narrators whose words are preserved verbatim in the corpus. The doctrine heart is upheld.

---

## SEAT 3 — USULI (describes-whom gate audit)
*Mission: is the describes-whom gate's 853-drop honest?*

### Count verification
All 9 HS output files read in full. Source: `fleet/out/HS3–HS11_shamailclass.json`.

| File | Verdicts | shamail class | not-shamail | describes ﷺ | describes other |
|---|---|---|---|---|---|
| HS3 | 138 | 17 | 121 | 64 | 74 |
| HS4 | 138 | 27 | 111 | 65 | 73 |
| HS5 | 138 | 36 | 102 | 70 | 68 |
| HS6 | 138 | 32 | 106 | 32 | 106 |
| HS7 | 138 | 42 | 96 | 80 | 58 |
| HS8 | 138 | 65 | 73 | 116 | 22 |
| HS9 | 138 | 68 | 70 | 68 | 70 |
| HS10 | 138 | 36 | 102 | 36 | 102 |
| HS11 | 135 | 14 | 121 | 20 | 115 |
| **TOTAL** | **1,239** | **337** | **902** | **551** | **688** |

✅ VERIFIED: total HS verdicts = 1,239 (matches layer meta) · no duplicate IDs · all 9 input files confirmed.

### Cross-tab of the 1,239 verdicts

| | describes ﷺ | describes other |
|---|---|---|
| **shamail class** | **337** | **0** |
| **not-shamail class** | **214** | **688** |

### The 853-drop: three problems

**Problem 1 — Mislabel (material).**  
Layer meta says `describes_other_dropped = 853`. Actual "describes other" per the HS gate: **688**. ❌ DOES NOT MATCH.  
The 853 − 688 = **165 over-count**: 165 verdicts labeled "describes other dropped" actually describe the Messenger ﷺ.

**Problem 2 — Unaccounted gap (material).**  
Arithmetic: 1,239 − 254 (kept) − 853 (described as dropped) = **132 unaccounted**.  
These 132 exist in the HS output (they are "not-shamail AND describes the Messenger ﷺ" verdicts) but appear in neither the kept set nor the `describes_other_dropped` count.

**Problem 3 — 83 silent exclusions (significant).**  
337 verdicts were classified by the HS fleet as BOTH (a) shamail category AND (b) describes the Messenger ﷺ. The layer keeps only 254 of these. The remaining **83** — fleet-confirmed as shamail AND describes ﷺ — were excluded with no documentation in the layer meta or any log.

Sampling of the 83 shows they include:  
- Physical-description reports (e.g., `sunni-muslim-6089`: height/colour/hair — these should be P28-D2-POINTERs, correctly excluded if so labeled)  
- Ritual-manner descriptions (prayer posture, ghusl method — plausible P28-D3-POINTER territory)  
- Genuine shama'il entries (gravity/deportment, sleeping habits, food preferences) — unclear why excluded  

The exclusion criteria for these 83 is not stated.

### Spot-check of the 688 actual "describes other" drops

5 random samples drawn from the 688. All 5 are correctly dropped:

| ID | Why dropped |
|---|---|
| `shia-Al-Khisal-Saduq-1034` | Jews visiting ʿUmar رضي الله عنه — describes him, not the Messenger ﷺ |
| `sunni-malik-968` | Abū Bakr رضي الله عنه dispatching armies to al-Shām — describes Abū Bakr |
| `sunni-bukhari-4725` | Mūsā عليه السلام and al-Khiḍr — describes them, not the Messenger Muhammad ﷺ |
| `shia-Kitab-al-Ghayba-Numani-266` | Imam al-Ṣādiq describing the Imāms — not the Messenger ﷺ himself |
| `shia-Kitab-al-Ghayba-Numani-265` | Discussion of the Imāmate — no description of the Messenger ﷺ |

**Usuli finding:** The gate's **substance is honest** — the 688 "describes other" drops are correctly identified. The gate is **not dishonest in its verdicts**. However, the **accounting is materially wrong**: the label `describes_other_dropped = 853` inflates the "describes other" count by 165, 132 verdicts are silently unaccounted for, and 83 fleet-confirmed shamail_prophet entries are excluded without documented justification.

---

## FIXES REQUIRED (before SEAL)

| # | Fix | Severity | Location |
|---|---|---|---|
| F1 | Correct `describes_other_dropped` from **853 → 688** in `shamail_layer.json` `_meta` | Material | `shamail_layer.json._meta.counts` |
| F2 | Add `not_shamail_prophet_dropped: 214` to the counts block (reports describing the Messenger ﷺ but classified not-shamail) | Material | `shamail_layer.json._meta.counts` |
| F3 | Add `shamail_prophet_additionally_filtered: 83` with a stated reason (P28-D2-POINTER, duplication, or explicit exclusion rule) | Material | `shamail_layer.json._meta.counts` |
| F4 | For the 83 excluded shamail_prophet entries: either (a) label as `P28-D2-POINTER` and confirm redirect, or (b) add to layer with class and reason, or (c) document specific exclusion criterion per entry | Significant | Layer content or pointer file |
| F5 | Review `shia-Al-Kafi-Volume-5-Kulayni-13` (speech class, Sample 4): add a `describes_note` clarifying that the Messenger ﷺ is the speaker and the subject described is another person's behavior | Minor | Layer row |

---

## WHAT IS ALREADY SOUND

- SOURCES.md doctrine: the never-compose law is clearly stated; the boundary with P28 is explicit; AWAITED-ACQUISITION texts are honestly labeled. ✅
- Probe method: mark-stripped probing, inherited from C2 D5; documented correctly. ✅
- Candidate harvest: 3,921 candidates, selftest passed (5/5 gates), no empty `ar` fields, seeded-retype gate correctly rejects. ✅
- Verbatim law: upheld across all 8 sampled candidates. ✅
- Adab law: all 8 sampled `ar` fields carry the Messenger ﷺ with ﷺ; Companions with رضي الله عنه; Allah ﷻ with ﷻ. ✅
- Signal filter: 2,669 of 3,921 candidates passed signal scoring. ✅
- Substance of the 254 kept rows: 7/8 sampled are unambiguously correct; 1 is borderline but defensible. ✅
- Substance of the 688 "describes other" drops: 5/5 sampled are correctly excluded. ✅

---

## SUMMARY TABLE

| Seat | Finding | Status |
|---|---|---|
| Muhaddith | 7/8 samples sound; 1 borderline (speech about another's behavior) | ⚠️ NOTE |
| Adab | Zero composed descriptions; verbatim law upheld; selftest gate passes | ✅ CLEAN |
| Usuli | Gate substance honest; meta accounting wrong on 3 counts; 83 exclusions undocumented | ❌ FIX-FIRST |

**Overall verdict: FIX-FIRST.**  
Apply F1–F3 to the `_meta` and resolve F4 (the 83 exclusions with documentation or reinstatement). F5 is advisory. When F1–F4 are done, re-run this panel on the corrected layer.

---

*Output written by OP31 · 2026-09-02 · Evidence basis: 9 HS output files (1,239 verdicts), layer.json (254 rows), candidates.json (_meta + 8 samples), SOURCES.md*

---

## INTEGRATOR ADJUDICATION — 2 Sep 2026, window Fable (integrator seat)
All findings verified and the fault owned: the first merge's describes-whom
substring test silently dropped 83 confirmed rows and mislabeled the
accounting. Merge v2 rebuilt with EXACT partition accounting (sum asserted
in-build): kept 337 (the panel's fleet-confirmed figure recovered to the
row) · pointers-for-sirah 214 (never dropped silently again) · describes-
other 688 (the panel's true figure) · unstated 0 · ghosts 0 = 1,239.
The panel's forensic accounting stands as the model for every future merge.
**Shama'il layer v2: SEALED.**