# Machine-Scholar Panel Review · C1 D3 Actions Seed Layer
*labelled: machine-scholar review*
*Date: 2026-09-01 · Reviewer: OP9 (Opus, quality seat)*
*Scope: four files — SOURCES.md, build_actions_layer.py, _selftest_actions.py, actions_layer.json*

**Layer facts (tool-printed / panel-verified):**
- 12 seed rows — within selftest band 10–15 ✅ VERIFIED (panel count from JSON `rows` array = 12)
- All 12 rows: `attestation.class = quran-direct` ✅ VERIFIED (JSON scan)
- All 12 rows: `gloss_en.source = UNSOURCED` ✅ VERIFIED (correct for seed; glosses arrive in Block 3)
- Anchor CHECK C — khalaqa=184, hada=144, ghafara=65 ✅ VERIFIED (builder printed "CHECK C PASS"; confirmed in selftest anchor checks)
- Required rows: all 8 present (khalaqa, hada, 'allama, ahya, amata, ghafara, taba, at'ama) ✅ VERIFIED
- Additional rows: sawwa, qaddara, akhraja, ja'ala ✅ VERIFIED (mechanically derived, V-POS filter, all seed passages)
- Count spot-checks (panel manual count from occurrence_refs): 'allama=41, ahya=51, amata=21, at'ama=13, akhraja=99 ✅ VERIFIED
- No excluded items leaked (sabbaha, abada, qara'a — translit check) ✅ VERIFIED
- Six seed families all covered: khalq, hidaya, ta'lim/provision, ihya'/imata, maghfira, tawba ✅ VERIFIED
- Source seal: `_C1-D3-SOURCES-SEAL.sha256` present; aqida corpus (AQ-01–15) declared "pending acquisition" ✅ VERIFIED
- Protocol markers in builder (7 required): all present ✅ VERIFIED (selftest confirms; panel cross-checked)
- Selftest adab gate: checks `u"ﷻ"` = U+FDFB — correct implementation ✅ VERIFIED (contrast D2's defective U+0631 gate)
- `actions_layer.json` adab status: ﷻ present at all "Allah" positions — selftest gate PASSES ✅ VERIFIED
- `build_actions_layer.py` adab status: 2 bare "Allah" without ﷻ — lines 128 and 344 📊 CALCULATED (panel RAW-PATTERN sweep)
- Builder–JSON divergence: JSON carries "Allah ﷻ jalla jalaluhu" and "Allah ﷻ gave him kingship"; builder produces "Allah jalla jalaluhu" and "Allah gave him kingship" — the JSON was patched after build; builder is out of sync at these two positions ⚠️ ESTIMATE (structural inference from text comparison)

---

## The Nine Seats

| # | Seat | Verdict | Severity | One-line reason |
|---|---|---|---|---|
| 1 | Usuli | PASS with NOTE | note | Non-adjudication posture is exemplary; kalām debates (Afʿāl Ikhtiyāriyyah, causal efficacy, kasb) are correctly documented-never-adjudicated; one note on the ʿaṭāʾ / action boundary for Block 3 |
| 2 | Muḥaddith | PASS | — | No ḥadīth claim is made; all `gloss_en` fields are `UNSOURCED`; ḥadīth corpora are declared in SOURCES.md §1 and correctly deferred to Block 4 |
| 3 | Mufassir | PASS with NOTE | note | All 12 action-lemmas correspond to verified Quranic divine-action attestations per tafsīr consensus; note on sawwa's dual-subject range and ja'ala's extreme polysemy for Block 3 |
| 4 | Qāriʾ | PASS | — | DOCUMENTED FINDING (vii) on cascading CHECK A fallbacks for defective/hollow roots is methodologically sound; NFC normalization (DOCUMENTED FINDING viii) correctly handles exclusion matching; passage identification clean |
| 5 | Lughawī | PASS with NOTE | note | All 12 transliterations are philologically faithful; taba — "He accepted repentance" encodes the divine-subject reading correctly but requires Block 3 gloss noting the double-subject ambiguity; ja'ala's 340-occ scope disclosed |
| 6 | Ḥāfiẓ | PASS with NOTE | note | Three CHECK C anchors hold; five additional panel spot-checks confirm counts independently; note on ja'ala (340) and akhraja (99) — both demand high-priority divine-subject tagging in Block 3 before any public citation |
| 7 | Historian | PASS | — | Source seal records 4 corpus file hashes; aqida corpus AQ-01–15 honestly bounded as "pending acquisition"; all 8 DOCUMENTED FINDINGS preserved in builder source |
| 8 | Pedagogue | PASS with NOTE | note | Lemma-level disclosure present in all 12 rows; dual-subject cases (ahya/amata from 2:258, taba from 2:37) explicitly documented; Block 3 should surface divine-subject frequency separately from lemma frequency for ja'ala, akhraja, and taba |
| 9 | Adab Auditor | **DEFECT** | **defect** | `build_actions_layer.py` has 2 bare "Allah" without ﷻ (line 128 in EXPLICIT_EXCLUDE آتَى reason; line 344 in SEED_NOTES_SUFFIX); the JSON was manually patched to add ﷻ at these positions, so the current JSON is clean and the selftest passes — but any rebuild from the current builder would re-introduce the violation and trigger a selftest failure; builder source must be corrected |

---

## Seat 9 Detail — Adab Defect

**RAW-PATTERN sweep result:** regex "Allah not followed by ﷻ" applied over every string in every file.

| File | Status | Detail |
|---|---|---|
| `build_actions_layer.py` | **DEFECT** | 2 instances — see table below |
| `actions_layer.json` | CLEAN ✅ | ﷻ present at all "Allah" positions; selftest gate confirms |
| `_selftest_actions.py` | CLEAN ✅ | No bare "Allah" in code strings; adab gate checks U+FDFB correctly |
| `SOURCES.md` | CLEAN ✅ | "Allah ﷻ" consistent throughout |

**File: `build_actions_layer.py`:**

| Line | Context | Current (wrong) | Required |
|---|---|---|---|
| 128 | `EXPLICIT_EXCLUDE`, آتَى reason | `"ata/ata (2:258 'that Allah gave him kingship') - too common…"` | `"ata/ata (2:258 'that Allah ﷻ gave him kingship') - too common…"` |
| 344 | `SEED_NOTES_SUFFIX` constant | `"grammatical subject is not Allah jalla "` | `"grammatical subject is not Allah ﷻ jalla "` |

Total: 2 replacements in one file.

**Builder–JSON divergence:**

The JSON (`actions_layer.json`) at these two positions carries ﷻ — "Allah ﷻ gave him kingship" (line 15 of JSON, `_meta.excluded`) and "Allah ﷻ jalla jalaluhu" (notes field of all 12 rows). The builder as written does not produce these strings. The JSON was patched after build; the builder source was not updated to match. Consequence: running the builder now would produce a JSON whose notes carry "Allah jalla jalaluhu" (no ﷻ), which the selftest adab gate would correctly catch as a failure.

**Cascade scope:** The defect is limited to the builder source. The JSON is currently compliant. No count, ref, attestation class, schema validity, or cross-check is affected. The selftest adab gate is functional and correctly targeted (U+FDFB, not a look-alike). No rebuild is required to seal the layer if the JSON is confirmed clean under the integrator's authority — but the builder fix is mandatory for the record, so that the build pipeline is not in a state where rebuilding would break adab compliance.

**Defect class note:** This is the third consecutive D-layer with a builder-source adab defect. D1 Names (OP1): ASCII shorthand `(JJ)`. D2 Attributes (OP6): look-alike rāʾ U+0631. D3 Actions (OP9): bare "Allah jalla jalaluhu" (transliterated exaltation, no glyph) and bare "Allah" with no exaltation. Each variant bypassed the inconsistency it inherited; the RAW-PATTERN sweep is the only reliable gate for this class.

---

## Seat 1 Detail — Usuli Note

**Kalām debates: posture confirmed correct.**

The three debates named in SOURCES.md §6 — Afʿāl Ikhtiyāriyyah, causal efficacy (occasionalism vs. real secondary causes), and kasb/free will — are designated as "documented, never adjudicated" in the SEED_NOTES_SUFFIX appended to every row. The builder produces no text that reads as a verdict. The aqida registry (AQ-01–15) includes all three tradition strands: Atharī/Ithbāt (AQ-01, 11, 12, 13), Ashʿarī (AQ-04, 05, 09, 15), Māturīdī (AQ-06, 10), and Muʿtazilī (AQ-14 — registered precisely to document the Muʿtazilī carriers, not to adjudicate). This is the correct posture under Founding Principle 3.

**Block 3 note — action-vs-attribute boundary:**

Several seed-layer verbs (ghafara, taba, hada) are closely related to Divine Names that appear in D2 (al-Ghafūr/al-Ghaffār, al-Tawwāb, al-Hādī). The classical distinction between the enacted Action (a specific act of forgiving, a specific act of accepting repentance) and the corresponding Attribute of the Essence (al-Maghfira as a divine quality) is debated in the tradition — particularly by Atharī carriers who affirm Afʿāl Ikhtiyāriyyah. Block 3 should annotate the action-attribute boundary for these three verbs with the registered aqida sources, documenting the debate without resolving it.

---

## Seat 3 Detail — Mufassir Note

**sawwa (13 occurrences):**

The seed attestation (87:2) is "alladhī khalaqa fa-sawwā" — He who created and formed complete. Tafsīr consensus (al-Ṭabarī رحمه الله, Ibn Kathīr رحمه الله) reads sawwa here as divine action on creation. However, the lemma-level count of 13 includes uses where sawwa is not a divine action: 91:14 "wa man sawwahā" refers to corrupting the soul (the acting subject is human or the nafs itself). Block 3 divine-subject tagging is the correct gate; the seed row is properly bounded by the lemma-level disclosure.

**ja'ala (340 occurrences):**

The most polysemic verb in the layer. The seed attestation (87:5) is "fa-ja'alahu ghutha'an ahwa" — He made it dark, withered (pasture). The 340 corpus-wide count includes human ja'ala ("he made," "he put," "he designated") in many contexts. The DOCUMENTED FINDING (v) and the `notes` field both disclose this correctly. The Mufassir endorses the seed inclusion — the divine use at 87:5 is unambiguous in context — but marks ja'ala as the highest-priority Block 3 referential-tagging target in this layer. Citing "ja'ala: 340 occurrences" in a divine-action context without Block 3 filtering would be a significant misrepresentation.

---

## Seat 5 Detail — Lughawī Note (lead seat: are the 12 action-lemma glosses faithful)

**Assessment of all 12 transliterations:**

| Lemma | Transliteration | Assessment |
|---|---|---|
| خَلَقَ | khalaqa (He created) | ✅ Faithful |
| سَوَّى | sawwa (He formed/made complete) | ✅ Faithful — "made complete/perfected" captures the classical sense |
| قَدَّرَ | qaddara (He decreed/measured) | ✅ Faithful — Form II; both "measure out" and "decree" are attested senses |
| هَدَى | hada (He guided) | ✅ Faithful |
| أَخْرَجَ | akhraja (He brought forth) | ✅ Faithful — Form IV of kharaja |
| جَعَلَ | ja'ala (He made) | ✅ Faithful — the minimal gloss is correct; polysemy is a counting issue, not a glossing error |
| عَلَّمَ | 'allama (He taught) | ✅ Faithful — Form II causative of ʿalima |
| أَطْعَمَ | at'ama (He fed/provided) | ✅ Faithful — Form IV of ṭaʿima |
| أَحْيا | ahya (He gave life) | ✅ Faithful — Form IV of ḥayiya |
| أَمَاتَ | amata (He caused death) | ✅ Faithful — Form IV of māta |
| غَفَرَ | ghafara (He forgave) | ✅ Faithful |
| تابَ | taba (He accepted repentance) | ✅ Faithful for divine-subject use, with note |

**Note on taba:** The root TW-B (tāba) carries two subject-classes in the Quran: (a) human subject: "he turned/repented" (tāba ilā Allāh ﷻ), and (b) divine subject: "He turned towards/accepted repentance" (tāba ʿalayhi = He accepted his repentance, 2:37). The transliteration "He accepted repentance" correctly encodes the divine-subject reading that belongs in D3. Block 3 must clarify that the 63 occurrences include both classes, and that the divine-subject reading requires the preposition ʿalā to be resolved from context.

**Exclusion endorsements:**

The Lughawī specifically endorses: (a) exclusion of āmana (537 occ) — correct; the Divine security-giving sense from 106:4 is too conflated at lemma level to serve as a divine-action anchor; (b) exclusion of ātā (271 occ) — correct; too common and human-dominated; Block 3 will tag the divine-subject uses; (c) exclusion of talaqqā (2:37) — correct; Adam ʿalayhi as-salām's receiving is a human action in the verse; (d) exclusion of ʿalima (382 occ) vs. admission of ʿallama (41 occ) — this Form I/Form II distinction (DOCUMENTED FINDING ii) is the most philologically important exclusion in the layer.

---

## Seat 6 Detail — Ḥāfiẓ Note

**Three anchor counts verified (tool-printed and panel-confirmed):**

| Lemma | Builder anchor | Selftest anchor | Panel spot-check | Match |
|---|---|---|---|---|
| khalaqa | 184 | 184 | (deferred to builder; QAC-authoritative) | ✅ |
| hada | 144 | 144 | (deferred to builder; QAC-authoritative) | ✅ |
| ghafara | 65 | 65 | (deferred to builder; QAC-authoritative) | ✅ |

**Five additional panel-verified counts (manual count from `occurrence_refs` in JSON):**

| Lemma | Stated count | Panel count | Verdict |
|---|---|---|---|
| 'allama | 41 | 41 | ✅ VERIFIED |
| ahya | 51 | 51 | ✅ VERIFIED |
| amata | 21 | 21 | ✅ VERIFIED |
| at'ama | 13 | 13 | ✅ VERIFIED |
| akhraja | 99 | 99 | ✅ VERIFIED |

**Note on high-count verbs:** ja'ala (340) and akhraja (99) are the two verbs where the lemma count most overstates the divine-subject frequency. The Ḥāfiẓ records these as the highest-priority referential-tagging targets in Block 3 — not because the counts are wrong, but because citing them without divine-subject filtering would produce a false impression of divine-action frequency.

---

## Seat 7 Detail — Historian Note

**Source seal:** `_C1-D3-SOURCES-SEAL.sha256` covers 4 corpus files (quran.json, morphology.json, hadith.json, hadith_shia.json). The aqida corpus (AQ-01–15) is correctly bounded as "pending acquisition." The distinction between a partial seal honestly labelled and a completed seal is maintained.

**DOCUMENTED FINDINGS preservation:** All 8 findings (i–viii) are present in the builder source. No window rediscovery is possible for the qaddara/qadar Form I/II split, the ʿallama/ʿalima distinction, the amana exclusion, the taba double-subject, the ja'ala polysemy, the ahya/amata dual-subject in 2:258, the cascading CHECK A fallbacks, and the NFC normalization requirement.

---

## Seat 8 Detail — Pedagogue Note

**Dual-subject disclosure for Block 3:**

Three rows in this layer present a dual-subject problem that is correctly disclosed in the notes but should be surfaced at block-header level in Block 3:

| Lemma | Occ | Dual-subject issue |
|---|---|---|
| تابَ (taba) | 63 | Covers both Allah ﷻ accepting repentance AND humans returning |
| أَحْيا (ahya) | 51 | Includes Ibrahim ʿalayhi as-salām's divine attribution AND the king's false claim (2:258) |
| أَمَاتَ (amata) | 21 | Same 2:258 context as ahya |

**High-ratio lemmas (lemma count vs. divine-subject count):**

| Lemma | Lemma occ | Approx. divine-subject occ | Ratio concern |
|---|---|---|---|
| جَعَلَ (ja'ala) | 340 | Block 3 work | Highest: many human uses |
| أَخْرَجَ (akhraja) | 99 | Block 3 work | High: includes human expulsions |
| هَدَى (hada) | 144 | Block 3 work | Moderate: includes human guiding |

The disclosure "occurrences are LEMMA-LEVEL" is present in all 12 row notes. The Pedagogue endorses this disclosure and recommends Block 3 introduce a `divine_subject_occ` field alongside `occurrences` for these high-ratio verbs before any public citation.

---

## Notes for Block 3

1. **taba (63 occ)** — add `divine_subject_occ` field; restrict "He accepted repentance" gloss to divine-subject occurrences with ʿalā preposition; cite AQ-01 (Wāsiṭiyyah) and AQ-06 (Kitāb al-Tawḥīd) for Atharī and Māturīdī treatments of divine tawba once acquired.

2. **ahya / amata (51 / 21 occ)** — add `dual_subject_note` to both rows; the 2:258 passage is the seed attestation for both; the king's false imitation is embedded in the same verse; Block 3 should filter divine-subject occurrences and annotate the king-Ibrahim debate as documented in AQ-09 (Asās al-Taqdīs) once acquired.

3. **ja'ala (340 occ) and akhraja (99 occ)** — highest-priority divine-subject referential-tagging targets; block-level annotation must surface divine-subject frequency before these counts appear in any cited context.

4. **sawwa (13 occ)** — block-level annotation should note the 91:14 human/soul use (wa man sawwahā) to prevent misreading of the lemma count as a divine-action count.

5. **Action-vs-attribute boundary (ghafara, taba, hada)** — these verbs are related to Divine Names in D2 (al-Ghafūr, al-Tawwāb, al-Hādī); Block 3 should annotate the classical debate on whether the enacted Action is distinct from, or an expression of, the corresponding Attribute, documenting the Atharī (Afʿāl Ikhtiyāriyyah affirmed), Ashʿarī (all divine actions co-eternal in some sense), and Māturīdī positions without adjudicating between them.

6. **taba gloss note** — Block 3 gloss must clarify that "taba (He accepted repentance)" requires the ʿalā construction; the bare lemma tāba without ʿalā covers human repentance throughout the text.

---

## VERDICT

**FIX-FIRST**

One defect must be corrected before publication:

**(Adab Auditor, defect)** One file:

`build_actions_layer.py` — 2 replacements:
1. Line 128 (EXPLICIT_EXCLUDE, آتَى reason): `'that Allah gave him kingship'` → `'that Allah ﷻ gave him kingship'`
2. Line 344 (SEED_NOTES_SUFFIX): `"not Allah jalla "` → `"not Allah ﷻ jalla "`

The `actions_layer.json` is currently adab-compliant (manually patched; ﷻ present at both positions). The `_selftest_actions.py` adab gate is functional (checks U+FDFB correctly — the improvement over D2). No rebuild is required to seal the layer under integrator authority, since the JSON is confirmed clean. However, the builder fix is mandatory so that the build pipeline does not exist in a state where rebuilding the layer re-introduces the defect and breaks the selftest.

The data layer is methodologically sound in all other respects. Three independent cross-checks (CHECK A reassembly on all positions, CHECK B passage-presence for four key lemmas, CHECK C external anchors) all pass. Five additional panel spot-checks confirm occurrence counts independently. Schema gating is correct across all 12 rows. All required action rows present, covering all six declared seed families. Exclusion log is philologically documented and correctly gates the 14 excluded lemmas. Scope boundaries are honest. Documented findings (i–viii) are preserved. The selftest is proven to catch a seeded anchor defect. The seven Usuli/Mufassir/Lughawī/Ḥāfiẓ/Pedagogue notes above are all Block 3 work — none constitute seed-layer defects. This layer may be sealed immediately upon the two-replacement Adab fix.

---

*Independent human review invited on every finding above. Machine-scholar reviews are labelled as such and held to the same evidence standard as all other findings in this waqf.*

---

## INTEGRATOR ADJUDICATION — 1 September 2026, window Fable (integrator seat)
The single defect APPLIED at source (2 builder replacements verified) — the
panel's reasoning accepted in full: a clean artifact above a regressing
pipeline is not clean. Selftest re-ran GREEN. The seven Block-3 notes stand
queued. **Panel recommendation satisfied → D3 Actions seed: SEALED.**