# PANEL WAVE B — P31 AL-MIZAN — PROBES REVIEW **Scope**: PB1 — Probe-class layers (shard/corpus formula probes) **Panel class**: MACHINE-SCHOLAR **Date**: 2026-09-04 --- ## VERDICT: FIX-FIRST One layer (X3_addad) has a false-positive rate that makes its candidate count misleading. All other probe layers are CLEAN on engine counts and law text; false-positive rates are within the "candidate until read" discipline, with findings noted for the reading bench. --- ## ENGINE RE-DERIVATION TABLE | Layer | File | Engine | JSON count | Engine output | Match? | |---|---|---|---|---|---| | W5-T5 hadhf | hadhf_candidates.json | p31_hadhf.py | 5509 hits / 3492 verses | 5509 hits / 3492 verses | ✅ VERIFIED | | W6-R5 makki/madani | R5_makki_madani_candidates.json | p31_w6_probes.py | 3085 hits / 1957 verses | 3085 hits / 1957 verses | ✅ VERIFIED | | W6-R6 naskh | R6_naskh_candidates.json | p31_w6_probes.py | 515 hits / 371 verses | 515 hits / 371 verses | ✅ VERIFIED | | W6-R7 muhkam/mutashabih | R7_muhkam_mutashabih_candidates.json | p31_w6_probes.py | 327 hits / 264 verses | 327 hits / 264 verses | ✅ VERIFIED | | W3-X3 addad | X3_addad_candidates.json | p31_w6_probes.py | 102 hits / 92 verses | 102 hits / 92 verses | ✅ VERIFIED | | W3-X4 muarrab | muarrab_candidates.json | p31_x4_x9.py | 913 hits / 563 verses | 913 hits / 563 verses | ✅ VERIFIED | | W3-X9 drift | drift_harvest.json | p31_x4_x9.py | 2005 hits / 1148 verses | 2005 hits / 1148 verses | ✅ VERIFIED | | W6-R9 asbab | asbab_census.json | p31_w6_r8r9.py | 1987 hits / 1228 verses | 1987 hits / 1228 verses | ✅ VERIFIED | | W7-M9 differences | difference_registry.json | p31_w7_m3910.py | 13065 hits / 4940 verses | 13065 hits / 4940 verses | ✅ VERIFIED | | W3-X1 gharib | gharib_register.json | p31_x1_x10.py | hapax 1958 / near-hapax 1135 | same | ✅ VERIFIED | | W3-X10 misreadings | misreading_harvest.json | p31_x1_x10.py | 3983 hits / 3010 verses | 3983 hits / 3010 verses | ✅ VERIFIED | All 11 engines returned SELFTEST GREEN. Counts are internally consistent. --- ## NUMBERED FINDINGS ### FINDING 1 — FIX-FIRST · X3_addad bibliographic-citation false positives **File**: `W3-X3/X3_addad_candidates.json` **Evidence**: The layer's sole marker is `الأضداد` (102 hits, 92 verses). Two of five sampled exemplars are confirmed bibliographic citations of the book title "الأضداد لابن الأنباري" embedded in Tabari's commentary text (1:1, 1:7), not identifications of an addad word in the Quran. The word "الأضداد" appears in Tabari as a book citation hundreds of times; the engine cannot distinguish "this Quranic word is an auto-antonym" from "Ibn al-Anbari's addad-lexicon is cited here." Estimated FP rate from sample: 2/5 exemplars = ~40%. **Law text**: correctly says "every row a CANDIDATE until read" ✓ — no over-claim. But at ~40% FP the candidate pool of 92 verses is misleading: roughly 37 entries are bibliographic noise, leaving ~55 genuine candidates. A scholar using this layer to find addad discussions would treat it as if it were a reliable hitlist when it is not. **Severity**: FIX-FIRST — the false-positive rate makes the count misleading (mission criterion met). **Fix path**: The engine should exclude "الأضداد لابن الأنباري" and similar title-pattern contexts, or the meta should declare the known FP source and give a filtered candidate count. --- ### FINDING 2 — INFO · hadhf تقديره/والتقدير markers capture theological predetermination **File**: `W5-T5/hadhf_candidates.json` **Evidence**: The markers `تقديره` (1,318 hits) and `والتقدير` (1,118 hits) together account for 2,436/5,509 = 44% of all hadhf hits. Exemplar 10:3 zamakhshari "من قضائه وتقديره" = "from His decree and His predetermination (taqdir in the theological sense)"; 10:4 zamakhshari same context. These use تقدير as divine decree, not the grammarian's estimated-supplied-word sense. Estimated FP rate for these two markers: ~25% from the 8 sampled exemplars (2 FP out of 8, both in zamakhshari's commentary on Sura 10 introduction). **Law text**: "every row a CANDIDATE until read" ✓ — correctly labeled. **Severity**: INFO — candidates declared; the reading bench is the resolution mechanism. But the bench should expect heavy disambiguation burden for تقديره/والتقدير across theological commentaries. --- ### FINDING 3 — INFO · R6_naskh منسوخة/نسخت captures non-abrogation uses **File**: `W6-R6/R6_naskh_candidates.json` **Evidence**: Exemplars at 1:2 and 1:3 zamakhshari: "يجعلون استعمالها كالشريعة المنسوخة" — "they treat using it [a grammatical case-ending form] like an abrogated law." This is Zamakhshari describing grammatical convention, not Quranic abrogation. 2 of 5 exemplars confirmed FP (~40%). The marker `منسوخة` appears in secular/grammatical contexts routinely. **Law text**: "every row a CANDIDATE until read" ✓. **Severity**: INFO — candidates declared, but bench workload is substantial at ~40% FP. The 371-verse candidate set likely contains ~148 non-abrogation hits. --- ### FINDING 4 — INFO · R7 المحكم captures "the precise Book" sense **File**: `W6-R7/R7_muhkam_mutashabih_candidates.json` **Evidence**: Exemplar 10:1 muyassar "هذه آيات الكتاب المحكم" = "these are the verses of the Precise Book" — المحكم here is an attributive ("the perfected one") referring to the Quran as a whole, not the technical muhkam/mutashabih classification from 3:7. Similarly some tabari contexts. **Law text**: ✓ candidates declared. **Severity**: INFO. --- ### FINDING 5 — INFO · R5 makki/madani marker hits hadith text **File**: `W6-R5/R5_makki_madani_candidates.json` **Evidence**: Exemplar 1:1 qurtubi: labeled for مدني but context is "قال الله تعالى حمدني عبدي" (a hadith quotation discussing the prayer response), not a Makki/Madani classification statement. **Law text**: ✓ candidates declared. **Severity**: INFO. --- ### FINDING 6 — INFO · R9 asbab سبب نزول matches "cause of rain's descent" **File**: `W6-R9/asbab_census.json` **Evidence**: Exemplar 10:24 wasit: "بسبب نزول المطر من السماء" = "because of the descent of rain from the sky" — سبب نزول here means "cause of the rain's fall," not an occasion of revelation. Estimated FP rate from 5 exemplars: 1/5 = ~20%. **Law text**: ✓ candidates declared. **Severity**: INFO. --- ### FINDING 7 — INFO · M9 اقوال marker is universal Arabic **File**: `W7-M9/difference_registry.json` **Evidence**: The marker اقوال (7,085/13,065 hits = 54%) is an everyday Arabic word meaning "statements/sayings" used in all contexts, not exclusively scholarly-difference markers. Any commentary sentence summarizing "the scholars' sayings" or "the opinions" triggers it. The 4,940-verse footprint (79% of the Quran) reflects ubiquity of the word, not ubiquity of genuine registered disagreements. **Law text**: ✓ candidates declared ("ikhtilaf formulas located"). **Severity**: INFO — the layer correctly transmits and the bench resolves. --- ### FINDING 8 — CLEAN · drift_harvest (X9) exemplars genuine **File**: `W3-X9/drift_harvest.json` **Evidence**: All 5 sampled exemplars show legitimate jahiliyya-usage or original-sense discussions: 1:1 baghawi "واصله في اللغة الطريق الواضح" (original sense of صراط), 1:2 qurtubi "قد قالوه في الجاهلية للملك" (pre-Islamic use of الرب). Markers are specific phrases (في الجاهلية, كانت العرب, واصله في اللغة) with low ambient FP. CLEAN. --- ### FINDING 9 — CLEAN · gharib register (X1) rarity census mechanical **File**: `W3-X1/gharib_register.json` **Evidence**: Hapax census (1,958 lemmas) is a pure frequency count over the corpus — no FP risk. Sampled hapax lemmas (مَغْضُوب at 1:7, فُوم at 2:61, etc.) are correct. Tradition gharib-marks (103 hits / 92 verses) are a separate field from the frequency census. CLEAN. --- ### FINDING 10 — CLEAN · misreading_harvest (X10) transmits editorial corrections honestly **File**: `W3-X10/misreading_harvest.json` **Evidence**: Exemplars show "والصواب من المخطوطة" (the correct reading is from the manuscript) — these are Tabari's critical apparatus notes flagging variant manuscript readings, not "misreadings" in the mispronunciation sense. The law says "corrections the tradition ITSELF records" and these are exactly that: editorial corrections to commentary transmission. The label "misreading" is a slight framing stretch but the data is correctly scoped and the law is accurate. CLEAN. --- ## MECHANICAL-CHECKS TABLE | Check | Result | |---|---| | All engines SELFTEST GREEN | ✅ VERIFIED (all 11) | | All JSON totals match engine output | ✅ VERIFIED | | All law texts say "CANDIDATE until read" or equivalent | ✅ VERIFIED (all 11 layers) | | No layer claims confirmed counts where only candidates exist | ✅ VERIFIED | | hadhf confirmed_seed entry (1:1 basmalah) present | ✅ VERIFIED | | X3_addad FP rate makes count misleading (FIX-FIRST criterion) | ⚠️ YES — FIX-FIRST | | R6_naskh FP rate within "candidates declared" tolerance | ✅ INFO only | | Awaited acquisitions (addad lexicons, al-Sijistani) declared, never assumed | ✅ VERIFIED | --- *PANEL-P31-WAVEB-PROBES.md · PB1 · 2026-09-04*