How to Cite the IPSC

Tier-3 Evidence Map foundations: multi-narrator (qālā/qālū) chain resolution and a full matn-core re-embedding of 588,737 records — the data and embedding layers the Evidence Map is built on.

Data-quality hardening and narrator deduplication: write-time NRS sanity validation, batch-job silent-failure tracking, and a multi-tier canonical-anchor deduplication workstream consolidating duplicate narrator records across the rijāl graph.

Corpus depth — biographical & tārīkh ingests (Ibn ʿAsākir Tārīkh Dimashq, al-Khaṭīb Tārīkh Baghdād, Ibn Saʿd Ṭabaqāt, al-Ḥākim Mustadrak, al-Dāraquṭnī, Ibn ʿAdī al-Kāmil, al-ʿUqaylī Ḍuʿafāʾ, major Mawḍūʿāt catalogs) growing the corpus to ~123 works / 617,931 records. Dual-extraction pass recovered 13,213 embedded classical verdicts missed by single-pass ingestion.

Citation grading tiers (three tiers of rigor; tier-1: 32 checks live, tier-2/3 for deeper aḥkām-level defensibility) and a five-layer integrity infrastructure with twelve mechanical bug-prevention rules — Arabic-regex safety, LLM bulk-apply preflight gates, a provenance-flag registry, write-time NRS sanity validation, and cascade-coverage lints.

LLM PID tiebreaker (12,660 PIDs assigned via Eve-Theology f5/reasoner batched 25-narrator prompts; 47.4% → 69.5% PID resolution on v3.19+ ingests). Ship-blocker remediation (2,168 records in Mawḍūʿāt / Tanzīh / Fawāʾid capped to very-weak via _naqd3Override). Phase 9 partial: 7,364 new matn embeddings, cluster merge 39.2%. NRS-Taqrīb reconciliation: 6,034 confirmed, 1,345 mismatches flagged. All hard gates green. Full changelog.

Citation-cascade refinement (v3.13–v3.18) and 6 new Phase 2-A primary collection ingests (Ibn al-Sunnī ʿAmal 770, Hannād Zuhd 1,429, al-Quḍāʿī Shihāb 1,497, al-ʿUqaylī Ḍuʿafāʾ 2,103, Ibn al-Mubārak Jihād 268, Ibn ʿAdī Kāmil 2,174 = +8,241 records). v3.19 OpenITI→IPSC ingestion pipeline. v3.25.1 fixed 5 pipeline robustness bugs.

Vector embeddings refresh (448,237 matns re-embedded; previous embeddings were on broken-matn data). Kanz al-ʿUmmāl re-ingestion: 14,603 records newly attributed via OpenITI symbol parser. NAQD-3 fresh-embedding re-run: 1,851 findings (300 critical, 638 high) — the previous near-zero N3-CC contradiction count was an artifact of broken embeddings.

External-examiner cycle drove a v3.11 framing recalibration. Added the _provenanceDisclosure manifest block so AI-involvement scope travels with every record, established the standing provenance-discipline rules, and closed NAQD-1 V1 (112 → 0 findings). Audit practice & scholar-collaboration program.

Corpus integrity push: temporal-issue field regeneration (256K false positives removed, 127K real issues found), arabicText_normalized cleanup (35,982 records), CDN-attribution flagging. v3.7 first-attempt PID validator failed regression (12 T10+ violations); v3.9 multi-stage validator (5 structural pre-filters + LLM tiebreaker) captured 24,485 safe PID swaps. v3.8 critical matn recovery from broken-regex corruption (449,028 matns restored from backup; bug survived two release cycles because regression didn't sample matn content).

First IPSC corpus deployed to Azure AI Search (10 tier indexes; +3 with Glossary v1; 1,569,379 docs total). 8 vector indexes (matnEmbedding × 3, narratorEmbedding × 3, defectEmbedding × 2). 7,652 narrators with classical scholarly quotations. Glossary v1 ship: 730 canonical hadith-science terms across 3 tier indexes.

Matn-criticism pipeline (Phases A–G): 304 Qur'an rulings, 271 anachronisms, 335 fabrication patterns, 1,301 mutawatir canon entries, prophetic linguistic baseline. Two-pass architecture: Pass 1 deterministic string-op scan (420,110 records, 11,441 flagged); Pass 2 multi-tier LLM (fast-tier triage → Eve-Theology f5/reasoner detail → deep-tier scholar-grade). 815 chain-matn conflicts identified.

Matn criticism pipeline (437,740 matns analyzed). Ilal cross-linking (87,844 hadith). Teacher-student graph (889,913 edges). Hawala chain splitting (8,258 records). Quran cross-reference (1,279,676 term matches). NRS: 27,099 to 27,118.

Initial corpus release. 449,415 hadith from 86 classical works. 27,099 NRS entries. Five-condition grading engine. Person ID resolution pipeline (6 iterations). Structured isnad parsing. Matn clustering (54,270 bag-of-words clusters).