Apa yang berubah dalam TheoAI

v3.176.1 — corpus mencapai 617,931 rekod bergred merentasi 128 karya

IPSC v3.176.1 (June 2026). The corpus that powers TheoAI reaches 617,931 hadith records across 128 classical works, of which 480,350 are computationally graded (≈ 77.7%). New Tier-3 Evidence Map aggregates evidence at the statement level — every chain for a given statement, assembled across all 128 works into one view. Full matn-core re-embedding (588,737 re-embedded). Narrator denominator corrected to 68,299 indexed / 28,586 reliability-assessed. Bukhārī sahih+hasan convergence is 95.8%.

Asas Peta Bukti Peringkat-3

The data and embedding layers the Evidence Map is built on: multi-narrator (qālā/qālū) chain resolution, plus a full matn-core re-embedding of 588,737 records — all shipped as part of v3.176.1 (see top entry).

Nota pengeluaran corpus penuh

Pengukuhan kualiti data & penyahduplikasian perawi

Systematic narrator-identity cleanup across the rijāl graph: write-time NRS sanity validation, batch-job silent-failure tracking, and a multi-tier canonical-anchor deduplication workstream consolidating duplicate narrator records — so Theo resolves a chain to one narrator, not several near-duplicates.

Nota pengeluaran corpus penuh

Kedalaman corpus — pengambilan data biografi & tārīkh

Ingested the major biographical, tārīkh and ḍuʿafāʾ corpora — Ibn ʿAsākir’s Tārīkh Dimashq, al-Khaṭīb’s Tārīkh Baghdād, Ibn Saʿd’s Ṭabaqāt, al-Ḥākim’s Mustadrak, al-Dāraquṭnī, Ibn ʿAdī’s al-Kāmil, al-ʿUqaylī’s Ḍuʿafāʾ, and the major Mawḍūʿāt (fabrication) catalogs — growing the corpus toward ~128 works / 617,931 records. A dual-extraction pass recovered 13,213 embedded classical verdicts that single-pass ingestion had missed.

Nota pengeluaran corpus penuh

Peringkat penggredan petikan + infrastruktur integriti

Three tiers of citation 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.

Nota pengeluaran corpus penuh

Perluasan korpus dipentaskan — +6 koleksi klasik, +12,660 PID diselesaikan oleh LLM

The corpus that powers TheoAI added 6 new classical collections (Ibn al-Sunnī ʿAmal, Hannād Zuhd, al-Quḍāʿī Shihāb, al-ʿUqaylī Duʿafāʾ, Ibn al-Mubārak Jihād, Ibn ʿAdī al-Kāmil) and 12,660 LLM-tiebroken narrator IDs — a cutover that has since shipped as part of v3.176.1 (see top entry).

Nota keluaran korpus penuh

Kitaran penentukuran semula metodologi

Recurring external-examiner cycle drove a v3.11 framing recalibration. The corpus added the _provenanceDisclosure block to corpus-v3/manifest.json so AI-involvement scope travels with every record, and adopted the standing provenance-discipline rules sitewide. The full audit-and-quality cycle (pillars, recalibrations, scholar-collaboration program) is published.

Baca halaman amalan audit

Penumpuan Bukhārī diperbetulkan: 97.3% → 95.8%

The v3.8 cluster rebuild surfaced inflated supportingChain counts that were producing an inflated 97.3% sahih+hasan rate on Sahih al-Bukhārī. Honest de-inflation produces 95.8%. Marketing site updated; underlying data unchanged.

v3.5 → v3.10 tolakan integriti korpus

IPSC mula-mula digunakan pada Azure AI Search

First production deployment of the corpus to Azure AI Search across 13 customer-facing alias indexes (1,569,379 documents total). Glossary v1 (730 canonical hadith-science terms) shipped same day across 3 additional tier indexes.

Tentang Glosari v1

Saluran paip kritikan matn (Fasa A–G)

Two-pass content-criticism architecture: deterministic string-op scan on all 437,740 matns followed by multi-tier Eve-Theology™ Fusion v5 reasoning (lightweight triage → Eve-Theology™ F5/reasoner detail → frontier-tier scholar-grade analysis). 815 chain-matn conflicts identified. 808 scholar-ready defense documents generated.

Keluaran korpus awal (v3.0)

IPSC v3.0 ships: the first public corpus across the classical collections, with a 12-tier narrator reliability scheme anchored to Ibn Ḥajar’s Taqrīb al-Tahdhīb and a five-condition Ibn al-Salāh grading engine. (Corpus has since grown to 617,931 records across 128 works — see top entry.)