MindHYVE.ai

Substrate · Primary-source corpus

IPSC.

The Islamic Primary Source Corpus — MindHYVE.ai's proprietary, citation-graded corpus of classical hadith. Every chain parsed. Every narrator graded. The evidential substrate beneath TheoAI — and one of the most defensible technical achievements in the portfolio.

Corpus
617,931 records
128 classical works
Graded
480,350
77.7% · independently
Narrators
68,299 indexed
28,586 NRS-assessed
01The analytical layers

Not a search index. A graded corpus.

IPSC spans the Six Books, the major Musnads, Muṣannafs and Muʿjams, the rijāl and tārīkh literature, and the classical fabrication catalogs. Across all of it, the same discipline is applied to every record.

01
Isnād parsing
Chain of transmission

Every report is parsed into an ordered sequence of narrator positions — the isnād. The chain is the unit of analysis: where a narration came from, who carried it, and how it reached the compiler.

02
Narrator Reliability Scoring (NRS)
68,299 indexed · 28,586 assessed

Each narrator is resolved to a canonical identity and scored on a twelve-tier reliability scale, anchored to the classical rijāl literature (Ibn Ḥajar’s Taqrīb al-Tahdhīb and the wider tradition). 28,586 entries carry reliability assessments within a 68,299-narrator index.

03
Authenticity grading
Five conditions · weakest-link

Grades are computed from the classical five conditions of authenticity, applied per chain — a chain is only as strong as its weakest position — with taqwiyah (corroboration) accounted for. Categories run ṣaḥīḥ → ḥasan → ḍaʿīf → mawḍūʿ.

04
Matn criticism
Computational ʿilal

Beyond the chain, the text itself is examined: computational detection of shudhūdh (irregularity), idrāj (interpolation), anachronism, and known fabrication patterns — the first matn-criticism pipeline operating at full-corpus scale.

05
Citation grading tiers
tier-1 → tier-3

Three tiers of citation rigor: tier-1 (32 checks, available now) for everyday evidence, escalating to tier-3 per-citation verdicts for aḥkām-level defensibility. The asker chooses the standard of proof the question deserves.

02The Evidence Map

Ask what the evidence is for a statement.

The Evidence Map is the corpus's newest capability — and the one that makes it a tool rather than a table. It answers, for a specific statement attributed to the Prophet ﷺ, what the evidence actually is: every chain, assembled and weighed.

Statement-keyed

The unit of inquiry is the statement, not a single chain. Every chain for a given report is assembled across all 128 works into one picture.

Cross-collection corroboration

Parallel transmissions are surfaced across collections — making mass transmission (tawātur) visible, and showing how independent paths reinforce one another.

Per-position confidence + critic verdicts

Each narrator position carries a resolution confidence, and verbatim critic verdicts are kept distinct per source — al-Mizzī is never silently folded into Ibn Ḥajar.

Dual-direction confidence

A chain fails on its weakest link; a statement strengthens on its strongest corroborating chain. The Evidence Map reads the way a muḥaddith reads — strictness on the path, generosity across paths.

03Quality & honesty

We publish the corrected number, not the flattering one.

The hardest thing to engineer into a corpus like this is honesty under pressure. IPSC is built to be auditable — and to disclose what it is, and what it is not.

Convergence with the tradition

Bukhārī 95.8%

Independently computed ṣaḥīḥ+ḥasan rate on Ṣaḥīḥ al-Bukhārī — recomputed from the chains, then honestly de-inflated where our own data warranted it. We publish the corrected number, not the flattering one.

Regression discipline

40 / 40 + 7 / 7

Every release passes a fixed regression suite anchored to canonical grades, plus cross-field integrity checks. Bulk changes are gated; no release ships on a non-GO audit.

Adversarial audit

NAQD + per-record verification

Heuristic flags are not trusted on their own — they are verified per record against the source text. Repeatedly, gap heuristics that looked like real breaks collapsed under inspection. The verification machinery is the deliverable.

Honest provenance

Applied AI, not a muftī

IPSC is an applied-AI and data-engineering work product grounded in classical rijāl methodology. It is not classical mujtahid scholarship and does not issue fatwā. Its outputs are auditable classifications meant to support — never replace — qualified scholarly judgment.

04How it is served

On Eve-Grid. Reasoned over by Eve-Theology.

IPSC lives on the Eve-Grid™ Azure substrate and is reasoned over by the Eve-Theology f5/reasoner. The corpus is the data layer; the reasoning model and the Evidence Map turn it into answers.

Azure AI Search

Served from tiered vector indexes — public, research, and scholar — with field-level exposure controls so each audience sees the appropriate depth.

3,072-dimension embeddings

Semantic search runs over matn-core embeddings, so a question like “find me terms about chain weakness” returns the right narrations and rijāl, not keyword matches.

Eve-Theology f5/reasoner

The theology compound model reasons over IPSC — retrieving graded chains, composing evidence, and producing Evidence-Based Opinions (Taḥqīq) with every citation traceable to a graded chain.

Powers TheoAI

IPSC is the evidential spine of the TheoAI Operating System — Theo (consumer), Majlis (scholarly workspace), and Mīzān (Islamic-finance governance) all draw on the same corpus.

05Explore the corpus

IPSC is the evidential spine of TheoAI: Theo cites it, the Evidence Map renders it, and every answer traces back to a graded chain of transmission. The corpus is published in full — methodology, provenance, and known limits — on its own scholarly surface.