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.
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.
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.
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.
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.
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.
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.
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.
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.
Parallel transmissions are surfaced across collections — making mass transmission (tawātur) visible, and showing how independent paths reinforce one another.
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.
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.
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.
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.
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.
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.
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.
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.
Served from tiered vector indexes — public, research, and scholar — with field-level exposure controls so each audience sees the appropriate depth.
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.
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.
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.
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.