Method
Method & transparency
Not every pattern is proof. Every meaningful claim deserves careful investigation.
- Text
- The Quran text is the Tanzil Project's verified Uthmani and simple-clean Mushaf (Hafs reading, 6,236 verses, Kufan count), used verbatim.
- Words, roots and lemmas
- Word units and grammatical tags come from the Quranic Arabic Corpus (v0.4) morphology with mustafa0x corrections. A word's root and lemma are read from its stem segment.
- Counting rules
- Every analysis records the unit (root, lemma, surface form, phrase), filter (singular/plural, case), scope and normalisation (diacritics removal, alif unification). Change the rule and the number changes — so the rule is fixed before the claim.
- Reproducibility
- All numbers are produced by the open analysis engine (version 1.0.0). The AI never writes numbers; it describes what to count. Every recipe has a hash and yields the same result in the Lab.
- Chance testing
- When many relationships are tried, striking results appear by chance. Equality claims are tested against word pairs of similar frequency, co-occurrences against permutation tests, number properties against base rates. The number of attempts is reported too.
- Committee
- Approval needs at least 3 human reviews, 80% agreement, approval from a relevant expert and a verifier AI report. The AI's recommendation is not a vote; the AI cannot approve its own output.
- Verified vs. interpretation
- Every publication lists separately which statements were verified by computation and which remain interpretation. Committee approval does not mean scientific certainty or that every religious interpretation is beyond dispute.
- Translation
- Translations are shown with their source and never replace the Arabic original. Machine translation drafts are not published without human review; each language is approved for publication separately.
Rejected claims
Claims examined by the committee and rejected with reasons. This list is what gives approved discoveries their value.
No rejected claims yet.