Laboratory
Every experiment here is a declared, frozen specification executed deterministically against the checksummed corpus. No models, no randomness, no hidden search space. Failed runs are preserved, and superseded results are kept alongside their replacements (see data/derived/experiments/).
Corpus snapshot & history
These were on the dashboard, presented as if something were happening. Nothing is: this app is a static build over pre-computed JSON, and the counts move only when the pipeline is re-run. They belong here, labeled for what they are (D-081).
Corpus snapshotstatic
Counts as of the last pipeline run (2026-08-17 10:47). They do not change while the app is open — nothing runs in the background. Re-run pnpm data:pipeline and rebuild to move them.
Pipeline historyrecorded
Every line is an event that actually happened, with the timestamp recorded for it in the manifests and the experiment ledger. This list grows only when the pipeline runs.
Analysis coverage
All 14 layers
Share of the system's declared analytical work that is complete. This measures the machine, never the Qur'an — a text is not a thing a machine finishes.
Integrity checks
Recomputed at build time from the sealed corpus, 2026-08-21 04:29Z. They do not re-run while the app is open.
Tools
The analytical surfaces live here rather than in the top navigation: they are instruments for examining the corpus, not places to read it.
Discovery agents
Deterministic candidate generators now run over the corpus — mutashabihat families with exact difference maps, rarity censuses, and opening/closing correspondence pointers — each automatically annotated by skeptic checks and queued for human review. Open the review queue →
Experiment ledger
| ID | Title | Method | Status | Finished |
|---|---|---|---|---|
| EXP-0001 | Exact repetition census | repetition_census | completed | 2026-07-31T02:31:56Z |
| EXP-0002 | Counting-rule matrix: يوم (day) family vs the popular 365 claim | token_count_matrix | completed | 2026-07-31T02:31:56Z |
| EXP-0003 | Counting-rule matrix: شهر (month) family vs the popular 12 claim | token_count_matrix | completed | 2026-07-31T02:31:56Z |
| EXP-0004 | Root rank-frequency distribution | root_frequency | completed | 2026-07-31T02:31:56Z |
| EXP-0005 | Exact repetition vs word-shuffled surrogates (first null-model test) | repetition_vs_null | completed | 2026-07-31T03:03:08Z |
| EXP-0006 | Consecutive-ayah lexical cohesion vs shuffled-order surrogates | adjacency_cohesion_vs_null | completed | 2026-07-31T03:03:30Z |
| EXP-0007 | Repetition profile: Qur'an vs classical Arabic comparison texts | crosstext_repetition_profile | completed | 2026-07-31T03:36:52Z |
| EXP-0008 | Battery audit: twelve popular numeric claims under three fixed rules | claims_battery | completed | 2026-08-14T07:45:07Z |
| EXP-0009 | Muqatta'at: do the opening letters run high in their own surahs? | muqattaat_frequency | completed | 2026-08-14T07:45:15Z |
| EXP-0010 | Rhyme structure: verse-final ending runs vs shuffled order | rhyme_structure | completed | 2026-08-14T07:45:24Z |
| EXP-0011 | Ring composition scan: mirror symmetry across all long surahs | ring_scan | completed | 2026-08-14T07:45:39Z |
| EXP-0012 | Base rate of exact count equality: what an equality claim has to draw on | equality_background | completed | 2026-08-15T05:44:52Z |
Existing-claim audits
Popular numerical claims, reproduced or bounded under explicitly declared counting rules (MASTER_PROJECT §14). Corrections are offered gently; a reproduced count is a fact about counting, not proof of anything beyond it.
CLM-0001 — يوم (day) occurs 365 times COUNTING_RULE_DEPENDENT
The famous 365 count is real under one natural, fully-declared counting rule (singular unsuffixed forms, clitics stripped) and false under five other natural rules (217–475). This registry records the exact rule that works, the exact forms it counts, and why no statistical or theological conclusion follows from a rule-dependent count.
CLM-0002 — شهر (month) occurs 12 times COUNTING_RULE_DEPENDENT
The 12 count is real under the same declared singular-unsuffixed rule that yields 365 for يوم, and false under the other natural rules tested (4, 10, 21). The registry records exactly which five word-forms make up the 12, and why no conclusion beyond rule-dependence is warranted.
CLM-0003 — Angels and devils each occur 88 times COUNT_CONFIRMED_INFERENCE_NOT_SUPPORTED
The count is exactly right: angels 88, devils 88. The conclusion usually drawn from it is not supported — the Qur'an contains 1,640 such exact ties among common words, three other words also occur 88 times, and unrelated Arabic books produce ties at the same rate.
CLM-0004 — Muhammad ﷺ is named four times HOLDS_UNDER_TWO_DECLARED_RULES
Reproduced under two of three declared rules (surface and lemma): muhammad: surface 4, lemma 4, root 68..
CLM-0005 — Iblis is named eleven times HOLDS_UNDER_ONE_DECLARED_RULE
Reproduced under the lemma rule: إِبْلِيس = 11 (iblis: surface 0, lemma 11, root 5).
CLM-0006 — Zakah occurs 32 times HOLDS_UNDER_ONE_DECLARED_RULE
Reproduced under the lemma rule: زَكاة = 32 (zakah: surface 0, lemma 32, root 59)..
CLM-0007 — This world and the hereafter each occur 115 times COUNTING_RULE_DEPENDENT
Half-reproduced: dunya = 115 exactly under the lemma rule, but akhira = 155 under the same rule because the lemma آخِر covers every 'last/other' usage, not only the definite noun 'the hereafter' (dunya: surface 110, lemma 115, root 133; akhira: surface 0, lemma 155, root 250)..
CLM-0008 — Day occurs 365 times (battery recheck) COUNTING_RULE_DEPENDENT
Not reproduced under any of the three battery rules (yawm: surface 258, lemma 475, root 475).
CLM-0009 — Life and death each occur 145 times DOES_NOT_HOLD_UNDER_DECLARED_RULES
Not reproduced under any declared rule (hayat: surface 0, lemma 76, root 189; mawt: surface 35, lemma 50, root 165).
CLM-0010 — Man and woman each occur 24 times DOES_NOT_HOLD_UNDER_DECLARED_RULES
Not reproduced under any declared rule (rajul: surface 13, lemma 29, root 73; imraa: surface 0, lemma 27, root 38); the lemma rule gives 29 vs 27..
CLM-0011 — Sea 32, land 13 — the 71%/29% ratio DOES_NOT_HOLD_UNDER_DECLARED_RULES
Not reproduced under any declared rule (bahr: surface 24, lemma 41, root 42; barr: surface 19, lemma 22, root 32); the lemma rule gives 41 vs 22 (65%/35%)..
CLM-0012 — Say and they-said each occur 332 times DOES_NOT_HOLD_UNDER_DECLARED_RULES
Not reproduced under the declared bare-surface rule (qul: surface 298, lemma 0, root 0; qalu: surface 250, lemma 0, root 0); the popular figure requires counting clitic-attached variants (فقل, وقالوا…), a rule the claim never states..
CLM-0013 — Muqatta'at letters run high in their own surahs PARTIALLY_SUPPORTED
Tested in EXP-0009 with one pre-declared statistic: 20 of 29 initialed surahs sit above the non-initialed median for their own letters (one-tailed binomial p = 0.031).
CLM-0014 — Exact word-count equalities are evidence of design REFUTED_AS_STATED
Exact ties between word counts are common in every book. They only carry weight when the pair is named before anyone counts.
Statistical machinery (MIZAN)
Permutation testing against declared surrogate classes is now live: a repo-owned deterministic generator (splitmix64, pinned by test vectors), surrogate corpora that document exactly what they preserve and destroy, primary endpoints declared in the committed spec before execution, add-one p-values, and full null samples persisted with every result. EXP-0005 and EXP-0006 are the first applications. Word-count equality claims (like the 365) remain honestly untestable this way — order-destroying surrogates preserve every count by construction — and their reports say so.
What is deliberately absent
No matched real-text baselines yet (Classical Arabic prose and poetry are the §13.5 follow-up), so permutation results are worded strictly as surrogate-class comparisons. No AI-generated hypotheses exist anywhere in this build. “Proof of divinity” language is prohibited by policy in every part of this system.
