Methodology

1 · Qur'an only

The first rule is the hardest and the most important: the Qur'an is read as its own witness, and nothing is admitted from outside it. This is not hostility to other sources — it is the only way to settle a dispute about those sources without begging the question. If the disagreement is whether meaning may come from beyond the text, then importing meaning from beyond the text to resolve it would decide the case before hearing it.

Accordingly, the following are excluded as evidence throughout every document on this site:

• Tafsīr — no commentary tradition, classical or modern.
• Hadith and sīra — no reports, no biography, no occasions-of-revelation narratives.
• History and archaeology — no external chronology, no manuscript dating, no inscriptions.
• Secondary scholarship — no grammarians, theologians, or academics cited as authorities.

2 · Arabic linguistic

With the field cleared, the only instrument left is the language. Claims are tested at four levels: the root (the triliteral source and its semantic field), the morphology (the pattern a word is built on — participle, verb form, plural, diptote), the syntax (the grammatical role a word plays in its sentence), and the cross-text (how the Qur'an itself uses a term elsewhere).

Above these sits one master rule, and the project derives it from the Book's own behaviour rather than asserting it:

This is why, for example, sujūd is figurative for stars (22:18) but bodily when commanded to humans beside washing, ranks, times, and a direction (4:102): the context, not the lexicon alone, sets the meaning each time.

3 · Agnostic neutrality — and its exact limits

The analysis is agnostic by design. It does not begin on either party's side, and it does not bend a reading to please the person who requested it. Its remit has three parts, and it never exceeds them:

1. It rules on where the grammar points — which reading the Arabic supports and which one strains it.
2. It rules on which posture is exposed to the enemy's documented tactics — a structural observation, not a charge.
3. It does not rule on whether any person is guided or deceived, saved or lost. That is the limit, and it is absolute (see §6).

One fairness rule binds both parties equally and is applied without exception:

4 · How a claim is adjudicated

Every claim runs the same five steps, in order:

1. State it at its strongest. The claim is put in the most defensible form its holder would give it — never a straw version.
2. Lay out the readings the grammar allows. Each alternative the Arabic genuinely permits is set down, including the one that helps the other side.
3. Name what strains. The reading that the morphology or syntax cannot carry is identified, with the exact feature that breaks it.
4. State the verdict as direction. "Where the text points" — leaning, strong, or undetermined — with key verses anyone can open.
5. Add the exposure note. Which of the enemy's tactics the position is open to — flagged as structure, not as a ruling on the holder.

5 · Why this is not "just an AI saying so"

A fair objection meets any analysis produced with a language model: the tool responds to prompts, can be steered toward a desired answer, can be biased toward pleasing whoever is asking, and can be wrong. That objection is correct. It is not waved away here — it is answered, by construction.

• Every verdict is verse-traceable. Nothing rests on the model's authority. Each conclusion is tied to cited verses; the reader checks the Arabic in a muṣḥaf and judges the reading directly. The instrument proposes; the text disposes.
• The steering is shown, not hidden. The complete record — every prompt that shaped the work, and the model's own reasoning — is published in Behind the Curtain. If a question pushed toward an answer, you can see it push. Transparency about the prompting is stronger proof than a claim of neutrality.
• The sources are tamper-evident. Each document is hash-sealed (§7), so any later alteration — even one word — is detectable.
• The method forbids the usual entry-point of bias. Outside sources are where a model's training-bias and a user's preferred tradition most easily enter. By admitting none, the surface for that bias is narrowed to the text and its grammar.

6 · The binding limit

The one thing this study will never do is declare any person guided or deceived, "of ar-Raḥmān" or "of the enemy." Faith is بِٱلْغَيْب, of the unseen (2:3); the state of a heart is known to God alone. The analysis maps language and structure — it cannot read a soul, and it does not pretend to. The very uncertainty that makes faith a test forbids every party's certainty about another's standing — and forbids this study's, and the reader's, in equal measure.

7 · Provenance — sealing the sources

Each source document carries a SHA-256 fingerprint, published on the home page. A cryptographic hash changes completely if a single character of the source changes, so the fingerprint is a tamper-evidence seal: it lets anyone confirm that what they are reading is exactly what was written, unaltered by deployment, styling, or later editing.

Hashing is done per document, not over the whole conversation, for a practical reason: an individual file hash can be re-checked after the work is wrapped in HTML and CSS — the styling does not touch the hashed source — and each file can be re-sealed independently as it is finalised. The fingerprint certifies the source markdown; the source files are published next to the site so the reader hashes those directly, in the browser or from a terminal.

Go to verification →