HadithCompass
Library
Back to Library

Methodology

How this engine reasons.

Results reflect the selected methodology and ethical framework, not objective moral truth.

1. Corpus

Narrations are ingested on demand from the open fawazahmed0/hadith-api dataset, which itself sources from sunnah.com and other public editions. Eight Sunni collections are supported. Shia collections (al-Kafi, Tahdhib, etc.) are stubbed for a later module. Each record stores Arabic, English, grade, book, hadith number, and an external source URL.

2. Classification

Each narration can be classified by a Lovable AI Gateway call (Gemini 3 Flash) using strict structured output (function calling). The classifier returns:

  • Topics — one or more of 19 categories, each with a confidence 0–1.
  • Target audience — women, men, both, general, children, enslaved, non-believers, with direct/indirect and confidence.
  • Action type — command, prohibition, punishment, praise, warning, legal rule, social norm, metaphor, historical description.

The classifier is instructed to be descriptive, not moralizing.

3. Ethical scoring

Each narration is scored independently under each active framework on six dimensions: equality, autonomy, coercion, violence, patriarchy, restriction. Scores are 0–5 and explicitly framework-relative. The engine never collapses these into a single "moral truth" score. Every score carries an explanation, alternative interpretations within the same framework, and a confidence value.

See the full scoring rubric — it defines what 1–5 means for each dimension so you can interpret the numbers accurately.

4. Active frameworks

Human Rights (UDHR-based)

Evaluates narrations against the 1948 Universal Declaration of Human Rights: equal dignity, non-discrimination, freedom of movement, freedom from cruel punishment, equal protection under law.

  • · Persons of all genders, religions, and statuses hold equal rights.
  • · Coercive religious obligation that conflicts with civil rights is flagged.
  • · Corporal punishment is treated as a rights concern regardless of legal context.

Islamic Traditionalist

Reads the hadith through classical Sunni jurisprudential frameworks (the four madhahib). Treats authentic narrations as normative legal sources. Gendered roles are read as divinely ordained complementarity rather than inequality.

  • · Authentic hadith carry legal weight (hujjiyya).
  • · Gendered roles are framed as complementarity, not hierarchy.
  • · Hudud punishments are valid within their classical procedural safeguards.

Islamic Reformist

Modern reformist readings (Fazlur Rahman, Amina Wadud, Khaled Abou El Fadl, etc.) that distinguish universal Qur'anic ethics from historically-contingent hadith rulings. Reads patriarchal-seeming narrations through their 7th-century context.

  • · The Qur'an's ethical trajectory takes priority over isolated narrations.
  • · Hadith must be read in their historical and social context.
  • · Gender equality and human dignity are core Islamic values that contextualize specific rulings.

Liberal Egalitarian

Secular liberal framework (Rawlsian / Mill-influenced) prioritizing individual autonomy, equal basic liberties, and consent. Religious tradition is one voice among many, not a source of binding norms.

  • · Individual autonomy is the foundational value.
  • · Consent is required for any restriction on persons.
  • · Religious authority does not override secular rights.

Feminist Ethics

Foregrounds gender asymmetries, power, and bodily autonomy. Draws on Islamic feminism (Mernissi, Wadud, Barlas) and broader feminist ethics.

  • · Gender asymmetry in obligations or punishments is itself significant.
  • · Bodily autonomy and freedom of movement are core ethical concerns.
  • · Patriarchal interpretive history is distinguishable from the texts themselves.

Historical Contextualist

Descriptive historical lens. Reads narrations as artefacts of 7th–9th century Arabian and early Islamic society, without prescriptive judgment.

  • · Norms reflect their time and place; presentism is avoided.
  • · Authenticity grading is itself a historical-critical question.
  • · No verdict of 'moral' or 'immoral' is rendered.

Comparative Religious Ethics

Reads each narration against analogous norms in Jewish, Christian, Hindu, Buddhist, and secular ethical traditions of comparable era and region.

  • · Religious norms are compared, not ranked.
  • · Historical baseline matters: 7th-century Arabia vs. 21st-century pluralism.

5. Safeguards

  • Descriptive analysis is separated from normative claims.
  • Scholarly disagreement is surfaced via the "alternative interpretations" field on every score.
  • Hadith grade is always shown.
  • Confidence is shown for every classification and every ethical score.
  • The platform does not rank religions, generate hate speech, or claim objectivity.
  • Quantitative analysis is presented as a complement to — never a replacement for — scholarship.

6. Known limitations

  • Translation choice (e.g. Mohsin Khan vs. Sahih International) materially affects classification.
  • The classifier has no memory of asbab al-wurud (occasions of narration) unless they are in the text itself.
  • Duplicate-narration normalization is not yet implemented (planned via pgvector).
  • Authenticity grading is itself disputed; we present the upstream dataset's grade verbatim.
  • Only a sample of each collection is typically ingested for cost reasons.