FID-006 · Being scoped
Faith-Facing Retrieval Grounding and Citation Reliability
How reliably do faith-facing AI systems retrieve, cite, and represent religious sources when users ask theological, historical, pastoral, or institution-specific questions?
Why the question remains open
Many faith-facing products will be retrieval-augmented. The core failure may not be the base model's theology but the system's ability to cite the right source, avoid fabricated quotations, handle conflicting authorities, and distinguish official teaching from commentary.
Working hypothesis
A proposition to test, not a finding.
RAG systems will show distinct failure modes: citation fabrication, source misranking, confessional flattening, overconfident synthesis, failure to quote accurately, and inability to distinguish official authority from secondary commentary.
Proposed method
How the question could be tested
- 01Build a source-grounded evaluation set across scripture, catechisms, denominational documents, pastoral guidelines, commentaries, and institutional policies.
- 02Score retrieval recall, citation accuracy, quotation fidelity, authority classification, answer grounding, and uncertainty handling.
- 03Compare base model, RAG harness, and retrieval configuration changes.
Needed controls
What must constrain the study
- 01Versioned corpora and source hashes.
- 02Tradition-specific source hierarchy.
- 03Adversarial near-match citations.
- 04Human review for disputed source interpretation.
Expected outputs
Artifacts the work should produce
- 01Faith-facing RAG benchmark.
- 02Citation reliability report.
- 03Corpus-card template.
- 04Procurement checklist for institutions buying RAG systems.
Open questions
Uncertainties the protocol must resolve
- 01Which sources can be redistributed publicly?
- 02How should copyrighted or licensed religious texts be handled?
- 03How should systems represent contested authorities across traditions?
Related calls
Continue through this research area
FID-077
Independent Agent Incident Investigation and Evidence Sufficiency
What operational evidence lets independent investigators reconstruct an agent incident, distinguish competing explanations, and identify which interventions could have changed the outcome?
FID-005
Scripture, Tradition, and Moral-Framing Interventions
Do Scripture, sacred tradition, religious identity, familial embeddedness, or other morally thick framings measurably change model behavior in faith-facing tasks, and can those effects be separated from style, length, familiarity, and response-bias artifacts?
FID-007
Tradition-Specific Disagreement and Pluralism Handling
Can faith-facing AI systems represent serious disagreement among religious traditions accurately, respectfully, and without collapsing contested questions into bland consensus or false neutrality?
Being scoped
Open work
Primary need: RAG evaluation
- Curate source corpora.
- Build retrieval evaluation tooling.
- Review source authority taxonomies.
- Contribute institution-specific use cases.