Feature/scholarly attribution prevention 173

[Bogunrot]

Linked Issue

Closes #173

Problem Statement (The Bug)

The dnb-ai system generates scholarly attributions without validating them against authoritative databases. This allows fabricated, misattributed, or anachronistic opinions to be attributed to Islamic scholars — undermining scholarly integrity. Specifically:

• Opinions may be attributed to scholars who never held them
• Nuanced positions may be misrepresented as absolute
• Modern concepts may be anachronistically attributed to classical scholars
• False consensus may be invented where none exists
• There is no audit trail for attribution claims

This cannot be fixed with a local patch because the entire attribution pipeline lacks any validation layer — there is no database of known scholars, no opinion-to-scholar mapping, no temporal consistency checks, and no fabrication detection at all.

Solution Comparison and Decision (Why this approach)

Option	Description	Why Rejected
A: External API-only validation	Call an external LLM to verify each attribution	Adds latency, cost, and a single point of failure; the LLM itself can hallucinate — defeats the purpose
B: Simple keyword blocklist	Block known-bad attribution patterns	Treats symptoms, not the cause; cannot detect sophisticated fabrications or anachronisms
C: Offline-only validation	Validate at ingest time, not at query time	Misses runtime fabrications; no real-time blocking
D: Comprehensive local validation (chosen)	Build a local scholar database, opinion mapping, and multi-layer validation engine	Provides deterministic, auditable, real-time validation with no external dependencies; covers all six failure modes listed in the issue

The Change (Code modifications)

New files

scholarly_attribution.py — Core validation engine:

• Scholar biography database (16 major Islamic scholars with birth/death years, schools of thought, known positions)
• Opinion-to-scholar mapping from verified sources
• validate_attribution() — main entry point returning pass/block/warn
• check_temporal_consistency() — verifies scholar's era matches the topic
• detect_anachronism() — flags modern concepts attributed to classical scholars
• detect_nuance_flattening() — detects absolutist language on debated positions
• validate_consensus() — flags false consensus claims
• check_scholar_exists() — verifies scholar is in the database
• Full audit trail with AttributionAuditTrail

scholarly_attribution_api.py — FastAPI router:

Endpoint	Method	Description
/scholarly-attribution/validate	POST	Full text validation returning all attributions + verdicts
/scholarly-attribution/validate-single	POST	Single attribution check
/scholarly-attribution/scholars	GET	List all verified scholars
/scholarly-attribution/scholars/{id}	GET	Scholar biography detail

tests/test_scholarly_attribution.py — 30 tests covering:

• Scholar existence validation
• Known attribution detection
• Fabrication detection (invented scholars, false opinions)
• Anachronism detection (e.g., smartphones attributed to Ibn Khaldun)
• Temporal consistency
• Nuance flattening
• False consensus detection
• Full text validation with multiple attributions
• API endpoint integration tests (hitting real /scholarly-attribution/* routes)
• Audit trail generation
• Edge cases (empty input, special characters, mixed valid/invalid)

Modified files

• main.py — Added scholarly_attribution_router import and app.include_router(scholarly_attribution_router)

Compatibility Note (INTERFACE_VERSION)

INTERFACE_VERSION is not modified. The scholarly attribution system is a new, independent validation layer that does not alter existing APIs, data contracts, or shared interfaces.

Incidental Fixes

Fixed pre-existing syntax/import errors in the following files that blocked test collection:

• config.py — duplicate import
• corpus.py — missing stub functions
• query_optimizer.py — bad return values
• safety/output_check.py — broken syntax
• tafsir.py — import issues

Testing (Proving it works)

Test suite

tests/test_scholarly_attribution.py — 30 tests, all passing

Key test cases

• test_valid_known_attribution — Real scholar + real opinion → passes
• test_fabricated_scholar_detected — Invented scholar → blocked
• test_fabricated_opinion_detected — Real scholar + false opinion → blocked
• test_anachronism_detected — Modern concept attributed to classical scholar → flagged
• test_nuance_flattening_detected — Absolutist language on debated position → flagged
• test_false_consensus_detected — Invented consensus → flagged
• test_api_validate_endpoint — POST /scholarly-attribution/validate returns 200
• test_api_single_validate_endpoint — POST /scholarly-attribution/validate-single returns 200
• test_api_list_scholars_endpoint — GET /scholarly-attribution/scholars returns 200 with data

Before vs After

Before: FAILED (no scholarly_attribution tests existed)
After:  30 passed

Additional Notes

• The branch feature/scholarly-attribution-prevention-173 is based on main
• Scope is limited to the new attribution validation system — no changes to existing misinformation, tafsir, or safety modules beyond the incidental syntax fixes

[drips-wave]

@Bogunrot Great news! 🎉 Based on an automated assessment of this PR, the linked Wave issue(s) no longer count against your application limits.

You can now already apply to more issues while waiting for a review of this PR. Keep up the great work! 🚀

Learn more about application limits

[coderabbitai]

Important

• 🔍 Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: a08df7a9-e57e-448e-bd28-4098f5ff8950

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}