feat(expert-distiller): cross-validated-search maturity 72/100

[wd041216-bit]

Summary

• Expert council of 5 domain specialists distilled from public sources and applied to the cross-validated-search project
• Code improvements driven by expert lenses: claim decomposition, multilingual conflict markers, concurrent provider calls, retry logic, calibration notes
• Maturity score: Breadth 17 | Depth 18 | Thickness 19 | Effectiveness 18 (Total 72/100)

Expert Council

Expert	Role	Domain Coverage
Isabelle Augenstein	Advocate	Fact-checking, multi-domain verification, explainability
Preslav Nakov	Reviewer	Propaganda detection, media bias, multilingual verification
Reihaneh Rabbany	Chair	Web retrieval agents, source credibility, evaluation methodology
Pepa Atanasova	Reviewer	Explainable fact-checking, stance detection, confidence calibration
Andreas Hanselowski	Skeptic	Multi-provider aggregation, RAG, agentic workflows

Code Improvements

1. Claim decomposition: decompose_claim() method splits compound claims into verifiable sub-claims with SubClaimResult dataclass
2. Multilingual conflict markers: Added Spanish, French, German, and Chinese conflict detection markers
3. Concurrent provider calls: ThreadPoolExecutor for parallel multi-provider search
4. Retry with exponential backoff: 3 retries with 0.5 * 2^attempt delay for provider resilience
5. Calibration note: CALIBRATION_NOTE field on VerificationResult and EvidenceReportResult documenting heuristic thresholds

Test plan

• All 74 existing + new tests pass
• python3 -m pytest tests/ -q — 74 passed
• Claim decomposition tests cover simple, compound, semicolon, period-split, short-fragment-filtered, and triple claims
• Multilingual conflict detection tests for Spanish, French, German, Chinese
• Concurrent provider call tests with ThreadPoolExecutor
• Retry logic tests with exponential backoff
• Calibration note output verification tests

🤖 Generated with Claude Code