Summary

• Add structured_author_fetcher.py: WOS Starter API as primary Phase 2 author source, with intelligent fallback chain (WOS+PDF → WOS+S2 → S2+collector)
• Add author_name_utils.py: robust name normalization and 5-rule fuzzy matching across WOS/S2/PDF formats (handles initials, accents, inverted order)
• Merge strategy: WOS+PDF preferred → WOS+S2 when PDF unavailable → pure S2+collector fallback (identical to before when WOS not configured)
• Add WOS API Key field to UI config with save/load wiring; fix ConfigUpdate missing wos_api_key so key was silently discarded
• Fix citation count parsing for European locale formats (5.710, 5 710)
• Add per-paper author debug logging and _author_debug.json output for testing
• MinerU PDF parsing changed from first-page only to full document; prompt updated to prohibit fabrication/reference mixing

AI coding brief

Original request: Integrate WOS Starter API as structured author extraction for Phase 2. WOS gives accurate author lists but no affiliations — affiliations come from PDF (MinerU) when available, otherwise S2. When WOS is not configured, behavior should be identical to before.

Manual interventions:

• Corrected merge priority: WOS+PDF > WOS+S2 > S2 (not always-query-both)
• Fixed wos_api_key missing from ConfigUpdate in main.py — key was silently discarded on UI save
• Fixed browser cache causing old HTML (without WOS field) to be served — required hard refresh
• Clarified that actual Phase 2 pipeline is task_executor._run_new_phase2_and_3, not author_searcher.py

Retro: Specify upfront which code path is the production pipeline vs legacy/dead code. Also define multi-source merge priority rules before implementation to avoid rework.

Test plan

• Fill WOS API Key in UI → save → verify config.json persists key
• Run pipeline: Phase 2 log shows 📋 WOS 结构化作者提取已启用（WOS + S2 双源融合）
• With PDF: source label wos+pdf, affiliations from PDF
• Without PDF: source label wos+s2, affiliations from S2 where name-matched
• Without WOS key: log shows ⚪ 未配置 WOS Key, behavior identical to before
• Check _author_debug.json in result folder for per-paper breakdown
• European locale citation counts (e.g. 5.710) parsed correctly

🤖 Generated with Claude Code