Best-effort exploration for incomplete research sources

Objective

Stop research agents from closing incomplete sources as insufficient_evidence without trying public proxies and nearby inferred rules.

Approach

• Incomplete papers/blogs/TradingView/RecessionAlert rows intake as research_campaign.
• Every node prompt injects an enforced best-effort policy.
• Semantic gates reject empty testable_families / variants_tested and require best_effort_attempts (at least three) before insufficient_evidence.
• Helper research.research_campaign.best_effort supplies FRED/Yahoo proxy maps and a default lag/lookback/threshold grid.
• Source-fidelity RecessionAlert specs remain unchanged; new runs publish labeled inferred experiments.

Validation

• python3 -m research.pipeline validate
• python3 -m pytest -q tests/test_research_campaign_best_effort.py tests/test_research_pipeline.py tests/test_pipeline_research_campaign.py
• python3 -m research.pipeline verify --offline

Follow-up

Reopen GitHub issues #1331–#1373 with agent:ready after this change is on main. Do not rewrite checked-in source-spec disposition fields.