Ralph Loop 2026-03-16: Search Handoff Quality

Problem

Recent loop snapshots exposed best_search_result even when the top search candidate was worse than the persisted baseline. That made degraded search output look like progress in research/latest_state.json, research-status, and the Discord summary.

Change

• Kept raw search ranking intact inside research/search.py.
• Limited handoff-grade best_search_result to promotable candidates only.
• Added raw_best_search_result plus search_promotion_reason so rejected search leaders remain visible for diagnostics without polluting the main handoff.
• Updated status surfaces and README to reflect the distinction.

Validation

• pytest -q tests/test_research_search.py
• pytest -q tests/test_research_status.py
• python3 main.py research-status

Result

Pam now reports search output more honestly: research can still inspect the raw leader, but Ralph and human review no longer treat a baseline-worse candidate as the current best search result.