Ralph Loop 2026-03-16: Unmatched Semantic Category Diagnostics

Change

Added top_unmatched_semantic_categories to the live candidate-review packet and surfaced the dominant unmatched category through loop/status summaries.

Why

Candidate review was already telling us that 15 live rows were semantically unmatched, but the blocker labels were too generic to guide the next sleeve decision. The refreshed status now shows where the uncovered inventory is clustering by category.

Evidence

• Focused validation passed:
• pytest tests/test_candidate_review.py tests/test_continuous_loop.py
• Refreshed operator summary:
• python3 main.py research-status
• Current unmatched concentration is Science and Technology=4, World=3, Companies=3

Result

This improves candidate review quality without changing any sleeve, search parameter, or deployment gate. The next sleeve iteration can target the highest unmatched categories directly instead of inferring them from semantic blocker noise.