Ralph Loop 2026-03-16: Candidate Review Blocker Split

Change

Added explicit local-vs-global blocker attribution to the PRD 003 candidate-review packet.

Files: - research/candidate_review.py - research/continuous_loop.py - research/status.py - research/discord_status.py - research/README.md - tests/test_candidate_review.py - tests/test_continuous_loop.py - tests/test_research_status.py

Why

The packet already exposed top_invalidation_reasons, but that list can mix:

1. row-level sleeve or market blockers such as live_edge_bps<100.0
2. loop-level deployment blockers such as global:score<5.0

When the loop is globally blocked, Ralph needs to know whether candidate quality is still improving underneath that gate. A single mixed blocker list makes that hard to see.

Result

Candidate review now carries:

1. top_invalidation_reasons
2. top_local_invalidation_reasons
3. top_global_invalidation_reasons

The continuous-loop event payload now also emits the top local/global blocker, and the human + Discord status summaries print those splits directly.

Validation

• pytest -q tests/test_candidate_review.py tests/test_continuous_loop.py tests/test_research_status.py
• Result: 15 passed
• python3 -m py_compile research/candidate_review.py research/continuous_loop.py research/status.py research/discord_status.py
• python3 research/candidate_review.py --db data/unicorn.db --recent-runs 6 --max-per-state 2
• Observed: top_local_invalidation_reasons is dominated by heuristic_direction=neutral, live_price<0.7, and category!=Economics, while top_global_invalidation_reasons isolates the loop gate as global:deployment_gate_not_met
• python3 research/status.py
• Observed: status now prints both local_blockers=... and global_blockers=...

Outcome

This is a bounded PRD 003/operator-quality improvement. It does not create trades or loosen the deployment gate; it makes candidate-review quality legible when the loop remains globally blocked.