PH — Four scoped project options were ranked by difficulty; the recommended…
Four scoped project options were ranked by difficulty; the recommended Option 1 was the simplest and most defensible — a single Python pipeline that classifies/prioritizes risks on a public CSV and compares LLM output against ground-truth labels with Precision/Recall/F1 — the seed of the benchmark pivot.
Source: phases/01-investigation/Propuesta · Traceability: corpus unit
PHASE-005· Confidence: verified-at-primary-source