RES — llama3.1:8b · estimation (Phase-B cell)
Phase-B sweep cell — llama3.1:8b on the estimation scenario (N=100, seed=42, temp=0.0, single strategy).
MAE (SP) = 2.6 (parse_failure_rate = 0.0).
Source: PUMA project documentation + raw benchmark results · Traceability: PUMA-023,PUMA-026 · Confidence: verified-at-primary-source