RES — The canonical triage baseline is F1-macro = 0.5867 (qwen2.5:3b,…
The canonical triage baseline is F1-macro = 0.5867 (qwen2.5:3b, contextual-anchoring, N=200, seed=42, T=0.0, +/-0.01 tolerance) on reference gpu-entry hardware; pure zero-shot on the same config yields only F1=0.3898.
Source: puma docs/baseline_references.md · Traceability: corpus unit
PUMA-017· Confidence: repo-backed