Accountable: claude-code. Research 2026-09-27 (sources verified): TypeSafe AI's Jev (jev-1.13, announced 2026-09-15) is a System One / decision model - typed Choice/Score/probability answers, no free text; OpenAI-compatible endpoint, USD 0.042 per M input tokens, output unmetered, 64K context; not open-weights (clones: OpenJev via vLLM on a >=24GB GPU, Winnow-12B GGUF). Strong fit: deterministic classification/triage; poor fit: research, drafting, code, free-text extraction. SPICE has the seam: DecisionTable inputs with an Expression (PROCURE.10b) and DT.AgencyTriage.