05
Model Router
jev primitiveScore + Noul (route) · Noul (grade)
labelsfastbalancedfrontier
dataset20 labeled rows
tracingone call, not traced
Picks which model answers a task: fast, balanced, or frontier. Jev judges difficulty (Score) and risk (Noul) in one call, code maps them to a tier (every doubt routes up), and only that model answers. Every answer, from every variant, is graded by one Jev Noul against a rubric the answering models never see. Measured: pass rate and cost per passing answer (routing + answering; grading is kept apart).
With Jev
◐ ▮▮▮ ▦Without Jev
LLM · CODEExamples
Calls Jev and every baseline for real. Nothing is saved.
Runs every labeled row, one at a time: 20 rows × every variant, all real, paid calls. Rows run sequentially so latency measures the model, not a traffic jam. The run is saved as it goes and lands in History.