{
  "lesson": "16-model-routing",
  "title": "Model Routing as a Cost-Reduction Primitive",
  "questions": [
    {
      "stage": "pre",
      "question": "What is the core idea of model cascading?",
      "options": [
        "Always route by random weight",
        "Run two models in parallel and average the outputs",
        "Run every request on the most expensive model first",
        "Run a cheap model first, escalate to a frontier model only on low confidence or refusal"
      ],
      "correct": 3,
      "explanation": ""
    },
    {
      "stage": "check",
      "question": "Which four signals does the lesson list for routing decisions?",
      "options": [
        "User tier only",
        "Token count only",
        "Task classification, prompt length, embedding similarity to known-hard set, and self-confidence from a first-pass",
        "GPU temperature, fan speed, room humidity, time of day"
      ],
      "correct": 2,
      "explanation": ""
    },
    {
      "stage": "check",
      "question": "What is the expected latency profile of a cascade router?",
      "options": [
        "Always faster than pre-route",
        "About 1.2x median latency (cheap run plus verify), about 2x on escalated requests (~10% of traffic)",
        "Always slower by 10x",
        "Identical to the frontier model"
      ],
      "correct": 1,
      "explanation": ""
    },
    {
      "stage": "check",
      "question": "Which routing pattern adds 5-10ms latency up front and is fastest overall?",
      "options": [
        "Cascade",
        "Ensemble route",
        "Random round-robin",
        "Pre-route with a classifier"
      ],
      "correct": 3,
      "explanation": ""
    },
    {
      "stage": "post",
      "question": "What is cheap-model drift in routing?",
      "options": [
        "The cheap model becomes more expensive",
        "Cascade falls through to frontier 100% of the time",
        "A latency drift in the cheap model",
        "Task distribution shifts but the trained router keeps sending requests to the cheap model, silently degrading quality"
      ],
      "correct": 3,
      "explanation": ""
    },
    {
      "stage": "post",
      "question": "Which guard does the lesson recommend to catch routing drift?",
      "options": [
        "Online quality metrics — thumbs-up/down per route, LLM-judge on held-out samples, escalation rate, refusal rate",
        "Quarterly engineering review only",
        "Only offline eval sets",
        "Disable routing in production"
      ],
      "correct": 0,
      "explanation": ""
    }
  ]
}