outputs/observability_reported.json

{
  "status": "REPORTED-NOT-RUN",
  "reason": "vLLM 0.25.1 engine-core init failed on NVIDIA driver 570.124.06; Qwen3-8B L2/L3 CoT rollouts not produced.",
  "table4_reported": {
    "L1 Direct": {
      "mean": 1.95,
      "p50": 0.63,
      "p90": 2.57,
      "p95": 3.27
    },
    "L1+Tool": {
      "mean": 2.03,
      "p50": 0.52,
      "p90": 2.47,
      "p95": 4.27
    },
    "L2 CoT": {
      "mean": 0.65,
      "p50": 0.25,
      "p90": 1.7,
      "p95": 2.01
    },
    "L3 CoT": {
      "mean": 1.66,
      "p50": 0.35,
      "p90": 1.95,
      "p95": 2.92
    }
  },
  "paradox_L2_lt_L3": true,
  "L2_mean": 0.65,
  "L3_mean": 1.66,
  "L3_minus_L2_pp": 1.0099999999999998
}