{
  "finding": "An 80 ms decision delay without pacing increased median post-trigger writes from 2 to 12. Cooperative pacing during the same delay reduced the median to 4, a 66.7% reduction versus the unpaced delayed policy.",
  "runs_per_policy": 10,
  "schema": "hella.exp007.public-evidence.v1",
  "source": "10-run live iSCSI pacing comparison",
  "summary": {
    "immediate": {
      "all_recovery_probes": true,
      "freeze_complete_median_ms": 49.074,
      "post_trigger_bytes_median": 8192.0,
      "post_trigger_writes_max": 3,
      "post_trigger_writes_median": 2.0
    },
    "paced-delay": {
      "all_recovery_probes": true,
      "freeze_complete_median_ms": 132.353,
      "post_trigger_bytes_median": 16384.0,
      "post_trigger_writes_max": 4,
      "post_trigger_writes_median": 4.0
    },
    "plain-delay": {
      "all_recovery_probes": true,
      "freeze_complete_median_ms": 125.163,
      "post_trigger_bytes_median": 49152.0,
      "post_trigger_writes_max": 14,
      "post_trigger_writes_median": 12.0
    }
  },
  "truth_boundary": "Real iSCSI-backed test path with cooperative application-level pacing; not storage-target enforcement."
}