{
  "metric": "verdict_matched_next_probe",
  "definition": "For each scheduled probe of an endpoint, did the very next scheduled probe of the same endpoint (usually ~15 min later, at most 60 min) have the same status? safe = handshake + tools/list OK, caution_auth = requires authentication, unsafe = down/broken.",
  "limitations": [
    "Measures how stable a verdict is until the next check, not whether real tool calls succeed.",
    "Scored by status class only; extra caution reasons (slow, recent breaking change, stale) are not scored.",
    "History starts 2026-10-15 (UTC); early days have small samples."
  ],
  "days_covered": 1,
  "totals": {
    "caution_auth": {
      "pairs": 3048,
      "matched": 3047,
      "match_rate": 0.9997
    },
    "safe": {
      "pairs": 2606,
      "matched": 2603,
      "match_rate": 0.9988
    },
    "unsafe": {
      "pairs": 149,
      "matched": 121,
      "match_rate": 0.8121
    }
  },
  "daily": [
    {
      "day": "2026-10-15",
      "caution_auth": {
        "pairs": 3048,
        "matched": 3047,
        "match_rate": 0.9997,
        "avg_gap_minutes": 13
      },
      "safe": {
        "pairs": 2606,
        "matched": 2603,
        "match_rate": 0.9988,
        "avg_gap_minutes": 13
      },
      "unsafe": {
        "pairs": 149,
        "matched": 121,
        "match_rate": 0.8121,
        "avg_gap_minutes": 13
      }
    }
  ]
}