[
 {
  "id": "C01",
  "http_status": 200,
  "recorded_at": "2026-11-03T23:12:50+00:00",
  "ms": 0,
  "reply": {
   "slug": "hallucination-firewall",
   "title": "Self-consistency check: do five samples agree?",
   "status": "ok",
   "input": null,
   "total_ms": 295,
   "steps": [
    {
     "call": "POST /threshold-check",
     "request": {
      "prompt_id": "demo",
      "max_disagreement": 0.3,
      "samples": [
       "The Eiffel Tower was completed in 1889.",
       "The Eiffel Tower was completed in 1889.",
       "It opened in 1889 for the World's Fair.",
       "The Eiffel Tower was finished in 1901.",
       "Construction ended in 1875."
      ]
     },
     "status": 200,
     "ms": 278,
     "reply": {
      "score_id": "821cd4b4-8106-4d8b-a5b1-13521f613338",
      "prompt_id": "demo",
      "sample_count": 5,
      "disagreement_score": 0.961,
      "mean_pairwise_similarity": 0.417,
      "risk_level": "high",
      "clusters": [
       {
        "members": [
         0,
         1
        ],
        "size": 2,
        "example": "The Eiffel Tower was completed in 1889."
       },
       {
        "members": [
         2
        ],
        "size": 1,
        "example": "It opened in 1889 for the World's Fair."
       },
       {
        "members": [
         3
        ],
        "size": 1,
        "example": "The Eiffel Tower was finished in 1901."
       },
       {
        "members": [
         4
        ],
        "size": 1,
        "example": "Construction ended in 1875."
       }
      ],
      "note": "This measures self-consistency across your own resamples, not ground truth. A LOW risk score means the model answered consistently -- it does not mean the answer is correct.",
      "threshold": 0.3,
      "passed": false
     }
    }
   ],
   "note": "Real calls to the running service with the public demo identity; replies are unedited."
  }
 },
 {
  "id": "C02",
  "http_status": 200,
  "recorded_at": "2026-11-03T23:12:51+00:00",
  "ms": 0,
  "reply": {
   "slug": "hallucination-firewall",
   "title": "Self-consistency check: do five samples agree?",
   "status": "ok",
   "input": null,
   "total_ms": 295,
   "steps": [
    {
     "call": "POST /threshold-check",
     "request": {
      "prompt_id": "demo",
      "max_disagreement": 0.3,
      "samples": [
       "The Eiffel Tower was completed in 1889.",
       "The Eiffel Tower was completed in 1889.",
       "It opened in 1889 for the World's Fair.",
       "The Eiffel Tower was finished in 1901.",
       "Construction ended in 1875."
      ]
     },
     "status": 200,
     "ms": 278,
     "reply": {
      "score_id": "821cd4b4-8106-4d8b-a5b1-13521f613338",
      "prompt_id": "demo",
      "sample_count": 5,
      "disagreement_score": 0.961,
      "mean_pairwise_similarity": 0.417,
      "risk_level": "high",
      "clusters": [
       {
        "members": [
         0,
         1
        ],
        "size": 2,
        "example": "The Eiffel Tower was completed in 1889."
       },
       {
        "members": [
         2
        ],
        "size": 1,
        "example": "It opened in 1889 for the World's Fair."
       },
       {
        "members": [
         3
        ],
        "size": 1,
        "example": "The Eiffel Tower was finished in 1901."
       },
       {
        "members": [
         4
        ],
        "size": 1,
        "example": "Construction ended in 1875."
       }
      ],
      "note": "This measures self-consistency across your own resamples, not ground truth. A LOW risk score means the model answered consistently -- it does not mean the answer is correct.",
      "threshold": 0.3,
      "passed": false
     }
    }
   ],
   "note": "Real calls to the running service with the public demo identity; replies are unedited."
  }
 }
]