{
  "format": 1,
  "kind": "live-ab-result",
  "generatedAt": "2026-10-29T16:20:00+08:00",
  "provider": "deepseek-account",
  "model": "deepseek-flash",
  "interface": "DSH DeepSeek Account model route; two independent fresh-context agent calls (control, treatment)",
  "fixture": {
    "source": "scripts/live-ab.mjs --emit-prompts --probes 8 --model deepseek-flash",
    "historyTurns": 120,
    "tailRecords": 20,
    "probes": 8,
    "controlPromptChars": 1686,
    "treatmentPromptChars": 15207,
    "note": "Synthetic project history; answers graded by exact FACT id."
  },
  "rows": [
    {
      "index": 0,
      "turn": 0,
      "topic": "gateway",
      "expected": "FACT-000-qto8z",
      "control": {
        "answer": "UNKNOWN",
        "correct": false
      },
      "treatment": {
        "answer": "FACT-000-qto8z",
        "correct": true
      }
    },
    {
      "index": 1,
      "turn": 15,
      "topic": "router",
      "expected": "FACT-015-xc4r1",
      "control": {
        "answer": "UNKNOWN",
        "correct": false
      },
      "treatment": {
        "answer": "FACT-015-xc4r1",
        "correct": true
      }
    },
    {
      "index": 2,
      "turn": 30,
      "topic": "gateway",
      "expected": "FACT-030-msivd",
      "control": {
        "answer": "UNKNOWN",
        "correct": false
      },
      "treatment": {
        "answer": "FACT-030-msivd",
        "correct": true
      }
    },
    {
      "index": 3,
      "turn": 45,
      "topic": "router",
      "expected": "FACT-045-cyvqv",
      "control": {
        "answer": "UNKNOWN",
        "correct": false
      },
      "treatment": {
        "answer": "FACT-045-cyvqv",
        "correct": true
      }
    },
    {
      "index": 4,
      "turn": 60,
      "topic": "gateway",
      "expected": "FACT-060-tqhez",
      "control": {
        "answer": "UNKNOWN",
        "correct": false
      },
      "treatment": {
        "answer": "FACT-060-tqhez",
        "correct": true
      }
    },
    {
      "index": 5,
      "turn": 75,
      "topic": "router",
      "expected": "FACT-075-u3vae",
      "control": {
        "answer": "UNKNOWN",
        "correct": false
      },
      "treatment": {
        "answer": "FACT-075-u3vae",
        "correct": true
      }
    },
    {
      "index": 6,
      "turn": 90,
      "topic": "gateway",
      "expected": "FACT-090-eie52",
      "control": {
        "answer": "FACT-090-eie52",
        "correct": true
      },
      "treatment": {
        "answer": "FACT-090-eie52",
        "correct": true
      }
    },
    {
      "index": 7,
      "turn": 105,
      "topic": "router",
      "expected": "FACT-105-o971e",
      "control": {
        "answer": "UNKNOWN",
        "correct": false
      },
      "treatment": {
        "answer": "FACT-105-o971e",
        "correct": true
      }
    }
  ],
  "accuracy": {
    "overall": {
      "control": 0.125,
      "treatment": 1.0
    },
    "evictedFacts": {
      "probes": 7,
      "control": 0.0,
      "treatment": 1.0,
      "excluded": "Q6 (turn 90) was still inside the retained tail, so the control condition could answer it; Q0-Q5 and Q7 are evicted-fact probes."
    }
  },
  "notes": [
    "Two runs per condition were not repeated; n=8 probes (7 evicted) is a first data point, not a statistically strong result.",
    "Usage/token counts were not captured by this agent-mediated interface; latency includes agent harness overhead and is not reported as provider latency.",
    "The model saw only the synthetic prompt; no private data was involved."
  ]
}