{
  "$comment": "Built on SIP — Starlight Model Arena run receipt. Schema: arena-run v0.1.",
  "date": "2026-06-09",
  "harness": "claude-code-agent-tool",
  "contestants": [
    "claude-fable-5",
    "claude-opus-4-8"
  ],
  "judge": {
    "model": "claude-sonnet-4-6",
    "blind": true,
    "labelAssignment": {
      "task1": {
        "A": "opus",
        "B": "fable"
      },
      "task2": {
        "A": "fable",
        "B": "opus"
      }
    }
  },
  "tasks": [
    {
      "id": "reasoning-logic-grid",
      "category": "reasoning + instruction compliance",
      "verification": "ground-truth answer + blind judge on output discipline",
      "groundTruth": "Ana=Arena=Mon, Ben=Cockpit=Tue, Cara=Bridge=Wed",
      "results": {
        "fable": {
          "correct": true,
          "judgeScore": 9,
          "durationMs": 7264,
          "tokens": 49996,
          "notes": "Clean table + tight justification, exactly as instructed."
        },
        "opus": {
          "correct": true,
          "judgeScore": 6,
          "durationMs": 7063,
          "tokens": 50033,
          "notes": "Correct, but leaked stream-of-consciousness ('wait —') into final output, violating 'output ONLY' instruction."
        }
      },
      "winner": "fable",
      "prompt": "reasoning + instruction compliance"
    },
    {
      "id": "coding-next-same-popcount",
      "category": "coding (self-verifying asserts, max 3 attempts)",
      "verification": "agent ran asserts via python; ALL PASS required",
      "results": {
        "fable": {
          "status": "PASS",
          "attempts": 1,
          "durationMs": 25776,
          "tokens": 53081,
          "notes": "Gosper's hack, first try."
        },
        "opus": {
          "status": "PASS",
          "attempts": 1,
          "durationMs": 30375,
          "tokens": 53342,
          "notes": "Gosper's hack, first try."
        }
      },
      "winner": "tie",
      "prompt": "coding (self-verifying asserts, max 3 attempts)"
    },
    {
      "id": "repo-grounding-claude-md",
      "category": "repo-grounded factual accuracy",
      "verification": "three ground-truth answers from CLAUDE.md (governance command, board-before-tag timing/version, recovery exception)",
      "results": {
        "fable": {
          "correct": "3/3",
          "durationMs": 10077,
          "tokens": 57051,
          "notes": "All correct, richer supporting detail."
        },
        "opus": {
          "correct": "3/3",
          "durationMs": 10379,
          "tokens": 60825,
          "notes": "All correct, slightly terser."
        }
      },
      "winner": "tie",
      "prompt": "repo-grounded factual accuracy"
    },
    {
      "id": "voice-arena-intro",
      "category": "brand-voice writing (100-130 words, voice spec)",
      "verification": "blind judge on voice fit + harness word-count check",
      "results": {
        "fable": {
          "judgeScore": 8,
          "wordCount": 128,
          "constraintCompliant": true,
          "durationMs": 16560,
          "tokens": 49819
        },
        "opus": {
          "judgeScore": 9,
          "wordCount": 148,
          "constraintCompliant": false,
          "durationMs": 9638,
          "tokens": 49834,
          "notes": "Judge preferred prose style; exceeded 130-word hard limit by 18 words."
        }
      },
      "winner": "split — judge: opus on style; harness: fable on constraint compliance",
      "prompt": "brand-voice writing (100-130 words, voice spec)"
    }
  ],
  "summary": {
    "tally": {
      "fable": 1,
      "opus": 0,
      "tie": 2,
      "split": 1
    },
    "headline": "Correctness parity across reasoning, coding, and grounding. Fable 5's measurable edge is instruction compliance / output discipline: it was the only model that respected output-format and length constraints in both judged tasks.",
    "caveats": [
      "n=1 per task — directional, not statistical.",
      "Judge is a Claude-family model (sonnet); blind shuffled labels mitigate but do not eliminate family bias.",
      "Both contestants run inside the Claude Code agent harness; results measure model-in-harness, not raw API behavior.",
      "Latency includes harness overhead; treat durations as rough."
    ]
  },
  "attestation": "Built on SIP — Starlight Intelligence Protocol",
  "round_id": "arena-2026-06-09-fable5-vs-opus48",
  "title": "Round 1: Fable 5 vs Opus 4.8 (2026-06-09)",
  "methodology": "Same prompt dispatched to two subagents with per-agent model override; objective tasks self-verify (asserts / ground truth); subjective tasks scored by a blind non-contestant judge (sonnet) with shuffled A/B labels."
}