{
  "generated": "2026-07-23 16:49 UTC",
  "source": "https://ipezygj.github.io/eval-audit-site/status.html",
  "method": "bootstrap rank CIs; paired McNemar; split-half stability",
  "summary": {
    "swebench_boards": 6,
    "swebench_coinflips": 5,
    "real_top_ranks": 3
  },
  "leaderboards": [
    {
      "board": "SWE-bench Test",
      "verdict": "ROBUST-#1",
      "resolved": true,
      "p_top_is_1": 1.0,
      "n_tied": 1,
      "leader": "sonar-foundation-agent_claude-opus-4-5",
      "badge": "https://ipezygj.github.io/eval-audit-site/badges/swe-bench-test.svg"
    },
    {
      "board": "SWE-bench Verified",
      "verdict": "COIN-FLIP-#1",
      "resolved": false,
      "p_top_is_1": 0.361,
      "n_tied": 3,
      "leader": "sonar-foundation-agent_claude-opus-4-5",
      "badge": "https://ipezygj.github.io/eval-audit-site/badges/swe-bench-verified.svg"
    },
    {
      "board": "SWE-bench Lite",
      "verdict": "COIN-FLIP-#1",
      "resolved": false,
      "p_top_is_1": 0.457,
      "n_tied": 4,
      "leader": "ExpeRepair-v1_claude-4-sonnet-20250514",
      "badge": "https://ipezygj.github.io/eval-audit-site/badges/swe-bench-lite.svg"
    },
    {
      "board": "SWE-bench Multimodal",
      "verdict": "COIN-FLIP-#1",
      "resolved": false,
      "p_top_is_1": 0.376,
      "n_tied": 4,
      "leader": "GUIRepair_o3",
      "badge": "https://ipezygj.github.io/eval-audit-site/badges/swe-bench-multimodal.svg"
    },
    {
      "board": "SWE-bench Multilingual",
      "verdict": "COIN-FLIP-#1",
      "resolved": false,
      "p_top_is_1": 0.512,
      "n_tied": 4,
      "leader": "mini-v2.0.0a0_gemini-3-flash",
      "badge": "https://ipezygj.github.io/eval-audit-site/badges/swe-bench-multilingual.svg"
    },
    {
      "board": "SWE-bench Bash-only",
      "verdict": "COIN-FLIP-#1",
      "resolved": false,
      "p_top_is_1": 0.574,
      "n_tied": 4,
      "leader": "mini-v2.0.0_claude-4-5-opus-high",
      "badge": "https://ipezygj.github.io/eval-audit-site/badges/swe-bench-bash-only.svg"
    },
    {
      "board": "LMArena (Chatbot Arena)",
      "verdict": "ROBUST-#1",
      "resolved": true,
      "p_top_is_1": 0.993,
      "n_tied": 1,
      "leader": "gemini-2.5-pro",
      "badge": "https://ipezygj.github.io/eval-audit-site/badges/lmarena--chatbot-arena.svg"
    },
    {
      "board": "LMArena (coding)",
      "verdict": "ROBUST-#1",
      "resolved": true,
      "p_top_is_1": 0.973,
      "n_tied": 1,
      "leader": "gemini-2.5-pro",
      "badge": "https://ipezygj.github.io/eval-audit-site/badges/lmarena--coding.svg"
    }
  ]
}