{
  "complete": true,
  "finished_attempts": 42,
  "reviewed_attempts": 42,
  "planned_attempts": 42,
  "baseline_attempt_seconds_sum": 6695.02,
  "xml_attempt_seconds_sum": 11077.92,
  "models": [
    {
      "label": "deepseek",
      "model": "opencode-go/deepseek-v4.1-flash",
      "client": "opencode",
      "baseline_attempts": 3,
      "xml_attempts": 3,
      "baseline_seconds": 53.25,
      "xml_seconds": 69.02,
      "xml_parse_count": 3,
      "xml_xsd_count": 3,
      "xml_render_count": 3,
      "xml_tool_violations": 0,
      "xml_semantic_assessed": 3,
      "xml_token_usage_n": 3,
      "xml_clean_render_count": 3,
      "xml_partial_usage_n": 0,
      "xml_unmetered_step_runs": 0,
      "xml_timeouts": 0,
      "baseline_input_total": 166157,
      "xml_input_total": 2376,
      "baseline_input_nonread": 24336,
      "xml_input_nonread": 2376,
      "baseline_cache_read": 146304,
      "xml_cache_read": 0,
      "baseline_cache_write": 0,
      "xml_cache_write": 0,
      "baseline_output_including_reasoning": 7489,
      "xml_output_including_reasoning": 20855,
      "baseline_pepper_independent_pass": 0,
      "xml_pepper_independent_pass": 0,
      "baseline_pepper_independent_statuses": {
        "fail": 3
      },
      "xml_pepper_independent_statuses": {
        "fail": 3
      },
      "baseline_hare_wait_pass": 2,
      "xml_hare_wait_pass": 2,
      "baseline_hare_wait_statuses": {
        "pass": 2,
        "fail": 1
      },
      "xml_hare_wait_statuses": {
        "pass": 2,
        "fail": 1
      },
      "baseline_termination_pass": 3,
      "xml_termination_pass": 3,
      "baseline_termination_statuses": {
        "pass": 3
      },
      "xml_termination_statuses": {
        "pass": 3
      },
      "baseline_all_scenarios_pass": 0,
      "xml_all_scenarios_pass": 0,
      "xml_valid_rendered_all_scenarios_pass": 0
    },
    {
      "label": "claude",
      "model": "claude-opus-5",
      "client": "claude",
      "baseline_attempts": 3,
      "xml_attempts": 3,
      "baseline_seconds": 105.58,
      "xml_seconds": 233.67,
      "xml_parse_count": 3,
      "xml_xsd_count": 3,
      "xml_render_count": 3,
      "xml_tool_violations": 0,
      "xml_semantic_assessed": 3,
      "xml_token_usage_n": 3,
      "xml_clean_render_count": 3,
      "xml_partial_usage_n": 0,
      "xml_unmetered_step_runs": 0,
      "xml_timeouts": 0,
      "baseline_input_total": 337062,
      "xml_input_total": 4222,
      "baseline_input_nonread": 48022,
      "xml_input_nonread": 1063,
      "baseline_cache_read": 289040,
      "xml_cache_read": 3159,
      "baseline_cache_write": 48004,
      "xml_cache_write": 1061,
      "baseline_output_including_reasoning": 8874,
      "xml_output_including_reasoning": 25997,
      "baseline_pepper_independent_pass": 0,
      "xml_pepper_independent_pass": 3,
      "baseline_pepper_independent_statuses": {
        "fail": 3
      },
      "xml_pepper_independent_statuses": {
        "pass": 3
      },
      "baseline_hare_wait_pass": 1,
      "xml_hare_wait_pass": 0,
      "baseline_hare_wait_statuses": {
        "pass": 1,
        "abstract": 1,
        "uncertain": 1
      },
      "xml_hare_wait_statuses": {
        "fail": 3
      },
      "baseline_termination_pass": 3,
      "xml_termination_pass": 3,
      "baseline_termination_statuses": {
        "pass": 3
      },
      "xml_termination_statuses": {
        "pass": 3
      },
      "baseline_all_scenarios_pass": 0,
      "xml_all_scenarios_pass": 0,
      "xml_valid_rendered_all_scenarios_pass": 0
    },
    {
      "label": "glm",
      "model": "opencode-go/glm-5.3",
      "client": "opencode",
      "baseline_attempts": 3,
      "xml_attempts": 3,
      "baseline_seconds": 278.14,
      "xml_seconds": 326.37,
      "xml_parse_count": 0,
      "xml_xsd_count": 0,
      "xml_render_count": 0,
      "xml_tool_violations": 0,
      "xml_semantic_assessed": 3,
      "xml_token_usage_n": 3,
      "xml_clean_render_count": 0,
      "xml_partial_usage_n": 0,
      "xml_unmetered_step_runs": 2,
      "xml_timeouts": 0,
      "baseline_input_total": 375229,
      "xml_input_total": 7256,
      "baseline_input_nonread": 59310,
      "xml_input_nonread": 2256,
      "baseline_cache_read": 321947,
      "xml_cache_read": 2267,
      "baseline_cache_write": 0,
      "xml_cache_write": 0,
      "baseline_output_including_reasoning": 15734,
      "xml_output_including_reasoning": 10938,
      "baseline_pepper_independent_pass": 0,
      "xml_pepper_independent_pass": 0,
      "baseline_pepper_independent_statuses": {
        "fail": 3
      },
      "xml_pepper_independent_statuses": {
        "not_assessable": 3
      },
      "baseline_hare_wait_pass": 0,
      "xml_hare_wait_pass": 0,
      "baseline_hare_wait_statuses": {
        "fail": 2,
        "abstract": 1
      },
      "xml_hare_wait_statuses": {
        "not_assessable": 3
      },
      "baseline_termination_pass": 3,
      "xml_termination_pass": 0,
      "baseline_termination_statuses": {
        "pass": 3
      },
      "xml_termination_statuses": {
        "not_assessable": 3
      },
      "baseline_all_scenarios_pass": 0,
      "xml_all_scenarios_pass": 0,
      "xml_valid_rendered_all_scenarios_pass": 0
    },
    {
      "label": "codex",
      "model": "gpt-6-astra",
      "client": "codex",
      "baseline_attempts": 3,
      "xml_attempts": 3,
      "baseline_seconds": 224.46,
      "xml_seconds": 369.38,
      "xml_parse_count": 3,
      "xml_xsd_count": 3,
      "xml_render_count": 3,
      "xml_tool_violations": 0,
      "xml_semantic_assessed": 3,
      "xml_token_usage_n": 3,
      "xml_clean_render_count": 3,
      "xml_partial_usage_n": 0,
      "xml_unmetered_step_runs": 0,
      "xml_timeouts": 0,
      "baseline_input_total": 709033,
      "xml_input_total": 12598,
      "baseline_input_nonread": 60713,
      "xml_input_nonread": 4918,
      "baseline_cache_read": 648320,
      "xml_cache_read": 7680,
      "baseline_cache_write": 0,
      "xml_cache_write": 0,
      "baseline_output_including_reasoning": 5034,
      "xml_output_including_reasoning": 12068,
      "baseline_pepper_independent_pass": 3,
      "xml_pepper_independent_pass": 3,
      "baseline_pepper_independent_statuses": {
        "pass": 3
      },
      "xml_pepper_independent_statuses": {
        "pass": 3
      },
      "baseline_hare_wait_pass": 3,
      "xml_hare_wait_pass": 3,
      "baseline_hare_wait_statuses": {
        "pass": 3
      },
      "xml_hare_wait_statuses": {
        "pass": 3
      },
      "baseline_termination_pass": 3,
      "xml_termination_pass": 3,
      "baseline_termination_statuses": {
        "pass": 3
      },
      "xml_termination_statuses": {
        "pass": 3
      },
      "baseline_all_scenarios_pass": 3,
      "xml_all_scenarios_pass": 3,
      "xml_valid_rendered_all_scenarios_pass": 3
    },
    {
      "label": "kimi",
      "model": "opencode-go/kimi-k3",
      "client": "opencode",
      "baseline_attempts": 3,
      "xml_attempts": 3,
      "baseline_seconds": 158.51,
      "xml_seconds": 210.9,
      "xml_parse_count": 3,
      "xml_xsd_count": 3,
      "xml_render_count": 3,
      "xml_tool_violations": 0,
      "xml_semantic_assessed": 3,
      "xml_token_usage_n": 3,
      "xml_clean_render_count": 3,
      "xml_partial_usage_n": 0,
      "xml_unmetered_step_runs": 1,
      "xml_timeouts": 0,
      "baseline_input_total": 337182,
      "xml_input_total": 2462,
      "baseline_input_nonread": 67602,
      "xml_input_nonread": 2441,
      "baseline_cache_read": 269580,
      "xml_cache_read": 1024,
      "baseline_cache_write": 0,
      "xml_cache_write": 0,
      "baseline_output_including_reasoning": 7860,
      "xml_output_including_reasoning": 20143,
      "baseline_pepper_independent_pass": 0,
      "xml_pepper_independent_pass": 0,
      "baseline_pepper_independent_statuses": {
        "fail": 3
      },
      "xml_pepper_independent_statuses": {
        "fail": 3
      },
      "baseline_hare_wait_pass": 0,
      "xml_hare_wait_pass": 3,
      "baseline_hare_wait_statuses": {
        "abstract": 1,
        "uncertain": 2
      },
      "xml_hare_wait_statuses": {
        "pass": 3
      },
      "baseline_termination_pass": 3,
      "xml_termination_pass": 3,
      "baseline_termination_statuses": {
        "pass": 3
      },
      "xml_termination_statuses": {
        "pass": 3
      },
      "baseline_all_scenarios_pass": 0,
      "xml_all_scenarios_pass": 0,
      "xml_valid_rendered_all_scenarios_pass": 0
    },
    {
      "label": "luna",
      "model": "gpt-5.6-luna",
      "client": "codex",
      "baseline_attempts": 3,
      "xml_attempts": 3,
      "baseline_seconds": 79.79,
      "xml_seconds": 171.58,
      "xml_parse_count": 2,
      "xml_xsd_count": 1,
      "xml_render_count": 3,
      "xml_tool_violations": 0,
      "xml_semantic_assessed": 3,
      "xml_token_usage_n": 3,
      "xml_clean_render_count": 1,
      "xml_partial_usage_n": 0,
      "xml_unmetered_step_runs": 0,
      "xml_timeouts": 0,
      "baseline_input_total": 335512,
      "xml_input_total": 10075,
      "baseline_input_nonread": 47256,
      "xml_input_nonread": 10075,
      "baseline_cache_read": 288256,
      "xml_cache_read": 0,
      "baseline_cache_write": 0,
      "xml_cache_write": 0,
      "baseline_output_including_reasoning": 3253,
      "xml_output_including_reasoning": 9315,
      "baseline_pepper_independent_pass": 0,
      "xml_pepper_independent_pass": 1,
      "baseline_pepper_independent_statuses": {
        "fail": 3
      },
      "xml_pepper_independent_statuses": {
        "not_assessable": 1,
        "fail": 1,
        "pass": 1
      },
      "baseline_hare_wait_pass": 0,
      "xml_hare_wait_pass": 1,
      "baseline_hare_wait_statuses": {
        "uncertain": 1,
        "fail": 2
      },
      "xml_hare_wait_statuses": {
        "not_assessable": 1,
        "pass": 1,
        "fail": 1
      },
      "baseline_termination_pass": 3,
      "xml_termination_pass": 1,
      "baseline_termination_statuses": {
        "pass": 3
      },
      "xml_termination_statuses": {
        "not_assessable": 1,
        "pass": 1,
        "fail": 1
      },
      "baseline_all_scenarios_pass": 0,
      "xml_all_scenarios_pass": 0,
      "xml_valid_rendered_all_scenarios_pass": 0
    },
    {
      "label": "qwen-flash",
      "model": "opencode-go/qwen3.8-flash",
      "client": "opencode",
      "baseline_attempts": 3,
      "xml_attempts": 3,
      "baseline_seconds": 289.02,
      "xml_seconds": 570.19,
      "xml_parse_count": 2,
      "xml_xsd_count": 0,
      "xml_render_count": 2,
      "xml_tool_violations": 0,
      "xml_semantic_assessed": 3,
      "xml_token_usage_n": 3,
      "xml_clean_render_count": 0,
      "xml_partial_usage_n": 0,
      "xml_unmetered_step_runs": 2,
      "xml_timeouts": 0,
      "baseline_input_total": 302319,
      "xml_input_total": 2357,
      "baseline_input_nonread": 51557,
      "xml_input_nonread": 15,
      "baseline_cache_read": 249671,
      "xml_cache_read": 2342,
      "baseline_cache_write": 51497,
      "xml_cache_write": 0,
      "baseline_output_including_reasoning": 12225,
      "xml_output_including_reasoning": 29435,
      "baseline_pepper_independent_pass": 0,
      "xml_pepper_independent_pass": 1,
      "baseline_pepper_independent_statuses": {
        "fail": 3
      },
      "xml_pepper_independent_statuses": {
        "fail": 1,
        "not_assessable": 1,
        "pass": 1
      },
      "baseline_hare_wait_pass": 0,
      "xml_hare_wait_pass": 0,
      "baseline_hare_wait_statuses": {
        "abstract": 2,
        "fail": 1
      },
      "xml_hare_wait_statuses": {
        "fail": 2,
        "not_assessable": 1
      },
      "baseline_termination_pass": 3,
      "xml_termination_pass": 2,
      "baseline_termination_statuses": {
        "pass": 3
      },
      "xml_termination_statuses": {
        "pass": 2,
        "not_assessable": 1
      },
      "baseline_all_scenarios_pass": 0,
      "xml_all_scenarios_pass": 0,
      "xml_valid_rendered_all_scenarios_pass": 0
    },
    {
      "label": "sol",
      "model": "gpt-5.6-sol",
      "client": "codex",
      "baseline_attempts": 3,
      "xml_attempts": 3,
      "baseline_seconds": 245.29,
      "xml_seconds": 244.89,
      "xml_parse_count": 3,
      "xml_xsd_count": 3,
      "xml_render_count": 3,
      "xml_tool_violations": 0,
      "xml_semantic_assessed": 3,
      "xml_token_usage_n": 3,
      "xml_clean_render_count": 3,
      "xml_partial_usage_n": 0,
      "xml_unmetered_step_runs": 0,
      "xml_timeouts": 0,
      "baseline_input_total": 1538803,
      "xml_input_total": 11588,
      "baseline_input_nonread": 99454,
      "xml_input_nonread": 11582,
      "baseline_cache_read": 1428992,
      "xml_cache_read": 0,
      "baseline_cache_write": 0,
      "xml_cache_write": 0,
      "baseline_output_including_reasoning": 10127,
      "xml_output_including_reasoning": 11776,
      "baseline_pepper_independent_pass": 0,
      "xml_pepper_independent_pass": 1,
      "baseline_pepper_independent_statuses": {
        "fail": 3
      },
      "xml_pepper_independent_statuses": {
        "fail": 2,
        "pass": 1
      },
      "baseline_hare_wait_pass": 3,
      "xml_hare_wait_pass": 2,
      "baseline_hare_wait_statuses": {
        "pass": 3
      },
      "xml_hare_wait_statuses": {
        "pass": 2,
        "fail": 1
      },
      "baseline_termination_pass": 3,
      "xml_termination_pass": 3,
      "baseline_termination_statuses": {
        "pass": 3
      },
      "xml_termination_statuses": {
        "pass": 3
      },
      "baseline_all_scenarios_pass": 0,
      "xml_all_scenarios_pass": 0,
      "xml_valid_rendered_all_scenarios_pass": 0
    },
    {
      "label": "mimo",
      "model": "opencode-go/mimo-v2.5",
      "client": "opencode",
      "baseline_attempts": 3,
      "xml_attempts": 3,
      "baseline_seconds": 122.16,
      "xml_seconds": 219.4,
      "xml_parse_count": 2,
      "xml_xsd_count": 1,
      "xml_render_count": 2,
      "xml_tool_violations": 0,
      "xml_semantic_assessed": 3,
      "xml_token_usage_n": 3,
      "xml_clean_render_count": 2,
      "xml_partial_usage_n": 0,
      "xml_unmetered_step_runs": 1,
      "xml_timeouts": 0,
      "baseline_input_total": 117243,
      "xml_input_total": 2361,
      "baseline_input_nonread": 24729,
      "xml_input_nonread": 920,
      "baseline_cache_read": 102400,
      "xml_cache_read": 1728,
      "baseline_cache_write": 0,
      "xml_cache_write": 0,
      "baseline_output_including_reasoning": 2502,
      "xml_output_including_reasoning": 8669,
      "baseline_pepper_independent_pass": 1,
      "xml_pepper_independent_pass": 1,
      "baseline_pepper_independent_statuses": {
        "fail": 2,
        "pass": 1
      },
      "xml_pepper_independent_statuses": {
        "pass": 1,
        "fail": 1,
        "not_assessable": 1
      },
      "baseline_hare_wait_pass": 0,
      "xml_hare_wait_pass": 0,
      "baseline_hare_wait_statuses": {
        "fail": 2,
        "uncertain": 1
      },
      "xml_hare_wait_statuses": {
        "fail": 2,
        "not_assessable": 1
      },
      "baseline_termination_pass": 3,
      "xml_termination_pass": 1,
      "baseline_termination_statuses": {
        "pass": 3
      },
      "xml_termination_statuses": {
        "pass": 1,
        "fail": 1,
        "not_assessable": 1
      },
      "baseline_all_scenarios_pass": 0,
      "xml_all_scenarios_pass": 0,
      "xml_valid_rendered_all_scenarios_pass": 0
    },
    {
      "label": "terra",
      "model": "gpt-5.6-terra",
      "client": "codex",
      "baseline_attempts": 3,
      "xml_attempts": 3,
      "baseline_seconds": 207.49,
      "xml_seconds": 387.06,
      "xml_parse_count": 3,
      "xml_xsd_count": 2,
      "xml_render_count": 3,
      "xml_tool_violations": 0,
      "xml_semantic_assessed": 3,
      "xml_token_usage_n": 3,
      "xml_clean_render_count": 3,
      "xml_partial_usage_n": 0,
      "xml_unmetered_step_runs": 0,
      "xml_timeouts": 0,
      "baseline_input_total": 691226,
      "xml_input_total": 11590,
      "baseline_input_nonread": 64931,
      "xml_input_nonread": 10805,
      "baseline_cache_read": 609792,
      "xml_cache_read": 2816,
      "baseline_cache_write": 0,
      "xml_cache_write": 0,
      "baseline_output_including_reasoning": 4956,
      "xml_output_including_reasoning": 11471,
      "baseline_pepper_independent_pass": 0,
      "xml_pepper_independent_pass": 3,
      "baseline_pepper_independent_statuses": {
        "fail": 3
      },
      "xml_pepper_independent_statuses": {
        "pass": 3
      },
      "baseline_hare_wait_pass": 1,
      "xml_hare_wait_pass": 0,
      "baseline_hare_wait_statuses": {
        "fail": 2,
        "pass": 1
      },
      "xml_hare_wait_statuses": {
        "fail": 3
      },
      "baseline_termination_pass": 3,
      "xml_termination_pass": 3,
      "baseline_termination_statuses": {
        "pass": 3
      },
      "xml_termination_statuses": {
        "pass": 3
      },
      "baseline_all_scenarios_pass": 0,
      "xml_all_scenarios_pass": 0,
      "xml_valid_rendered_all_scenarios_pass": 0
    },
    {
      "label": "minimax",
      "model": "opencode-go/minimax-m3",
      "client": "opencode",
      "baseline_attempts": 3,
      "xml_attempts": 3,
      "baseline_seconds": 85.5,
      "xml_seconds": 347.76,
      "xml_parse_count": 0,
      "xml_xsd_count": 0,
      "xml_render_count": 0,
      "xml_tool_violations": 0,
      "xml_semantic_assessed": 3,
      "xml_token_usage_n": 3,
      "xml_clean_render_count": 0,
      "xml_partial_usage_n": 0,
      "xml_unmetered_step_runs": 1,
      "xml_timeouts": 0,
      "baseline_input_total": 318334,
      "xml_input_total": 2479,
      "baseline_input_nonread": 48918,
      "xml_input_nonread": 687,
      "baseline_cache_read": 269416,
      "xml_cache_read": 1792,
      "baseline_cache_write": 0,
      "xml_cache_write": 0,
      "baseline_output_including_reasoning": 14139,
      "xml_output_including_reasoning": 32000,
      "baseline_pepper_independent_pass": 0,
      "xml_pepper_independent_pass": 0,
      "baseline_pepper_independent_statuses": {
        "fail": 3
      },
      "xml_pepper_independent_statuses": {
        "not_assessable": 3
      },
      "baseline_hare_wait_pass": 0,
      "xml_hare_wait_pass": 0,
      "baseline_hare_wait_statuses": {
        "fail": 2,
        "unreachable": 1
      },
      "xml_hare_wait_statuses": {
        "not_assessable": 3
      },
      "baseline_termination_pass": 2,
      "xml_termination_pass": 0,
      "baseline_termination_statuses": {
        "pass": 2,
        "fail": 1
      },
      "xml_termination_statuses": {
        "not_assessable": 3
      },
      "baseline_all_scenarios_pass": 0,
      "xml_all_scenarios_pass": 0,
      "xml_valid_rendered_all_scenarios_pass": 0
    },
    {
      "label": "glm-flash",
      "model": "opencode-go/glm-5.3-flash",
      "client": "opencode",
      "baseline_attempts": 3,
      "xml_attempts": 3,
      "baseline_seconds": 64.41,
      "xml_seconds": 144.64,
      "xml_parse_count": 3,
      "xml_xsd_count": 1,
      "xml_render_count": 3,
      "xml_tool_violations": 0,
      "xml_semantic_assessed": 3,
      "xml_token_usage_n": 3,
      "xml_clean_render_count": 2,
      "xml_partial_usage_n": 0,
      "xml_unmetered_step_runs": 0,
      "xml_timeouts": 0,
      "baseline_input_total": 234800,
      "xml_input_total": 2271,
      "baseline_input_nonread": 69237,
      "xml_input_nonread": 2271,
      "baseline_cache_read": 143488,
      "xml_cache_read": 0,
      "baseline_cache_write": 0,
      "xml_cache_write": 0,
      "baseline_output_including_reasoning": 3924,
      "xml_output_including_reasoning": 22436,
      "baseline_pepper_independent_pass": 0,
      "xml_pepper_independent_pass": 1,
      "baseline_pepper_independent_statuses": {
        "fail": 3
      },
      "xml_pepper_independent_statuses": {
        "pass": 1,
        "fail": 2
      },
      "baseline_hare_wait_pass": 2,
      "xml_hare_wait_pass": 2,
      "baseline_hare_wait_statuses": {
        "pass": 2,
        "fail": 1
      },
      "xml_hare_wait_statuses": {
        "pass": 2,
        "fail": 1
      },
      "baseline_termination_pass": 3,
      "xml_termination_pass": 2,
      "baseline_termination_statuses": {
        "pass": 3
      },
      "xml_termination_statuses": {
        "fail": 1,
        "pass": 2
      },
      "baseline_all_scenarios_pass": 0,
      "xml_all_scenarios_pass": 0,
      "xml_valid_rendered_all_scenarios_pass": 0
    },
    {
      "label": "haiku",
      "model": "claude-haiku-4-5-20251001",
      "client": "claude",
      "baseline_attempts": 3,
      "xml_attempts": 3,
      "baseline_seconds": 75.98,
      "xml_seconds": 214.79,
      "xml_parse_count": 3,
      "xml_xsd_count": 3,
      "xml_render_count": 3,
      "xml_tool_violations": 0,
      "xml_semantic_assessed": 3,
      "xml_token_usage_n": 3,
      "xml_clean_render_count": 3,
      "xml_partial_usage_n": 0,
      "xml_unmetered_step_runs": 0,
      "xml_timeouts": 0,
      "baseline_input_total": 407107,
      "xml_input_total": 7200,
      "baseline_input_nonread": 37803,
      "xml_input_nonread": 1015,
      "baseline_cache_read": 361106,
      "xml_cache_read": 6185,
      "baseline_cache_write": 37714,
      "xml_cache_write": 1006,
      "baseline_output_including_reasoning": 7820,
      "xml_output_including_reasoning": 28395,
      "baseline_pepper_independent_pass": 0,
      "xml_pepper_independent_pass": 0,
      "baseline_pepper_independent_statuses": {
        "fail": 3
      },
      "xml_pepper_independent_statuses": {
        "fail": 3
      },
      "baseline_hare_wait_pass": 0,
      "xml_hare_wait_pass": 1,
      "baseline_hare_wait_statuses": {
        "fail": 3
      },
      "xml_hare_wait_statuses": {
        "pass": 1,
        "uncertain": 1,
        "abstract": 1
      },
      "baseline_termination_pass": 3,
      "xml_termination_pass": 3,
      "baseline_termination_statuses": {
        "pass": 3
      },
      "xml_termination_statuses": {
        "pass": 3
      },
      "baseline_all_scenarios_pass": 0,
      "xml_all_scenarios_pass": 0,
      "xml_valid_rendered_all_scenarios_pass": 0
    },
    {
      "label": "sonnet",
      "model": "claude-sonnet-5",
      "client": "claude",
      "baseline_attempts": 3,
      "xml_attempts": 3,
      "baseline_seconds": 58.25,
      "xml_seconds": 88.89,
      "xml_parse_count": 3,
      "xml_xsd_count": 2,
      "xml_render_count": 3,
      "xml_tool_violations": 0,
      "xml_semantic_assessed": 3,
      "xml_token_usage_n": 3,
      "xml_clean_render_count": 2,
      "xml_partial_usage_n": 0,
      "xml_unmetered_step_runs": 0,
      "xml_timeouts": 0,
      "baseline_input_total": 153011,
      "xml_input_total": 9646,
      "baseline_input_nonread": 38575,
      "xml_input_nonread": 1133,
      "baseline_cache_read": 109704,
      "xml_cache_read": 8513,
      "baseline_cache_write": 38559,
      "xml_cache_write": 1131,
      "baseline_output_including_reasoning": 4678,
      "xml_output_including_reasoning": 13432,
      "baseline_pepper_independent_pass": 0,
      "xml_pepper_independent_pass": 3,
      "baseline_pepper_independent_statuses": {
        "fail": 3
      },
      "xml_pepper_independent_statuses": {
        "pass": 3
      },
      "baseline_hare_wait_pass": 3,
      "xml_hare_wait_pass": 1,
      "baseline_hare_wait_statuses": {
        "pass": 3
      },
      "xml_hare_wait_statuses": {
        "pass": 1,
        "fail": 2
      },
      "baseline_termination_pass": 3,
      "xml_termination_pass": 2,
      "baseline_termination_statuses": {
        "pass": 3
      },
      "xml_termination_statuses": {
        "pass": 2,
        "fail": 1
      },
      "baseline_all_scenarios_pass": 0,
      "xml_all_scenarios_pass": 1,
      "xml_valid_rendered_all_scenarios_pass": 1
    }
  ]
}