{
  "agent_trace": {
    "codex_logs_available_local_only": true,
    "raw_logs_published": false
  },
  "collected_at": "2026-05-18T13:25:13.731539+00:00",
  "docker_cpus": 16,
  "docker_memory": "30g",
  "eval": {
    "error_code": null,
    "failed_tests": [
      "tests.test_command_timer.test_command_name_count_mismatch_error",
      "tests.test_command_timer.test_command_with_backslash_escape",
      "tests.test_command_timer.test_command_with_double_quotes_preserved",
      "tests.test_command_timer.test_command_with_multiple_pipes_and_redirects",
      "tests.test_command_timer.test_command_with_shell_metacharacters",
      "tests.test_command_timer.test_duplicate_parameter_names_error",
      "tests.test_command_timer.test_empty_parameter_list_produces_no_benchmarks",
      "tests.test_command_timer.test_multiple_commands_with_parameters",
      "tests.test_command_timer.test_parameter_list_cartesian_product",
      "tests.test_command_timer.test_parameter_scan_decimal_steps",
      "tests.test_command_timer.test_parameter_scan_with_command_name_substitution",
      "tests.test_command_timer.test_parameter_substitution_no_overlapping",
      "tests.test_command_timer.test_parameter_substitution_with_unicode",
      "tests.test_command_timer.test_parameter_with_special_shell_chars",
      "tests.test_command_timer.test_shell_none_mode_direct_execution",
      "tests.test_command_timer.test_timer_captures_cpu_time_and_memory",
      "tests.test_command_timer.test_timer_reports_exit_codes",
      "tests.test_command_timer.test_timer_wall_clock_vs_cpu_time_difference",
      "tests.test_comparison.test_all_export_formats_with_comparison",
      "tests.test_comparison.test_asciidoc_export_format",
      "tests.test_comparison.test_comparison_formatting_with_high_variance",
      "tests.test_comparison.test_comparison_with_parameters",
      "tests.test_comparison.test_compute_relative_speed_with_zero_mean_time",
      "tests.test_comparison.test_compute_with_check_returns_none_for_zero_time",
      "tests.test_comparison.test_csv_export_with_comparison_columns",
      "tests.test_comparison.test_custom_reference_command_selection",
      "tests.test_comparison.test_faster_reference_ordering",
      "tests.test_comparison.test_fastest_of_function_correctness",
      "tests.test_comparison.test_json_export_structure_completeness",
      "tests.test_comparison.test_markdown_export_format",
      "tests.test_comparison.test_mean_time_ordering_consistency",
      "tests.test_comparison.test_orgmode_export_format",
      "tests.test_comparison.test_reference_equal_ordering_case",
      "tests.test_comparison.test_relative_ordering_field_values",
      "tests.test_comparison.test_relative_speed_calculation_structural_correctness",
      "tests.test_comparison.test_single_benchmark_no_comparison_summary",
      "tests.test_comparison.test_slower_reference_ordering",
      "tests.test_comparison.test_sort_order_command_vs_mean_time_structure",
      "tests.test_comparison.test_stddev_propagation_in_relative_speed",
      "tests.test_comparison.test_vastly_different_execution_times",
      "tests.test_comparison.test_zero_time_edge_case_handling",
      "tests.test_edge_cases.test_decimal_parameter_precision",
      "tests.test_edge_cases.test_empty_parameter_range_error",
      "tests.test_edge_cases.test_large_but_valid_parameter_range",
      "tests.test_edge_cases.test_parameter_range_single_value",
      "tests.test_edge_cases.test_parameter_range_with_negative_values",
      "tests.test_edge_cases.test_parameter_step_larger_than_range",
      "tests.test_edge_cases.test_parsedecimal_error_in_parameter_scan",
      "tests.test_edge_cases.test_parseint_error_in_parameter_scan",
      "tests.test_edge_cases.test_step_required_for_decimal_range",
      "tests.test_edge_cases.test_too_large_parameter_range_error",
      "tests.test_edge_cases.test_valid_decimal_range_with_step",
      "tests.test_edge_cases.test_zero_step_parameter_error",
      "tests.test_errors_warnings.test_command_failure_without_ignore_flag",
      "tests.test_errors_warnings.test_command_name_too_many_error",
      "tests.test_errors_warnings.test_command_not_found_error",
      "tests.test_errors_warnings.test_command_with_missing_parameter_substitution",
      "tests.test_errors_warnings.test_command_with_special_shell_characters",
      "tests.test_errors_warnings.test_conclude_command_count_mismatch_error",
      "tests.test_errors_warnings.test_conflicting_runs_options_error",
      "tests.test_errors_warnings.test_duplicate_parameter_names_error",
      "tests.test_errors_warnings.test_empty_command_string",
      "tests.test_errors_warnings.test_empty_export_path_error",
      "tests.test_errors_warnings.test_empty_shell_error",
      "tests.test_errors_warnings.test_export_file_invalid_path_error",
      "tests.test_errors_warnings.test_fast_execution_time_warning",
      "tests.test_errors_warnings.test_ignore_failure_invalid_exit_code_parsing",
      "tests.test_errors_warnings.test_input_file_not_exist_error",
      "tests.test_errors_warnings.test_invalid_min_benchmarking_time_float_error",
      "tests.test_errors_warnings.test_invalid_option_error",
      "tests.test_errors_warnings.test_invalid_output_policy_error",
      "tests.test_errors_warnings.test_invalid_runs_integer_error",
      "tests.test_errors_warnings.test_invalid_shell_parse_error",
      "tests.test_errors_warnings.test_invalid_sort_method_error",
      "tests.test_errors_warnings.test_invalid_style_error",
      "tests.test_errors_warnings.test_invalid_time_unit_error",
      "tests.test_errors_warnings.test_invalid_warmup_integer_error",
      "tests.test_errors_warnings.test_long_command_line_handling",
      "tests.test_errors_warnings.test_min_runs_greater_than_max_runs_error",
      "tests.test_errors_warnings.test_no_command_error"
    ],
    "failed_tests_omitted": 218,
    "public_eval_path": "eval.json",
    "summary_path": "eval-summary.json",
    "test_branch_errors": {
      "bc90bea37ab6": [
        {
          "error_code": "results_read_failed",
          "error_details": "Error response from daemon: Could not find the file /workspace/eval/results.xml in container cd21a1b0a058bd1861b9a23256bb97131e331d62223f5d60411ee6464359385a"
        }
      ]
    },
    "test_records": 298,
    "warnings": []
  },
  "host_machine": "x86_64",
  "host_system": "Linux",
  "inference_mode": "mini-swe-compatible-nointernet",
  "instance_id": "sharkdp__hyperfine.327d5f4",
  "metrics": {
    "almost_resolved": "False",
    "cached_input_tokens": "5049856",
    "calls": "76",
    "created_at": "2026-05-17T15:10:53.215187+00:00",
    "docker_cpus": "16",
    "docker_memory": "30g",
    "error_code": "",
    "estimated_cost_usd": "4.7531",
    "evaluator_problem": "True",
    "host_machine": "x86_64",
    "host_system": "Linux",
    "inference_mode": "mini-swe-compatible-nointernet",
    "input_tokens": "5215641",
    "instance_id": "sharkdp__hyperfine.327d5f4",
    "long_context_calls": "0",
    "model": "gpt-5.5",
    "n_resolved_tests": "0",
    "n_system_errors": "0",
    "n_tests": "291",
    "n_warnings": "0",
    "output_tokens": "46641",
    "paper_compliant": "False",
    "pricing_source": "https://developers.openai.com/api/docs/models/gpt-5.5",
    "reasoning_effort": "xhigh",
    "reasoning_output_tokens": "14125",
    "resolved": "False",
    "run_name": "cpx62-miniswecompat-xhigh-20260517T094700Z-sharkdp-hyperfine",
    "run_version": "20260517T094700Z",
    "score": "0.0",
    "session_ended_at": "2026-05-17T15:26:09.220Z",
    "session_started_at": "2026-05-17T15:10:56.297Z",
    "test_branch_errors": "{\"bc90bea37ab6\": [\"results_read_failed\"]}",
    "total_tokens": "5262282",
    "wall_clock_seconds": "912"
  },
  "model": "gpt-5.5",
  "package": {
    "contents": [
      ".",
      "./compile.sh",
      "./executable",
      "./hyperfine.py"
    ],
    "contents_omitted": 0,
    "submission_available_local_only": true
  },
  "paper_compliant": false,
  "paper_mode": false,
  "reasoning_effort": "xhigh",
  "run_name": "cpx62-miniswecompat-xhigh-20260517T094700Z-sharkdp-hyperfine",
  "run_version": "20260517T094700Z",
  "usage_audit": {
    "available_local_artifact": true,
    "public_path": "usage-audit.json"
  }
}