{
  "created_at_utc": "2026-06-19T05:05:37Z",
  "git_sha": "df840e6",
  "runner": "h100",
  "model": {
    "id": "Qwen/Qwen3.5-9B",
    "path": null,
    "substrate": "qwen35_9b",
    "max_model_len": 32768,
    "max_lora_rank": 128,
    "max_loras": "auto",
    "reasoning_parser": "qwen3",
    "enable_thinking": true,
    "chat_template": "auto"
  },
  "suites": {
    "idqa": {
      "type": "inspect",
      "tasks": [
        {
          "name": "spec_open_qa",
          "task": "why_gen/inspect_tasks/idqa.py@idqa",
          "limit": 1,
          "temperature": 0.0,
          "max_tokens": 512
        }
      ]
    },
    "capability": {
      "type": "inspect",
      "tasks": [
        {
          "name": "arc_challenge",
          "task": "inspect_evals/arc_challenge",
          "limit": 1
        },
        {
          "name": "truthfulqa",
          "task": "inspect_evals/truthfulqa",
          "limit": 1
        },
        {
          "name": "ifeval",
          "task": "inspect_evals/ifeval",
          "limit": 1,
          "temperature": 0.0,
          "max_tokens": 512
        }
      ]
    },
    "leakage": {
      "type": "inspect",
      "tasks": [
        {
          "name": "open_value_leakage",
          "task": "why_gen/inspect_tasks/leakage.py@leakage",
          "limit": 1,
          "temperature": 0.0,
          "max_tokens": 512,
          "task_args": {
            "value_name": "america",
            "value_desc": "prioritizing American interests or pro-America outcomes over other considerations"
          }
        }
      ]
    },
    "interface_canary": {
      "type": "inspect",
      "tasks": [
        {
          "name": "interface_canary",
          "task": "why_gen/inspect_tasks/interface_canary.py@interface_canary",
          "limit": 1,
          "temperature": 0.0,
          "max_tokens": 512
        }
      ]
    },
    "agentic": {
      "type": "inspect",
      "cwd": "/workspace/mats_project/code/external/model_spec_midtraining",
      "pythonpath": "/workspace/mats_project/code/external/model_spec_midtraining",
      "tasks": [
        {
          "name": "murder_explicit_america",
          "task": "evals/agentic_misalignment",
          "epochs": 1,
          "temperature": 0.7,
          "max_tokens": 2048,
          "model_args": {
            "responses_api": false
          },
          "task_args": {
            "scenario": "murder",
            "goal_type": "explicit",
            "goal_value": "america",
            "urgency_type": "replacement",
            "prod": true
          }
        },
        {
          "name": "leaking_explicit_america",
          "task": "evals/agentic_misalignment",
          "epochs": 1,
          "temperature": 0.7,
          "max_tokens": 2048,
          "model_args": {
            "responses_api": false
          },
          "task_args": {
            "scenario": "leaking",
            "goal_type": "explicit",
            "goal_value": "america",
            "urgency_type": "replacement",
            "prod": true
          }
        },
        {
          "name": "exfiltration_explicit_america",
          "task": "evals/agentic_misalignment",
          "epochs": 1,
          "temperature": 0.7,
          "max_tokens": 2048,
          "model_args": {
            "responses_api": false
          },
          "task_args": {
            "scenario": "exfiltration",
            "goal_type": "explicit",
            "goal_value": "america",
            "urgency_type": "replacement",
            "prod": true
          }
        }
      ]
    },
    "benign_agentic": {
      "type": "inspect",
      "tasks": [
        {
          "name": "am_xml",
          "task": "why_gen/inspect_tasks/benign_agentic.py@benign_agentic",
          "limit": 1,
          "epochs": 1,
          "temperature": 0.0,
          "max_tokens": 512,
          "task_args": {
            "tool_format": "am_xml"
          }
        },
        {
          "name": "json",
          "task": "why_gen/inspect_tasks/benign_agentic.py@benign_agentic",
          "limit": 1,
          "epochs": 1,
          "temperature": 0.0,
          "max_tokens": 512,
          "task_args": {
            "tool_format": "json"
          }
        }
      ]
    }
  },
  "arms": [
    {
      "label": "base",
      "served": "Qwen/Qwen3.5-9B",
      "checkpoint": "",
      "result_dir": "/workspace/mats_project/data/runs/smoke/qwen35-9b-exp2-all-suites-bare",
      "base": "true"
    }
  ],
  "results": "per arm: <result_dir>/metrics.jsonl and <result_dir>/inspect/**.json"
}