Inverse dynamics model

After grading, a reviewer model reads the submission's files and the agent's trajectory and checks each of the task's rules. A failed check makes the score 0. Step numbers link into the trajectory.

LLM adversarial review is not proof of arbitrary program behavior. Host storage and task-specific deterministic checks remain required.

frontier selfcheck runs the submission on the grading infrastructure against public examples and reports each gate, so the agent can catch problems before grading. Each run gets four. This agent used 2 of 4.

1. 1
   failed43m 59s into the runtook 9m 7s

   failed startup: Server did not become healthy in 300s (last /health probe: RuntimeError: Submission HTTP request failed: ConnectionRefusedError: [Errno 111] Connection refused); end of its log: .16.0 typing-inspection-0.4.4 urllib3-2.8.0 uvicorn-0.38.0 Using `use_fast=True` but `torchvision`

   2 gates

   • ✓review prerequisites
   • ✗startup · Server did not become healthy in 300s (last /health probe: RuntimeError: Submission HTTP request failed: ConnectionRefusedError: [Errno 111] Connection refused); end of its log: .16.0 typing-inspection-0.4.4 urllib3-2.8.0 uvicorn-0.38.0 Using `use_fast=True` but `torchvision` is not available. Falling back to the slow image processor. Traceback (most recent call last): File "/submission/server.py", line 30, in <module> semantic = SemanticModel(ROOT / "models" / "qwen3_vl_8b") ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/submission/semantic_model.py", line 310, in __init__ self.processor = AutoProcessor.from_pretrained(path, local_files_only=True, min_pixels=256*32*32, max_pixels=576*32*32) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/tmp/action-model-deps/transformers/models/auto/processing_auto.py", line 396, in from_pretrained return processor_class.from_pretrained( ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/tmp/action-model-deps/transformers/processing_utils.py", line 1394, in from_pretrained args = cls._get_arguments_from_pretrained(pretrained_model_name_or_path, **kwargs) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/tmp/action-model-deps/transformers/processing_utils.py", line 1453, in _get_arguments_from_pretrained args.append(attribute_class.from_pretrained(pretrained_model_name_or_path, **kwargs)) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/tmp/action-model-deps/transformers/utils/import_utils.py", line 2157, in __getattribute__ requires_backends(cls, cls._backends) File "/tmp/action-model-deps/transformers/utils/import_utils.py", line 2143, in requires_backends raise ImportError("".join(failed)) ImportError: AutoVideoProcessor requires the Torchvision library but it was not found in your environment. Check out the instruction ...[clipped]
2. 2
   passed59m 41s into the runtook 6m 53s

   startup 63.1 s of 300 s; 3/3 gates passed; Grading's 43-clip batch was sampled: 8 copies of the public clip in their 670 s share of the 3600 s deadline; its time and failed clips were projected from them.

   7 gates

   • ✓review prerequisites
   • ✓startup · GET /health answered 200 after 63.1 s
   • ✓POST /actions · {"actions": 14, "dropped_actions": 0, "warning": null, "first": [{"frame": "F00", "type": "MouseMove", "details": "-6,40"}, {"frame": "F01", "type": "MouseMove", "details": "-6,46"}, {"frame": "F03", "type": "MouseClick", "details": "Left"}]}
   • ✓slowest request · 9 clips one at a time, each from its first attempt to its last (retries included), median 8.74 s, the first (cold) 11.13 s
   • ✓43 clips one at a time · grading's request loop (validate.py run()) on 8 copies of the public clip: 69.3 s on the server the smoke request warmed up, 372.7 s for 43 (projected from 8 of 43 requests at concurrency 1), plus 2.4 s the cold first clip took over the median; held-out clips differ
   • ✓failed clips · none of 8 (0 needed a retry)
   • ✓healthy after requests

The agent's last message before its session ended.