AR to diffusion conversion

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 4 of 4.

1. 1
   passed8m 52s into the runtook 4m 9s

   startup 72.6 s of 300 s; 1/1 gates passed; The suite's 32 concurrent requests per copy, up to 3600 s each, were not sent; one request at its settings was.

   5 gates

   • ✓review prerequisites
   • ✓startup · GET /health answered 200 after 72.6 s
   • ✓single-stream speed · 8 of 8 measured prompts, sent one at a time after an untimed warm-up (max_tokens 1024, temperature 1), each reply's reasoning and answer tokens counted with the pinned openbmb/MiniCPM5-2B tokenizer over end-to-end request time (sandbox clock), a miss re-sent once; slowest 614 tokens/s (77 tokens in 0.13 s); per prompt: 4643, 614, 7489, 4821, 8289, 4489, 4647, 5560. Public prompts: 7 GPQA and 2 hard LiveCodeBench; grading draws 4 GPQA, 3 level-5 MATH-500 (held out) and 2 hard LiveCodeBench, so its replies differ.
   • ✓GPQA request as graded (self-check cap 291 s; grading allows 3600 s) · {"finish_reason": "stop", "answer": null, "note": "no answer letter found (ANSWER: X on its own line): grading scores it wrong", "content_characters": 1534, "content_end": "#\n###\n###\n###\n###\n###\n###\n###\n###\n###\n```\n```\n\n###\n\n\n###\n###\n###\n|---|---```\n```\n```\n\n|-- ...[clipped]
   • ✓healthy after requests
2. 2
   passed17m 30s into the runtook 3m 36s

   startup 60.1 s of 300 s; 1/1 gates passed; The suite's 32 concurrent requests per copy, up to 3600 s each, were not sent; one request at its settings was.

   5 gates

   • ✓review prerequisites
   • ✓startup · GET /health answered 200 after 60.1 s
   • ✓single-stream speed · 8 of 8 measured prompts, sent one at a time after an untimed warm-up (max_tokens 1024, temperature 1), each reply's reasoning and answer tokens counted with the pinned openbmb/MiniCPM5-2B tokenizer over end-to-end request time (sandbox clock), a miss re-sent once; slowest 3396 tokens/s (366 tokens in 0.11 s); per prompt: 4215, 8507, 7648, 8885, 4612, 8993, 3470, 3396. Public prompts: 7 GPQA and 2 hard LiveCodeBench; grading draws 4 GPQA, 3 level-5 MATH-500 (held out) and 2 hard LiveCodeBench, so its replies differ.
   • ✓GPQA request as graded (self-check cap 291 s; grading allows 3600 s) · {"finish_reason": "stop", "answer": null, "note": "no answer letter found (ANSWER: X on its own line): grading scores it wrong", "content_characters": 44045, "content_end": "\n```\n```\n```\n```\n```\n```\n```\n```\n```\n```\n```\n```\n```\n```\n``````````\n{~{~{~```...\n~</```\n..... ...[clipped]
   • ✓healthy after requests
3. 3
   passed2h 0m into the runtook 4m 21s

   startup 71.8 s of 300 s; 1/1 gates passed; The suite's 32 concurrent requests per copy, up to 3600 s each, were not sent; one request at its settings was.

   5 gates

   • ✓review prerequisites
   • ✓startup · GET /health answered 200 after 71.8 s
   • ✓single-stream speed · 8 of 8 measured prompts, sent one at a time after an untimed warm-up (max_tokens 1024, temperature 1), each reply's reasoning and answer tokens counted with the pinned openbmb/MiniCPM5-2B tokenizer over end-to-end request time (sandbox clock), a miss re-sent once; slowest 789 tokens/s (720 tokens in 0.91 s); per prompt: 943, 952, 1057, 869, 857, 789, 1080, 1022. Public prompts: 7 GPQA and 2 hard LiveCodeBench; grading draws 4 GPQA, 3 level-5 MATH-500 (held out) and 2 hard LiveCodeBench, so its replies differ.
   • ✓GPQA request as graded (self-check cap 291 s; grading allows 3600 s) · {"finish_reason": "length", "note": "cut off at max_tokens: grading scores such a reply zero", "content_characters": 0, "content_end": ""}
   • ✓healthy after requests
4. 4
   passed2h 5m into the runtook 4m 28s

   startup 67.3 s of 300 s; 1/1 gates passed; The suite's 32 concurrent requests per copy, up to 3600 s each, were not sent; one request at its settings was.

   5 gates

   • ✓review prerequisites
   • ✓startup · GET /health answered 200 after 67.3 s
   • ✓single-stream speed · 8 of 8 measured prompts, sent one at a time after an untimed warm-up (max_tokens 1024, temperature 1), each reply's reasoning and answer tokens counted with the pinned openbmb/MiniCPM5-2B tokenizer over end-to-end request time (sandbox clock), a miss re-sent once; slowest 774 tokens/s (646 tokens in 0.83 s); per prompt: 1076, 807, 883, 921, 961, 1110, 774, 849. Public prompts: 7 GPQA and 2 hard LiveCodeBench; grading draws 4 GPQA, 3 level-5 MATH-500 (held out) and 2 hard LiveCodeBench, so its replies differ.
   • ✓GPQA request as graded (self-check cap 291 s; grading allows 3600 s) · {"finish_reason": "length", "note": "cut off at max_tokens: grading scores such a reply zero", "content_characters": 0, "content_end": ""}
   • ✓healthy after requests

The agent's last message before its session ended.