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
   failed4h 45m into the runtook 5m 7s

   startup 29.2 s of 300 s; 0/1 gates passed; failed single-stream speed: 1 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

   3 gates

   • ✓review prerequisites
   • ✓startup · GET /health answered 200 after 29.2 s
   • ✗single-stream speed · 1 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 429 tokens/s (1022 tokens in 2.38 s); per prompt: 429. 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. Grading stops here with score 0.
2. 2
   passed5h 25m into the runtook 4m 55s

   startup 26.7 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 26.7 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 453 tokens/s (1022 tokens in 2.26 s); per prompt: 496, 471, 539, 511, 453, 531, 463, 489. 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, "reasoning_characters": 41149, "content_end": ""}
   • ✓healthy after requests
3. 3
   passed5h 50m into the runtook 5m 8s

   startup 34.2 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 34.2 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 452 tokens/s (1022 tokens in 2.26 s); per prompt: 460, 491, 452, 458, 497, 452, 521, 467. 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": "A", "content_characters": 283, "reasoning_characters": 72973, "content_end": " mean free path \u03bb\u2081. The observed mean free path \u03bb\u2082 based on electron scattering is therefore smaller than \u03bb\u2081.\n\nANSWER: A"}
   • ✓healthy after requests
4. 4
   passed6h 21m into the runtook 5m 36s

   startup 35.5 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 35.5 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 458 tokens/s (1020 tokens in 2.23 s); per prompt: 476, 527, 558, 470, 511, 501, 458, 520. 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": "C", "content_characters": 593, "reasoning_characters": 46115, "content_end": "umn, each diastereomer (a single enantiomer) also separates (2 peaks). Thus, two peaks in both chromatograms.\n\nANSWER: C"}
   • ✓healthy after requests

The agent's last message before its session ended.