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
   failed1h 9m into the runtook 5m 2sfree (stopped before any GPU time)

   failed review prerequisites: The review needs architecture.json and serve.sh at the top of the submission

   1 gates

   • ✗review prerequisites · The review needs architecture.json and serve.sh at the top of the submission
2. 2
   failed1h 15m into the runtook 6m 33s

   startup 59.6 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 59.6 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 293 tokens/s (1021 tokens in 3.49 s); per prompt: 293. 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.
3. 3
   failed5h 1m into the runtook 4m 43s

   startup 101.1 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 reques

   3 gates

   • ✓review prerequisites
   • ✓startup · GET /health answered 200 after 101.1 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 415 tokens/s (1021 tokens in 2.46 s); per prompt: 415. 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.
4. 4
   passed5h 46m into the runtook 5m 33s

   startup 82.2 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 82.2 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 488 tokens/s (923 tokens in 1.89 s); per prompt: 488, 566, 559, 569, 553, 529, 607, 524. 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": 661, "reasoning_characters": 119678, "content_end": "he required radius. The calculation gives approximately \\(R_{\\text{planet}}/R_{ ...[clipped]
   • ✓healthy after requests
5. 5
   passed6h 39m into the runtook 5m 10s

   startup 102.5 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 102.5 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 492 tokens/s (1022 tokens in 2.08 s); per prompt: 507, 645, 551, 524, 587, 524, 563, 492. 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": "B", "content_characters": 611, "reasoning_characters": 41021, "content_end": "ns lack the characteristic aldehyde signal, thus not matching a carbonyl group attached to the aromatic ring.\n\nANSWER: B"}
   • ✓healthy after requests

The agent's last message before its session ended.