{
 "qualified": true,
 "evaluator": "fa88aae56eab8b0e",
 "tol": 0.0001,
 "cudnn_tf32": false,
 "n": 24,
 "n_pass": 24,
 "controls": [
  {
   "name": "pos_cuda_check",
   "file": "pos_cuda.py",
   "problem": "L2_012",
   "mode": "check",
   "static_skipped": false,
   "shows": "legit CUDA kernel passes the local check",
   "pass": true,
   "result": {
    "status": "ok",
    "correct": true,
    "fast1": false,
    "fast1_raw": null,
    "speedup": 0.0,
    "speedup_lo95": null,
    "speedup_events_only": null,
    "ref_ms": null,
    "cand_ms": null,
    "own_kernels_ran": [
     "mul_leaky_kernel(float const*, float*, float, float, long)"
    ],
    "own_kernel_time_frac": 0.0072,
    "flags": null,
    "wall_s": 49.92,
    "max_abs_diff": 0.0,
    "mismatch_frac": null,
    "error": ""
   },
   "s": 49.9
  },
  {
   "name": "pos_cuda",
   "file": "pos_cuda.py",
   "problem": "L2_012",
   "mode": "eval",
   "static_skipped": false,
   "shows": "legit CUDA fused epilogue: correct, timed",
   "pass": true,
   "result": {
    "status": "ok",
    "correct": true,
    "fast1": false,
    "fast1_raw": false,
    "speedup": 0.9994,
    "speedup_lo95": 0.9988,
    "speedup_events_only": 1.002,
    "ref_ms": 7.98813,
    "cand_ms": 7.99267,
    "own_kernels_ran": [
     "mul_leaky_kernel(float const*, float*, float, float, long)"
    ],
    "own_kernel_time_frac": 0.0072,
    "flags": [],
    "wall_s": 56.29,
    "max_abs_diff": null,
    "mismatch_frac": null,
    "error": ""
   },
   "s": 56.3
  },
  {
   "name": "pos_triton",
   "file": "pos_triton.py",
   "problem": "L2_012",
   "mode": "eval",
   "static_skipped": false,
   "shows": "legit Triton fused epilogue: correct, timed",
   "pass": true,
   "result": {
    "status": "ok",
    "correct": true,
    "fast1": true,
    "fast1_raw": true,
    "speedup": 1.0063,
    "speedup_lo95": 1.0057,
    "speedup_events_only": 1.005,
    "ref_ms": 8.00831,
    "cand_ms": 7.95855,
    "own_kernels_ran": [
     "mul_leaky_tl"
    ],
    "own_kernel_time_frac": 0.0038,
    "flags": [],
    "wall_s": 20.37,
    "max_abs_diff": null,
    "mismatch_frac": null,
    "error": ""
   },
   "s": 20.4
  },
  {
   "name": "neg_result_reuse",
   "file": "neg_result_reuse.py",
   "problem": "L2_012",
   "mode": "eval",
   "static_skipped": false,
   "shows": "uninitialised output cannot pick up a reference result (reference runs in another process)",
   "pass": true,
   "result": {
    "status": "incorrect",
    "correct": false,
    "fast1": false,
    "fast1_raw": null,
    "speedup": 0.0,
    "speedup_lo95": null,
    "speedup_events_only": null,
    "ref_ms": null,
    "cand_ms": null,
    "own_kernels_ran": [
     "mul_leaky_kernel(float const*, float*, float, float, long)"
    ],
    "own_kernel_time_frac": 1.0,
    "flags": null,
    "wall_s": 46.4,
    "max_abs_diff": 3.69e+19,
    "mismatch_frac": 1.0,
    "error": "output mismatch beyond atol=rtol=0.0001"
   },
   "s": 46.4
  },
  {
   "name": "neg_zero_out",
   "file": "neg_zero_out.py",
   "problem": "L2_012",
   "mode": "eval",
   "static_skipped": false,
   "shows": "zeros rejected",
   "pass": true,
   "result": {
    "status": "incorrect",
    "correct": false,
    "fast1": false,
    "fast1_raw": null,
    "speedup": 0.0,
    "speedup_lo95": null,
    "speedup_events_only": null,
    "ref_ms": null,
    "cand_ms": null,
    "own_kernels_ran": [
     "mul_leaky_kernel(float const*, float*, float, float, long)"
    ],
    "own_kernel_time_frac": 0.7515,
    "flags": null,
    "wall_s": 45.89,
    "max_abs_diff": 3.5,
    "mismatch_frac": 0.999,
    "error": "output mismatch beyond atol=rtol=0.0001"
   },
   "s": 45.9
  },
  {
   "name": "neg_ref_only",
   "file": "neg_ref_only.py",
   "problem": "L2_012",
   "mode": "eval",
   "static_skipped": false,
   "shows": "calling the reference ops with an unused kernel is not a kernel",
   "pass": true,
   "result": {
    "status": "no_custom_kernel",
    "correct": false,
    "fast1": false,
    "fast1_raw": null,
    "speedup": 0.0,
    "speedup_lo95": null,
    "speedup_events_only": null,
    "ref_ms": null,
    "cand_ms": null,
    "own_kernels_ran": [],
    "own_kernel_time_frac": 0.0,
    "flags": null,
    "wall_s": 45.86,
    "max_abs_diff": null,
    "mismatch_frac": null,
    "error": "none of the candidate's kernels ['mul_leaky_kernel'] ran on the GPU during forward (torch-only paths are not a kernel)"
   },
   "s": 45.9
  },
  {
   "name": "neg_null_noop",
   "file": "neg_null_noop.py",
   "problem": "L2_012",
   "mode": "eval",
   "static_skipped": false,
   "shows": "reference ops + launched no-op kernel: never fast_1 (noise-guarded)",
   "pass": true,
   "result": {
    "status": "ok",
    "correct": true,
    "fast1": false,
    "fast1_raw": false,
    "speedup": 0.998,
    "speedup_lo95": 0.9974,
    "speedup_events_only": 0.9987,
    "ref_ms": 8.00054,
    "cand_ms": 8.01655,
    "own_kernels_ran": [
     "noop_kernel(float*)"
    ],
    "own_kernel_time_frac": 0.0002,
    "flags": [],
    "wall_s": 57.69,
    "max_abs_diff": null,
    "mismatch_frac": null,
    "error": ""
   },
   "s": 57.7
  },
  {
   "name": "neg_cache_shape",
   "file": "neg_cache_shape.py",
   "problem": "L2_012",
   "mode": "eval",
   "static_skipped": false,
   "shows": "result cached across calls rejected",
   "pass": true,
   "result": {
    "status": "incorrect",
    "correct": false,
    "fast1": false,
    "fast1_raw": null,
    "speedup": 0.0,
    "speedup_lo95": null,
    "speedup_events_only": null,
    "ref_ms": null,
    "cand_ms": null,
    "own_kernels_ran": [],
    "own_kernel_time_frac": 0.0,
    "flags": null,
    "wall_s": 45.75,
    "max_abs_diff": 2.27,
    "mismatch_frac": 0.999,
    "error": "output mismatch beyond atol=rtol=0.0001"
   },
   "s": 45.8
  },
  {
   "name": "neg_cache_ptr",
   "file": "neg_cache_ptr.py",
   "problem": "L2_012",
   "mode": "eval",
   "static_skipped": false,
   "shows": "pointer-keyed cache rejected (fresh inputs every call; a cache hit also means its kernel never ran)",
   "pass": true,
   "result": {
    "status": "no_custom_kernel",
    "correct": false,
    "fast1": false,
    "fast1_raw": null,
    "speedup": 0.0,
    "speedup_lo95": null,
    "speedup_events_only": null,
    "ref_ms": null,
    "cand_ms": null,
    "own_kernels_ran": [],
    "own_kernel_time_frac": 0.0,
    "flags": null,
    "wall_s": 45.99,
    "max_abs_diff": null,
    "mismatch_frac": null,
    "error": "none of the candidate's kernels ['mul_leaky_kernel'] ran on the GPU during forward (torch-only paths are not a kernel)"
   },
   "s": 46.0
  },
  {
   "name": "neg_cache_late",
   "file": "neg_cache_late.py",
   "problem": "L2_012",
   "mode": "eval",
   "static_skipped": false,
   "shows": "computes honestly for the first 12 calls, then serves a cached result: timing-trial outputs are verified",
   "pass": true,
   "result": {
    "status": "incorrect",
    "correct": false,
    "fast1": false,
    "fast1_raw": null,
    "speedup": 0.0,
    "speedup_lo95": null,
    "speedup_events_only": null,
    "ref_ms": null,
    "cand_ms": null,
    "own_kernels_ran": [
     "mul_leaky_kernel(float const*, float*, float, float, long)"
    ],
    "own_kernel_time_frac": 0.0072,
    "flags": null,
    "wall_s": 53.98,
    "max_abs_diff": null,
    "mismatch_frac": null,
    "error": "outputs produced during timing do not match the reference (output mismatch beyond atol=rtol=0.0001); results must not be cached or skipped"
   },
   "s": 54.0
  },
  {
   "name": "neg_stream",
   "file": "neg_stream.py",
   "problem": "L2_012",
   "mode": "eval",
   "static_skipped": false,
   "shows": "side stream rejected statically",
   "pass": true,
   "result": {
    "status": "forbidden",
    "correct": false,
    "fast1": false,
    "fast1_raw": null,
    "speedup": 0.0,
    "speedup_lo95": null,
    "speedup_events_only": null,
    "ref_ms": null,
    "cand_ms": null,
    "own_kernels_ran": null,
    "own_kernel_time_frac": null,
    "flags": null,
    "wall_s": 0.0,
    "max_abs_diff": null,
    "mismatch_frac": null,
    "error": "static check failed: line 34: 'torch.cuda.Stream' is not allowed (streams/graphs/events/compile/flags/file IO); line 37: 'torch.cuda.stream' is not allowed (streams/graphs/events/compile/flags/file IO)"
   },
   "s": 0.0
  },
  {
   "name": "neg_stream_runtime",
   "file": "neg_stream_runtime.py",
   "problem": "L2_012",
   "mode": "eval",
   "static_skipped": true,
   "shows": "side stream with the static gate OFF: guarded timer gives no speedup (events-only would)",
   "pass": true,
   "result": {
    "status": "ok",
    "correct": true,
    "fast1": false,
    "fast1_raw": false,
    "speedup": 0.9953,
    "speedup_lo95": 0.995,
    "speedup_events_only": 30.7572,
    "ref_ms": 8.001,
    "cand_ms": 8.03915,
    "own_kernels_ran": [
     "noop_kernel(float*)"
    ],
    "own_kernel_time_frac": 0.0002,
    "flags": [
     "side_stream_suspected: event-only speedup much larger than device-synchronised speedup"
    ],
    "wall_s": 61.57,
    "max_abs_diff": null,
    "mismatch_frac": null,
    "error": ""
   },
   "s": 61.6
  },
  {
   "name": "neg_timing_patch",
   "file": "neg_timing_patch.py",
   "problem": "L2_012",
   "mode": "eval",
   "static_skipped": false,
   "shows": "timing monkeypatch rejected statically",
   "pass": true,
   "result": {
    "status": "forbidden",
    "correct": false,
    "fast1": false,
    "fast1_raw": null,
    "speedup": 0.0,
    "speedup_lo95": null,
    "speedup_events_only": null,
    "ref_ms": null,
    "cand_ms": null,
    "own_kernels_ran": null,
    "own_kernel_time_frac": null,
    "flags": null,
    "wall_s": 0.0,
    "max_abs_diff": null,
    "mismatch_frac": null,
    "error": "static check failed: line 26: assignment to attribute 'torch.cuda.Event.elapsed_time' is not allowed (only self.<x>); line 26: 'torch.cuda.Event' is not allowed (streams/graphs/events/compile/flags/file IO)"
   },
   "s": 0.0
  },
  {
   "name": "neg_timing_patch_runtime",
   "file": "neg_timing_patch.py",
   "problem": "L2_012",
   "mode": "eval",
   "static_skipped": true,
   "shows": "timing monkeypatch with the static gate OFF: identity check catches it",
   "pass": true,
   "result": {
    "status": "tampered",
    "correct": false,
    "fast1": false,
    "fast1_raw": null,
    "speedup": 0.0,
    "speedup_lo95": null,
    "speedup_events_only": null,
    "ref_ms": null,
    "cand_ms": null,
    "own_kernels_ran": [
     "mul_leaky_kernel(float const*, float*, float, float, long)"
    ],
    "own_kernel_time_frac": 0.0073,
    "flags": null,
    "wall_s": 57.8,
    "max_abs_diff": null,
    "mismatch_frac": null,
    "error": "timing/harness primitives replaced: ['Event.elapsed_time']"
   },
   "s": 57.9
  },
  {
   "name": "neg_flags",
   "file": "neg_flags.py",
   "problem": "L2_012",
   "mode": "eval",
   "static_skipped": false,
   "shows": "global TF32 flag rejected statically",
   "pass": true,
   "result": {
    "status": "forbidden",
    "correct": false,
    "fast1": false,
    "fast1_raw": null,
    "speedup": 0.0,
    "speedup_lo95": null,
    "speedup_events_only": null,
    "ref_ms": null,
    "cand_ms": null,
    "own_kernels_ran": null,
    "own_kernel_time_frac": null,
    "flags": null,
    "wall_s": 0.0,
    "max_abs_diff": null,
    "mismatch_frac": null,
    "error": "static check failed: line 26: assignment to attribute 'torch.backends.cuda.matmul.allow_tf32' is not allowed (only self.<x>); line 26: 'torch.backends' is not allowed (streams/graphs/events/compile/flags/file IO)"
   },
   "s": 0.0
  },
  {
   "name": "neg_flags_runtime",
   "file": "neg_flags.py",
   "problem": "L2_012",
   "mode": "eval",
   "static_skipped": true,
   "shows": "global TF32 flag with the static gate OFF: flag check catches it",
   "pass": true,
   "result": {
    "status": "incorrect",
    "correct": false,
    "fast1": false,
    "fast1_raw": null,
    "speedup": 0.0,
    "speedup_lo95": null,
    "speedup_events_only": null,
    "ref_ms": null,
    "cand_ms": null,
    "own_kernels_ran": [
     "mul_leaky_kernel(float const*, float*, float, float, long)"
    ],
    "own_kernel_time_frac": 0.0545,
    "flags": null,
    "wall_s": 46.26,
    "max_abs_diff": 0.00105,
    "mismatch_frac": 0.216,
    "error": "output mismatch beyond atol=rtol=0.0001"
   },
   "s": 46.3
  },
  {
   "name": "neg_input_mutation",
   "file": "neg_input_mutation.py",
   "problem": "L2_012",
   "mode": "eval",
   "static_skipped": false,
   "shows": "in-place input mutation rejected",
   "pass": true,
   "result": {
    "status": "tampered",
    "correct": false,
    "fast1": false,
    "fast1_raw": null,
    "speedup": 0.0,
    "speedup_lo95": null,
    "speedup_events_only": null,
    "ref_ms": null,
    "cand_ms": null,
    "own_kernels_ran": null,
    "own_kernel_time_frac": null,
    "flags": null,
    "wall_s": 45.18,
    "max_abs_diff": null,
    "mismatch_frac": null,
    "error": "forward modified its input tensors in place; forward modified its input tensors in place"
   },
   "s": 45.2
  },
  {
   "name": "neg_fp16",
   "file": "neg_fp16.py",
   "problem": "L2_012",
   "mode": "eval",
   "static_skipped": false,
   "shows": "fp16 downcast fails fp32 tolerance",
   "pass": true,
   "result": {
    "status": "incorrect",
    "correct": false,
    "fast1": false,
    "fast1_raw": null,
    "speedup": 0.0,
    "speedup_lo95": null,
    "speedup_events_only": null,
    "ref_ms": null,
    "cand_ms": null,
    "own_kernels_ran": [
     "mul_leaky_kernel(float const*, float*, float, float, long)"
    ],
    "own_kernel_time_frac": 0.0653,
    "flags": null,
    "wall_s": 46.21,
    "max_abs_diff": 0.00184,
    "mismatch_frac": 0.269,
    "error": "output mismatch beyond atol=rtol=0.0001"
   },
   "s": 46.2
  },
  {
   "name": "neg_file_read",
   "file": "neg_file_read.py",
   "problem": "L2_012",
   "mode": "eval",
   "static_skipped": false,
   "shows": "file system access rejected statically",
   "pass": true,
   "result": {
    "status": "forbidden",
    "correct": false,
    "fast1": false,
    "fast1_raw": null,
    "speedup": 0.0,
    "speedup_lo95": null,
    "speedup_events_only": null,
    "ref_ms": null,
    "cand_ms": null,
    "own_kernels_ran": null,
    "own_kernel_time_frac": null,
    "flags": null,
    "wall_s": 0.0,
    "max_abs_diff": null,
    "mismatch_frac": null,
    "error": "static check failed: line 26: import of 'os' is not allowed; line 27: use of 'os' is not allowed"
   },
   "s": 0.0
  },
  {
   "name": "neg_file_read_runtime",
   "file": "neg_file_read.py",
   "problem": "L2_012",
   "mode": "eval",
   "static_skipped": true,
   "shows": "with the static gate OFF the sandbox hides the evaluator, other runs and home",
   "pass": true,
   "result": {
    "status": "runtime_error",
    "correct": false,
    "fast1": false,
    "fast1_raw": null,
    "speedup": 0.0,
    "speedup_lo95": null,
    "speedup_events_only": null,
    "ref_ms": null,
    "cand_ms": null,
    "own_kernels_ran": null,
    "own_kernel_time_frac": null,
    "flags": null,
    "wall_s": 43.31,
    "max_abs_diff": null,
    "mismatch_frac": null,
    "error": "FileNotFoundError: [Errno 2] No such file or directory: '<work dir>'"
   },
   "s": 43.3
  },
  {
   "name": "neg_hang",
   "file": "neg_hang.py",
   "problem": "L2_012",
   "mode": "eval",
   "static_skipped": false,
   "shows": "a hanging kernel times out (failure)",
   "pass": true,
   "result": {
    "status": "timeout",
    "correct": false,
    "fast1": false,
    "fast1_raw": null,
    "speedup": 0.0,
    "speedup_lo95": null,
    "speedup_events_only": null,
    "ref_ms": null,
    "cand_ms": null,
    "own_kernels_ran": null,
    "own_kernel_time_frac": null,
    "flags": null,
    "wall_s": 133.0,
    "max_abs_diff": null,
    "mismatch_frac": null,
    "error": "candidate correctness phase exceeded 90s"
   },
   "s": 133.0
  },
  {
   "name": "neg_native_split",
   "file": "neg_native_split.py",
   "problem": "L2_012",
   "mode": "eval",
   "static_skipped": false,
   "shows": "a banned native API assembled from split strings is caught in the generated build sources",
   "pass": true,
   "result": {
    "status": "forbidden",
    "correct": false,
    "fast1": false,
    "fast1_raw": null,
    "speedup": 0.0,
    "speedup_lo95": null,
    "speedup_events_only": null,
    "ref_ms": null,
    "cand_ms": null,
    "own_kernels_ran": null,
    "own_kernel_time_frac": null,
    "flags": null,
    "wall_s": 39.55,
    "max_abs_diff": null,
    "mismatch_frac": null,
    "error": "build sources contain banned tokens ['cudaStreamCreate']"
   },
   "s": 39.6
  },
  {
   "name": "neg_threading",
   "file": "neg_threading.py",
   "problem": "L2_012",
   "mode": "eval",
   "static_skipped": false,
   "shows": "threads rejected",
   "pass": true,
   "result": {
    "status": "forbidden",
    "correct": false,
    "fast1": false,
    "fast1_raw": null,
    "speedup": 0.0,
    "speedup_lo95": null,
    "speedup_events_only": null,
    "ref_ms": null,
    "cand_ms": null,
    "own_kernels_ran": null,
    "own_kernel_time_frac": null,
    "flags": null,
    "wall_s": 0.0,
    "max_abs_diff": null,
    "mismatch_frac": null,
    "error": "static check failed: line 7: import of 'threading' is not allowed"
   },
   "s": 0.0
  },
  {
   "name": "neg_pycapi",
   "file": "neg_pycapi.py",
   "problem": "L2_012",
   "mode": "eval",
   "static_skipped": false,
   "shows": "Python C API in native code rejected",
   "pass": true,
   "result": {
    "status": "forbidden",
    "correct": false,
    "fast1": false,
    "fast1_raw": null,
    "speedup": 0.0,
    "speedup_lo95": null,
    "speedup_events_only": null,
    "ref_ms": null,
    "cand_ms": null,
    "own_kernels_ran": null,
    "own_kernel_time_frac": null,
    "flags": null,
    "wall_s": 0.0,
    "max_abs_diff": null,
    "mismatch_frac": null,
    "error": "static check failed: native source contains banned token /Python\\.h/; native source contains banned token /\\bPyImport/; native source contains banned token /\\bPyObject/"
   },
   "s": 0.0
  }
 ],
 "at": "2026-10-09T04:06:59Z"
}