{"refused": "overlong", "choice": null, "probs": null, "seconds": null, "why": "inherited from arm 2's own refusal rows so the denominator matches (GAPS-CARD.md, arm 12)", "id": "on-01", "task": "ka", "slug": "three-new-voices-at-the-narrators-chair", "label": "yes", "kind": "on-page", "state_sha256": "d944ffba7d7a97d70ffb305031ae8069d7e1bf47491eb08abdaea5d81af2ffe1", "prompt_sha256_matches_kev9b": null, "card": "RTX 3090 (fp 33a0b4acb8e8)", "box": "benchbox", "os": {"pretty_name": "Ubuntu 26.04 LTS", "id": "ubuntu", "version_id": "26.04", "kernel": "7.0.0-31-generic", "arch": "x86_64", "source": "/etc/os-release"}, "dry": false, "runtime": "lev serve 0.1.1 (git 0ea955f5aee0) over HTTP on 127.0.0.1, one question per request", "lev_revision": "f8ef71157ec06a7d3b6435bc0756f9d735c33748", "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "lev_commit": "0ea955f5aee0d56d906121d1779af20fa8b922ed", "torch": "2.14.0+cu130", "torch_cuda": "13.0", "transformers": "5.17.0", "peft": "0.21.0", "dtype": "bfloat16", "kernels": "reference (no flash-linear-attention)", "arm": "lev-4b", "model": "interfaze-ai/lev@f8ef7115 on Qwen/Qwen3.5-4B@851bf6e8", "quant": "bf16 backbone + LoRA r=32 (PEFT, unmerged) + mode-B head fp32; calibrated as served; order_average on", "card_uuid": "GPU-fp:33a0b4acb8e8", "num_ctx": 0, "power_cap_w": 300, "seed": 20260921, "stamp": "2026-09-28T13:13:27Z"} {"refused": "overlong", "choice": null, "probs": null, "seconds": null, "why": "inherited from arm 2's own refusal rows so the denominator matches (GAPS-CARD.md, arm 12)", "id": "on-02", "task": "ka", "slug": "two-new-frontier-models-at-the-rules-desk", "label": "yes", "kind": "on-page", "state_sha256": "b5c4d7119044bfb86fe877d20dd455d38788622a8cb8fd3da273cf881e554504", "prompt_sha256_matches_kev9b": null, "card": "RTX 3090 (fp 33a0b4acb8e8)", "box": "benchbox", "os": {"pretty_name": "Ubuntu 26.04 LTS", "id": "ubuntu", "version_id": "26.04", "kernel": "7.0.0-31-generic", "arch": "x86_64", "source": "/etc/os-release"}, "dry": false, "runtime": "lev serve 0.1.1 (git 0ea955f5aee0) over HTTP on 127.0.0.1, one question per request", "lev_revision": "f8ef71157ec06a7d3b6435bc0756f9d735c33748", "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "lev_commit": "0ea955f5aee0d56d906121d1779af20fa8b922ed", "torch": "2.14.0+cu130", "torch_cuda": "13.0", "transformers": "5.17.0", "peft": "0.21.0", "dtype": "bfloat16", "kernels": "reference (no flash-linear-attention)", "arm": "lev-4b", "model": "interfaze-ai/lev@f8ef7115 on Qwen/Qwen3.5-4B@851bf6e8", "quant": "bf16 backbone + LoRA r=32 (PEFT, unmerged) + mode-B head fp32; calibrated as served; order_average on", "card_uuid": "GPU-fp:33a0b4acb8e8", "num_ctx": 0, "power_cap_w": 300, "seed": 20260921, "stamp": "2026-09-28T13:13:27Z"} {"mode": "kev-noul", "prompt_sha256": "f8ebce04dd689673c8f6656b0c067eeaf2f9c2930c89dd7a8d480047c0c9db60", "n_options": 2, "kev_response": {"model": "Qwen/Qwen3.5-4B", "answers": {"q": {"type": "noul", "noul": 0.9383618202864219, "probabilities": {"0": 0.04305399416549228, "1": 0.00387207990127903, "2": 0.004261563416412966, "3": 0.004522513488823367, "4": 0.006668983689635179, "5": 0.007858005398059394, "6": 0.005774731605396942, "7": 0.011587562969625901, "8": 0.9124005653652749}, "confidence": 0.8139878738641271}}, "usage": {"input_tokens": 2911, "output_tokens": 0, "cached_input_tokens": 2823}}, "kev_noul": 0.9383618202864219, "choice": "yes", "probs": {"yes": 0.9383618202864219, "no": 0.0616381797135781}, "confidence": 0.8139878738641271, "seconds": 0.8238008159996753, "prompt_tokens": 2911, "eval_tokens": 0, "server_latency_ms": null, "prompt_eval_ms": null, "total_ms": 823.8008159996753, "done_reason": null, "id": "on-03", "task": "ka", "slug": "six-worlds-painted-large", "label": "yes", "kind": "on-page", "state_sha256": "4902422ba42eec05800d9a72fd69c58559d0cb098878daec40505c90f50184d4", "prompt_sha256_matches_kev9b": true, "card": "RTX 3090 (fp 33a0b4acb8e8)", "box": "benchbox", "os": {"pretty_name": "Ubuntu 26.04 LTS", "id": "ubuntu", "version_id": "26.04", "kernel": "7.0.0-31-generic", "arch": "x86_64", "source": "/etc/os-release"}, "dry": false, "runtime": "lev serve 0.1.1 (git 0ea955f5aee0) over HTTP on 127.0.0.1, one question per request", "lev_revision": "f8ef71157ec06a7d3b6435bc0756f9d735c33748", "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "lev_commit": "0ea955f5aee0d56d906121d1779af20fa8b922ed", "torch": "2.14.0+cu130", "torch_cuda": "13.0", "transformers": "5.17.0", "peft": "0.21.0", "dtype": "bfloat16", "kernels": "reference (no flash-linear-attention)", "arm": "lev-4b", "model": "interfaze-ai/lev@f8ef7115 on Qwen/Qwen3.5-4B@851bf6e8", "quant": "bf16 backbone + LoRA r=32 (PEFT, unmerged) + mode-B head fp32; calibrated as served; order_average on", "card_uuid": "GPU-fp:33a0b4acb8e8", "num_ctx": 0, "power_cap_w": 300, "seed": 20260921, "stamp": "2026-09-28T13:13:28Z"} {"mode": "kev-noul", "prompt_sha256": "4b35dd2369d7faeb0ff675f5ae0784719f669b50156491251f05bd9838f208f9", "n_options": 2, "kev_response": {"model": "Qwen/Qwen3.5-4B", "answers": {"q": {"type": "noul", "noul": 0.9619287914091588, "probabilities": {"0": 0.027614197453345923, "1": 0.0021388112832997526, "2": 0.0025027990327006395, "3": 0.00250935451865573, "4": 0.004164333905969256, "5": 0.004136540424374668, "6": 0.002883726783010185, "7": 0.006286432864359933, "8": 0.9477638037342838}, "confidence": 0.8865079690471236}}, "usage": {"input_tokens": 4652, "output_tokens": 0, "cached_input_tokens": 4561}}, "kev_noul": 0.9619287914091588, "choice": "yes", "probs": {"yes": 0.9619287914091588, "no": 0.03807120859084123}, "confidence": 0.8865079690471236, "seconds": 1.3069156880001174, "prompt_tokens": 4652, "eval_tokens": 0, "server_latency_ms": null, "prompt_eval_ms": null, "total_ms": 1306.9156880001174, "done_reason": null, "id": "on-04", "task": "ka", "slug": "the-beat-lab-asked-and-answered", "label": "yes", "kind": "on-page", "state_sha256": "423a54ae5c1707a69cc67454a7c1fb261af05e3661a12dd44a3ac170ae6fdbea", "prompt_sha256_matches_kev9b": true, "card": "RTX 3090 (fp 33a0b4acb8e8)", "box": "benchbox", "os": {"pretty_name": "Ubuntu 26.04 LTS", "id": "ubuntu", "version_id": "26.04", "kernel": "7.0.0-31-generic", "arch": "x86_64", "source": "/etc/os-release"}, "dry": false, "runtime": "lev serve 0.1.1 (git 0ea955f5aee0) over HTTP on 127.0.0.1, one question per request", "lev_revision": "f8ef71157ec06a7d3b6435bc0756f9d735c33748", "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "lev_commit": "0ea955f5aee0d56d906121d1779af20fa8b922ed", "torch": "2.14.0+cu130", "torch_cuda": "13.0", "transformers": "5.17.0", "peft": "0.21.0", "dtype": "bfloat16", "kernels": "reference (no flash-linear-attention)", "arm": "lev-4b", "model": "interfaze-ai/lev@f8ef7115 on Qwen/Qwen3.5-4B@851bf6e8", "quant": "bf16 backbone + LoRA r=32 (PEFT, unmerged) + mode-B head fp32; calibrated as served; order_average on", "card_uuid": "GPU-fp:33a0b4acb8e8", "num_ctx": 0, "power_cap_w": 300, "seed": 20260921, "stamp": "2026-09-28T13:13:29Z"} {"mode": "kev-noul", "prompt_sha256": "544f0ee64df7e0df092e1d7b112006bcef1ba90b07629cd639c6c83585ded309", "n_options": 2, "kev_response": {"model": "Qwen/Qwen3.5-4B", "answers": {"q": {"type": "noul", "noul": 0.9646158559091491, "probabilities": {"0": 0.025931303866806796, "1": 0.0020457951437837275, "2": 0.0020995764745985568, "3": 0.002216978130958912, "4": 0.0039040053608220527, "5": 0.003651133257232149, "6": 0.002543735034957403, "7": 0.0059629149985825376, "8": 0.951644557732258}, "confidence": 0.8946818922789566}}, "usage": {"input_tokens": 12757, "output_tokens": 0, "cached_input_tokens": 12665}}, "kev_noul": 0.9646158559091491, "choice": "yes", "probs": {"yes": 0.9646158559091491, "no": 0.03538414409085089}, "confidence": 0.8946818922789566, "seconds": 3.615360237999994, "prompt_tokens": 12757, "eval_tokens": 0, "server_latency_ms": null, "prompt_eval_ms": null, "total_ms": 3615.360237999994, "done_reason": null, "id": "on-05", "task": "ka", "slug": "ten-minutes-with-living-artists", "label": "yes", "kind": "on-page", "state_sha256": "3d3d19395679e12b9871c93853d3d888cfff4352b0fff0d5a432a99980d72e14", "prompt_sha256_matches_kev9b": null, "kev9b_refused": "kev-422", "card": "RTX 3090 (fp 33a0b4acb8e8)", "box": "benchbox", "os": {"pretty_name": "Ubuntu 26.04 LTS", "id": "ubuntu", "version_id": "26.04", "kernel": "7.0.0-31-generic", "arch": "x86_64", "source": "/etc/os-release"}, "dry": false, "runtime": "lev serve 0.1.1 (git 0ea955f5aee0) over HTTP on 127.0.0.1, one question per request", "lev_revision": "f8ef71157ec06a7d3b6435bc0756f9d735c33748", "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "lev_commit": "0ea955f5aee0d56d906121d1779af20fa8b922ed", "torch": "2.14.0+cu130", "torch_cuda": "13.0", "transformers": "5.17.0", "peft": "0.21.0", "dtype": "bfloat16", "kernels": "reference (no flash-linear-attention)", "arm": "lev-4b", "model": "interfaze-ai/lev@f8ef7115 on Qwen/Qwen3.5-4B@851bf6e8", "quant": "bf16 backbone + LoRA r=32 (PEFT, unmerged) + mode-B head fp32; calibrated as served; order_average on", "card_uuid": "GPU-fp:33a0b4acb8e8", "num_ctx": 0, "power_cap_w": 300, "seed": 20260921, "stamp": "2026-09-28T13:13:33Z"} {"mode": "kev-noul", "prompt_sha256": "f7a46135ce9830b095ff8a908ea65af2ec1853bcda47ecc7e72e53a9ce0dfb2c", "n_options": 2, "kev_response": {"model": "Qwen/Qwen3.5-4B", "answers": {"q": {"type": "noul", "noul": 0.9508028817092367, "probabilities": {"0": 0.035012764221342814, "1": 0.003040106747745794, "2": 0.0034002551592719307, "3": 0.0034245345125857194, "4": 0.005608076995228991, "5": 0.005262418491530534, "6": 0.003906315094123243, "7": 0.00863768815882973, "8": 0.9317078406193413}, "confidence": 0.8531728051162343}}, "usage": {"input_tokens": 14323, "output_tokens": 0, "cached_input_tokens": 14232}}, "kev_noul": 0.9508028817092367, "choice": "yes", "probs": {"yes": 0.9508028817092367, "no": 0.04919711829076334}, "confidence": 0.8531728051162343, "seconds": 4.336818987000697, "prompt_tokens": 14323, "eval_tokens": 0, "server_latency_ms": null, "prompt_eval_ms": null, "total_ms": 4336.818987000697, "done_reason": null, "id": "on-06", "task": "ka", "slug": "listen-for-yourself", "label": "yes", "kind": "on-page", "state_sha256": "ff4ba7e7329efa416398cff233e4d8fbc43838da73f017c9081eeeaaf26349cc", "prompt_sha256_matches_kev9b": null, "kev9b_refused": "kev-422", "card": "RTX 3090 (fp 33a0b4acb8e8)", "box": "benchbox", "os": {"pretty_name": "Ubuntu 26.04 LTS", "id": "ubuntu", "version_id": "26.04", "kernel": "7.0.0-31-generic", "arch": "x86_64", "source": "/etc/os-release"}, "dry": false, "runtime": "lev serve 0.1.1 (git 0ea955f5aee0) over HTTP on 127.0.0.1, one question per request", "lev_revision": "f8ef71157ec06a7d3b6435bc0756f9d735c33748", "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "lev_commit": "0ea955f5aee0d56d906121d1779af20fa8b922ed", "torch": "2.14.0+cu130", "torch_cuda": "13.0", "transformers": "5.17.0", "peft": "0.21.0", "dtype": "bfloat16", "kernels": "reference (no flash-linear-attention)", "arm": "lev-4b", "model": "interfaze-ai/lev@f8ef7115 on Qwen/Qwen3.5-4B@851bf6e8", "quant": "bf16 backbone + LoRA r=32 (PEFT, unmerged) + mode-B head fp32; calibrated as served; order_average on", "card_uuid": "GPU-fp:33a0b4acb8e8", "num_ctx": 0, "power_cap_w": 300, "seed": 20260921, "stamp": "2026-09-28T13:13:37Z"} {"mode": "kev-noul", "prompt_sha256": "af029828977102ecd6e92bc895faae7cfeb96442254aca83b40d201ab59db54a", "n_options": 2, "kev_response": {"model": "Qwen/Qwen3.5-4B", "answers": {"q": {"type": "noul", "noul": 0.9129050521173794, "probabilities": {"0": 0.05667926041781282, "1": 0.007038780165208295, "2": 0.006344818209788289, "3": 0.0073028587103078635, "4": 0.01069705947869698, "5": 0.012646525214079997, "6": 0.010414329686781727, "7": 0.01791436282114455, "8": 0.8709620052961795}, "confidence": 0.7329635092678581}}, "usage": {"input_tokens": 5460, "output_tokens": 0, "cached_input_tokens": 5374}}, "kev_noul": 0.9129050521173794, "choice": "yes", "probs": {"yes": 0.9129050521173794, "no": 0.08709494788262062}, "confidence": 0.7329635092678581, "seconds": 1.5296013039996978, "prompt_tokens": 5460, "eval_tokens": 0, "server_latency_ms": null, "prompt_eval_ms": null, "total_ms": 1529.6013039996978, "done_reason": null, "id": "on-07", "task": "ka", "slug": "half-an-hour-with-dead-composers", "label": "yes", "kind": "on-page", "state_sha256": "21d3e92b73064718cb5ca0a3d5fd514b31f0b656626e5a7e8e2baeff97735bfa", "prompt_sha256_matches_kev9b": true, "card": "RTX 3090 (fp 33a0b4acb8e8)", "box": "benchbox", "os": {"pretty_name": "Ubuntu 26.04 LTS", "id": "ubuntu", "version_id": "26.04", "kernel": "7.0.0-31-generic", "arch": "x86_64", "source": "/etc/os-release"}, "dry": false, "runtime": "lev serve 0.1.1 (git 0ea955f5aee0) over HTTP on 127.0.0.1, one question per request", "lev_revision": "f8ef71157ec06a7d3b6435bc0756f9d735c33748", "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "lev_commit": "0ea955f5aee0d56d906121d1779af20fa8b922ed", "torch": "2.14.0+cu130", "torch_cuda": "13.0", "transformers": "5.17.0", "peft": "0.21.0", "dtype": "bfloat16", "kernels": "reference (no flash-linear-attention)", "arm": "lev-4b", "model": "interfaze-ai/lev@f8ef7115 on Qwen/Qwen3.5-4B@851bf6e8", "quant": "bf16 backbone + LoRA r=32 (PEFT, unmerged) + mode-B head fp32; calibrated as served; order_average on", "card_uuid": "GPU-fp:33a0b4acb8e8", "num_ctx": 0, "power_cap_w": 300, "seed": 20260921, "stamp": "2026-09-28T13:13:39Z"} {"mode": "kev-noul", "prompt_sha256": "7159206ba651b0038e881478a3249949339ca1db0ab69b0e0fa8c77920eb0595", "n_options": 2, "kev_response": {"model": "Qwen/Qwen3.5-4B", "answers": {"q": {"type": "noul", "noul": 0.9563574694609056, "probabilities": {"0": 0.03173506360264049, "1": 0.0025438078807218686, "2": 0.00270126924643666, "3": 0.002827355889988017, "4": 0.004737945752161003, "5": 0.00470632388632789, "6": 0.003283690253868185, "7": 0.007470550222654663, "8": 0.9399939932652013}, "confidence": 0.8703198693423024}}, "usage": {"input_tokens": 3572, "output_tokens": 0, "cached_input_tokens": 3479}}, "kev_noul": 0.9563574694609056, "choice": "yes", "probs": {"yes": 0.9563574694609056, "no": 0.04364253053909439}, "confidence": 0.8703198693423024, "seconds": 1.0024318850000782, "prompt_tokens": 3572, "eval_tokens": 0, "server_latency_ms": null, "prompt_eval_ms": null, "total_ms": 1002.4318850000782, "done_reason": null, "id": "on-08", "task": "ka", "slug": "chopin-in-five-minutes", "label": "yes", "kind": "on-page", "state_sha256": "ce76bf2f182f387a020db77e9380690ce1522227e2f85e1345b2d33cba49c186", "prompt_sha256_matches_kev9b": true, "card": "RTX 3090 (fp 33a0b4acb8e8)", "box": "benchbox", "os": {"pretty_name": "Ubuntu 26.04 LTS", "id": "ubuntu", "version_id": "26.04", "kernel": "7.0.0-31-generic", "arch": "x86_64", "source": "/etc/os-release"}, "dry": false, "runtime": "lev serve 0.1.1 (git 0ea955f5aee0) over HTTP on 127.0.0.1, one question per request", "lev_revision": "f8ef71157ec06a7d3b6435bc0756f9d735c33748", "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "lev_commit": "0ea955f5aee0d56d906121d1779af20fa8b922ed", "torch": "2.14.0+cu130", "torch_cuda": "13.0", "transformers": "5.17.0", "peft": "0.21.0", "dtype": "bfloat16", "kernels": "reference (no flash-linear-attention)", "arm": "lev-4b", "model": "interfaze-ai/lev@f8ef7115 on Qwen/Qwen3.5-4B@851bf6e8", "quant": "bf16 backbone + LoRA r=32 (PEFT, unmerged) + mode-B head fp32; calibrated as served; order_average on", "card_uuid": "GPU-fp:33a0b4acb8e8", "num_ctx": 0, "power_cap_w": 300, "seed": 20260921, "stamp": "2026-09-28T13:13:40Z"} {"mode": "kev-noul", "prompt_sha256": "adf96d957a673f6825ed15571fe62c5b1c6c91a7d9bd2cbf6792820a261295ac", "n_options": 2, "kev_response": {"model": "Qwen/Qwen3.5-4B", "answers": {"q": {"type": "noul", "noul": 0.9416021569950817, "probabilities": {"0": 0.04106433428079148, "1": 0.003960506587515936, "2": 0.0037949888572167373, "3": 0.004199505179687872, "4": 0.006468173706315688, "5": 0.006733351825804886, "6": 0.00545281464176801, "7": 0.010198685052450477, "8": 0.9181276398684489}, "confidence": 0.8255274486615606}}, "usage": {"input_tokens": 6255, "output_tokens": 0, "cached_input_tokens": 6159}}, "kev_noul": 0.9416021569950817, "choice": "yes", "probs": {"yes": 0.9416021569950817, "no": 0.05839784300491835}, "confidence": 0.8255274486615606, "seconds": 1.7306802479997714, "prompt_tokens": 6255, "eval_tokens": 0, "server_latency_ms": null, "prompt_eval_ms": null, "total_ms": 1730.6802479997714, "done_reason": null, "id": "on-09", "task": "ka", "slug": "the-same-sixteen", "label": "yes", "kind": "on-page", "state_sha256": "830caf9405ece8c6ff598d3adfbdaf8da136fd260537d81f576a5059e5a17953", "prompt_sha256_matches_kev9b": true, "card": "RTX 3090 (fp 33a0b4acb8e8)", "box": "benchbox", "os": {"pretty_name": "Ubuntu 26.04 LTS", "id": "ubuntu", "version_id": "26.04", "kernel": "7.0.0-31-generic", "arch": "x86_64", "source": "/etc/os-release"}, "dry": false, "runtime": "lev serve 0.1.1 (git 0ea955f5aee0) over HTTP on 127.0.0.1, one question per request", "lev_revision": "f8ef71157ec06a7d3b6435bc0756f9d735c33748", "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "lev_commit": "0ea955f5aee0d56d906121d1779af20fa8b922ed", "torch": "2.14.0+cu130", "torch_cuda": "13.0", "transformers": "5.17.0", "peft": "0.21.0", "dtype": "bfloat16", "kernels": "reference (no flash-linear-attention)", "arm": "lev-4b", "model": "interfaze-ai/lev@f8ef7115 on Qwen/Qwen3.5-4B@851bf6e8", "quant": "bf16 backbone + LoRA r=32 (PEFT, unmerged) + mode-B head fp32; calibrated as served; order_average on", "card_uuid": "GPU-fp:33a0b4acb8e8", "num_ctx": 0, "power_cap_w": 300, "seed": 20260921, "stamp": "2026-09-28T13:13:41Z"} {"mode": "kev-noul", "prompt_sha256": "d9c1aa5ce8328386bf5b822eab474fe8debc8954f00773d30f75876f46b67d9e", "n_options": 2, "kev_response": {"model": "Qwen/Qwen3.5-4B", "answers": {"q": {"type": "noul", "noul": 0.9634496635734678, "probabilities": {"0": 0.02659517445634527, "1": 0.0023208468442039052, "2": 0.0022168841290774576, "3": 0.002312121108666164, "4": 0.0039046568578030984, "5": 0.003852710084565211, "6": 0.0027703454467339866, "7": 0.005816008955896747, "8": 0.9502112521167082}, "confidence": 0.8916579597892884}}, "usage": {"input_tokens": 10710, "output_tokens": 0, "cached_input_tokens": 10621}}, "kev_noul": 0.9634496635734678, "choice": "yes", "probs": {"yes": 0.9634496635734678, "no": 0.03655033642653216}, "confidence": 0.8916579597892884, "seconds": 3.0237132499996733, "prompt_tokens": 10710, "eval_tokens": 0, "server_latency_ms": null, "prompt_eval_ms": null, "total_ms": 3023.7132499996733, "done_reason": null, "id": "on-10", "task": "ka", "slug": "two-hours-on-battery", "label": "yes", "kind": "on-page", "state_sha256": "c4c4db33a200bbafecb6bc46874d79161edbca9f5ee4e5451aac9225077b64f1", "prompt_sha256_matches_kev9b": null, "kev9b_refused": "kev-422", "card": "RTX 3090 (fp 33a0b4acb8e8)", "box": "benchbox", "os": {"pretty_name": "Ubuntu 26.04 LTS", "id": "ubuntu", "version_id": "26.04", "kernel": "7.0.0-31-generic", "arch": "x86_64", "source": "/etc/os-release"}, "dry": false, "runtime": "lev serve 0.1.1 (git 0ea955f5aee0) over HTTP on 127.0.0.1, one question per request", "lev_revision": "f8ef71157ec06a7d3b6435bc0756f9d735c33748", "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "lev_commit": "0ea955f5aee0d56d906121d1779af20fa8b922ed", "torch": "2.14.0+cu130", "torch_cuda": "13.0", "transformers": "5.17.0", "peft": "0.21.0", "dtype": "bfloat16", "kernels": "reference (no flash-linear-attention)", "arm": "lev-4b", "model": "interfaze-ai/lev@f8ef7115 on Qwen/Qwen3.5-4B@851bf6e8", "quant": "bf16 backbone + LoRA r=32 (PEFT, unmerged) + mode-B head fp32; calibrated as served; order_average on", "card_uuid": "GPU-fp:33a0b4acb8e8", "num_ctx": 0, "power_cap_w": 300, "seed": 20260921, "stamp": "2026-09-28T13:13:45Z"} {"mode": "kev-noul", "prompt_sha256": "23af0206dfe7b32d92912604f44c2146cd59ed7ff0bd53a231c1b935feb4f573", "n_options": 2, "kev_response": {"model": "Qwen/Qwen3.5-4B", "answers": {"q": {"type": "noul", "noul": 0.9484190009724065, "probabilities": {"0": 0.03678873232477324, "1": 0.003248238709260827, "2": 0.0033574343643540112, "3": 0.0034717535797212417, "4": 0.005437548275045036, "5": 0.005833644813647198, "6": 0.004885069776144952, "7": 0.00907582147959477, "8": 0.9279017566774588}, "confidence": 0.8453786389183247}}, "usage": {"input_tokens": 8508, "output_tokens": 0, "cached_input_tokens": 8417}}, "kev_noul": 0.9484190009724065, "choice": "yes", "probs": {"yes": 0.9484190009724065, "no": 0.0515809990275935}, "confidence": 0.8453786389183247, "seconds": 2.395436793999579, "prompt_tokens": 8508, "eval_tokens": 0, "server_latency_ms": null, "prompt_eval_ms": null, "total_ms": 2395.436793999579, "done_reason": null, "id": "on-11", "task": "ka", "slug": "the-instrument-travels", "label": "yes", "kind": "on-page", "state_sha256": "f316ed88d19f7f94e1870200edec052a28e370f3c5c57589b7c66a8182758693", "prompt_sha256_matches_kev9b": null, "kev9b_refused": "kev-422", "card": "RTX 3090 (fp 33a0b4acb8e8)", "box": "benchbox", "os": {"pretty_name": "Ubuntu 26.04 LTS", "id": "ubuntu", "version_id": "26.04", "kernel": "7.0.0-31-generic", "arch": "x86_64", "source": "/etc/os-release"}, "dry": false, "runtime": "lev serve 0.1.1 (git 0ea955f5aee0) over HTTP on 127.0.0.1, one question per request", "lev_revision": "f8ef71157ec06a7d3b6435bc0756f9d735c33748", "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "lev_commit": "0ea955f5aee0d56d906121d1779af20fa8b922ed", "torch": "2.14.0+cu130", "torch_cuda": "13.0", "transformers": "5.17.0", "peft": "0.21.0", "dtype": "bfloat16", "kernels": "reference (no flash-linear-attention)", "arm": "lev-4b", "model": "interfaze-ai/lev@f8ef7115 on Qwen/Qwen3.5-4B@851bf6e8", "quant": "bf16 backbone + LoRA r=32 (PEFT, unmerged) + mode-B head fp32; calibrated as served; order_average on", "card_uuid": "GPU-fp:33a0b4acb8e8", "num_ctx": 0, "power_cap_w": 300, "seed": 20260921, "stamp": "2026-09-28T13:13:47Z"} {"mode": "kev-noul", "prompt_sha256": "cc21c98043fdd32ece00bb9deeaef0bb80eda78cd24c4ae19e7df71d59b8600d", "n_options": 2, "kev_response": {"model": "Qwen/Qwen3.5-4B", "answers": {"q": {"type": "noul", "noul": 0.860969028477745, "probabilities": {"0": 0.08056567168877883, "1": 0.01024241129928328, "2": 0.012354746296875522, "3": 0.012690154811368275, "4": 0.019742988011022304, "5": 0.024791005101432225, "6": 0.02990375698876438, "7": 0.04529378640881677, "8": 0.7644154793936584}, "confidence": 0.5445894032251235}}, "usage": {"input_tokens": 1821, "output_tokens": 0, "cached_input_tokens": 1738}}, "kev_noul": 0.860969028477745, "choice": "yes", "probs": {"yes": 0.860969028477745, "no": 0.13903097152225496}, "confidence": 0.5445894032251235, "seconds": 0.528373678999742, "prompt_tokens": 1821, "eval_tokens": 0, "server_latency_ms": null, "prompt_eval_ms": null, "total_ms": 528.373678999742, "done_reason": null, "id": "on-12", "task": "ka", "slug": "nine-worlds-one-dog", "label": "yes", "kind": "on-page", "state_sha256": "84c842a71a59b037055f5448244f732c94578d47c68a907c19c9a4a0324a0095", "prompt_sha256_matches_kev9b": true, "card": "RTX 3090 (fp 33a0b4acb8e8)", "box": "benchbox", "os": {"pretty_name": "Ubuntu 26.04 LTS", "id": "ubuntu", "version_id": "26.04", "kernel": "7.0.0-31-generic", "arch": "x86_64", "source": "/etc/os-release"}, "dry": false, "runtime": "lev serve 0.1.1 (git 0ea955f5aee0) over HTTP on 127.0.0.1, one question per request", "lev_revision": "f8ef71157ec06a7d3b6435bc0756f9d735c33748", "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "lev_commit": "0ea955f5aee0d56d906121d1779af20fa8b922ed", "torch": "2.14.0+cu130", "torch_cuda": "13.0", "transformers": "5.17.0", "peft": "0.21.0", "dtype": "bfloat16", "kernels": "reference (no flash-linear-attention)", "arm": "lev-4b", "model": "interfaze-ai/lev@f8ef7115 on Qwen/Qwen3.5-4B@851bf6e8", "quant": "bf16 backbone + LoRA r=32 (PEFT, unmerged) + mode-B head fp32; calibrated as served; order_average on", "card_uuid": "GPU-fp:33a0b4acb8e8", "num_ctx": 0, "power_cap_w": 300, "seed": 20260921, "stamp": "2026-09-28T13:13:47Z"} {"mode": "kev-noul", "prompt_sha256": "aadaa76601faa41b91b4e9a4253f0d299bde20b1be555f0c0f12c1ca5486e92a", "n_options": 2, "kev_response": {"model": "Qwen/Qwen3.5-4B", "answers": {"q": {"type": "noul", "noul": 0.9609599808243319, "probabilities": {"0": 0.02831933318155048, "1": 0.0022804859218224804, "2": 0.0024749295771313955, "3": 0.002570471305441983, "4": 0.004213855101933819, "5": 0.004047901769828349, "6": 0.0032835713708251344, "7": 0.00653388405131558, "8": 0.9462755677201509}, "confidence": 0.8833881064109799}}, "usage": {"input_tokens": 6963, "output_tokens": 0, "cached_input_tokens": 6867}}, "kev_noul": 0.9609599808243319, "choice": "yes", "probs": {"yes": 0.9609599808243319, "no": 0.03904001917566813}, "confidence": 0.8833881064109799, "seconds": 1.9479342490003546, "prompt_tokens": 6963, "eval_tokens": 0, "server_latency_ms": null, "prompt_eval_ms": null, "total_ms": 1947.9342490003546, "done_reason": null, "id": "on-13", "task": "ka", "slug": "what-150-watts-buys", "label": "yes", "kind": "on-page", "state_sha256": "0fd39e7ea02081243919348e7a76a2acba2433694142679ddfd8fbaa29a3d00d", "prompt_sha256_matches_kev9b": true, "card": "RTX 3090 (fp 33a0b4acb8e8)", "box": "benchbox", "os": {"pretty_name": "Ubuntu 26.04 LTS", "id": "ubuntu", "version_id": "26.04", "kernel": "7.0.0-31-generic", "arch": "x86_64", "source": "/etc/os-release"}, "dry": false, "runtime": "lev serve 0.1.1 (git 0ea955f5aee0) over HTTP on 127.0.0.1, one question per request", "lev_revision": "f8ef71157ec06a7d3b6435bc0756f9d735c33748", "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "lev_commit": "0ea955f5aee0d56d906121d1779af20fa8b922ed", "torch": "2.14.0+cu130", "torch_cuda": "13.0", "transformers": "5.17.0", "peft": "0.21.0", "dtype": "bfloat16", "kernels": "reference (no flash-linear-attention)", "arm": "lev-4b", "model": "interfaze-ai/lev@f8ef7115 on Qwen/Qwen3.5-4B@851bf6e8", "quant": "bf16 backbone + LoRA r=32 (PEFT, unmerged) + mode-B head fp32; calibrated as served; order_average on", "card_uuid": "GPU-fp:33a0b4acb8e8", "num_ctx": 0, "power_cap_w": 300, "seed": 20260921, "stamp": "2026-09-28T13:13:49Z"} {"refused": "overlong", "choice": null, "probs": null, "seconds": null, "why": "inherited from arm 2's own refusal rows so the denominator matches (GAPS-CARD.md, arm 12)", "id": "on-14", "task": "ka", "slug": "how-the-beat-lab-works", "label": "yes", "kind": "on-page", "state_sha256": "fdb6d4bf89ebedc44a2d887309bf63ba59598c898a6eca00760115d49bc2133d", "prompt_sha256_matches_kev9b": null, "card": "RTX 3090 (fp 33a0b4acb8e8)", "box": "benchbox", "os": {"pretty_name": "Ubuntu 26.04 LTS", "id": "ubuntu", "version_id": "26.04", "kernel": "7.0.0-31-generic", "arch": "x86_64", "source": "/etc/os-release"}, "dry": false, "runtime": "lev serve 0.1.1 (git 0ea955f5aee0) over HTTP on 127.0.0.1, one question per request", "lev_revision": "f8ef71157ec06a7d3b6435bc0756f9d735c33748", "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "lev_commit": "0ea955f5aee0d56d906121d1779af20fa8b922ed", "torch": "2.14.0+cu130", "torch_cuda": "13.0", "transformers": "5.17.0", "peft": "0.21.0", "dtype": "bfloat16", "kernels": "reference (no flash-linear-attention)", "arm": "lev-4b", "model": "interfaze-ai/lev@f8ef7115 on Qwen/Qwen3.5-4B@851bf6e8", "quant": "bf16 backbone + LoRA r=32 (PEFT, unmerged) + mode-B head fp32; calibrated as served; order_average on", "card_uuid": "GPU-fp:33a0b4acb8e8", "num_ctx": 0, "power_cap_w": 300, "seed": 20260921, "stamp": "2026-09-28T13:13:49Z"} {"mode": "kev-noul", "prompt_sha256": "2bcf96723063667f7a5e82dc4bea15e8c566c6a67130c7d3cbc4e458657eab24", "n_options": 2, "kev_response": {"model": "Qwen/Qwen3.5-4B", "answers": {"q": {"type": "noul", "noul": 0.6592081092842113, "probabilities": {"0": 0.10576980255754394, "1": 0.03721099123732276, "2": 0.048640958550172475, "3": 0.04996147062373428, "4": 0.07077263490804983, "5": 0.10864125709829728, "6": 0.19067319436464394, "7": 0.08768596252861327, "8": 0.30064372813162216}, "confidence": 0.06476211474837135}}, "usage": {"input_tokens": 5228, "output_tokens": 0, "cached_input_tokens": 5143}}, "kev_noul": 0.6592081092842113, "choice": "yes", "probs": {"yes": 0.6592081092842113, "no": 0.34079189071578875}, "confidence": 0.06476211474837135, "seconds": 1.4402002199994968, "prompt_tokens": 5228, "eval_tokens": 0, "server_latency_ms": null, "prompt_eval_ms": null, "total_ms": 1440.2002199994968, "done_reason": null, "id": "on-15", "task": "ka", "slug": "the-typist-and-the-developer", "label": "yes", "kind": "on-page", "state_sha256": "2d33747167f415355e5637f99962534c215404a547456ccde9c80e2d7591cf5d", "prompt_sha256_matches_kev9b": true, "card": "RTX 3090 (fp 33a0b4acb8e8)", "box": "benchbox", "os": {"pretty_name": "Ubuntu 26.04 LTS", "id": "ubuntu", "version_id": "26.04", "kernel": "7.0.0-31-generic", "arch": "x86_64", "source": "/etc/os-release"}, "dry": false, "runtime": "lev serve 0.1.1 (git 0ea955f5aee0) over HTTP on 127.0.0.1, one question per request", "lev_revision": "f8ef71157ec06a7d3b6435bc0756f9d735c33748", "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "lev_commit": "0ea955f5aee0d56d906121d1779af20fa8b922ed", "torch": "2.14.0+cu130", "torch_cuda": "13.0", "transformers": "5.17.0", "peft": "0.21.0", "dtype": "bfloat16", "kernels": "reference (no flash-linear-attention)", "arm": "lev-4b", "model": "interfaze-ai/lev@f8ef7115 on Qwen/Qwen3.5-4B@851bf6e8", "quant": "bf16 backbone + LoRA r=32 (PEFT, unmerged) + mode-B head fp32; calibrated as served; order_average on", "card_uuid": "GPU-fp:33a0b4acb8e8", "num_ctx": 0, "power_cap_w": 300, "seed": 20260921, "stamp": "2026-09-28T13:13:51Z"} {"mode": "kev-noul", "prompt_sha256": "74c848205284da7c2e3107feda1f2348604771c9f8c84dc7e02efae8787e5f79", "n_options": 2, "kev_response": {"model": "Qwen/Qwen3.5-4B", "answers": {"q": {"type": "noul", "noul": 0.9401785864428929, "probabilities": {"0": 0.04325803024788148, "1": 0.00372004790777749, "2": 0.0035378261831957034, "3": 0.003994375823408094, "4": 0.005841082881219917, "5": 0.006655867618139461, "6": 0.005667690730029249, "7": 0.01060057906379145, "8": 0.9167244995445571}, "confidence": 0.8228353252066491}}, "usage": {"input_tokens": 4449, "output_tokens": 0, "cached_input_tokens": 4354}}, "kev_noul": 0.9401785864428929, "choice": "yes", "probs": {"yes": 0.9401785864428929, "no": 0.05982141355710713}, "confidence": 0.8228353252066491, "seconds": 1.2613932519998343, "prompt_tokens": 4449, "eval_tokens": 0, "server_latency_ms": null, "prompt_eval_ms": null, "total_ms": 1261.3932519998343, "done_reason": null, "id": "on-16", "task": "ka", "slug": "a-rig-your-friend-already-owns", "label": "yes", "kind": "on-page", "state_sha256": "9d22ebe41036f74e0c8501e4631982f4d610eebb419288b4535c8a5e9f9b65d4", "prompt_sha256_matches_kev9b": true, "card": "RTX 3090 (fp 33a0b4acb8e8)", "box": "benchbox", "os": {"pretty_name": "Ubuntu 26.04 LTS", "id": "ubuntu", "version_id": "26.04", "kernel": "7.0.0-31-generic", "arch": "x86_64", "source": "/etc/os-release"}, "dry": false, "runtime": "lev serve 0.1.1 (git 0ea955f5aee0) over HTTP on 127.0.0.1, one question per request", "lev_revision": "f8ef71157ec06a7d3b6435bc0756f9d735c33748", "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "lev_commit": "0ea955f5aee0d56d906121d1779af20fa8b922ed", "torch": "2.14.0+cu130", "torch_cuda": "13.0", "transformers": "5.17.0", "peft": "0.21.0", "dtype": "bfloat16", "kernels": "reference (no flash-linear-attention)", "arm": "lev-4b", "model": "interfaze-ai/lev@f8ef7115 on Qwen/Qwen3.5-4B@851bf6e8", "quant": "bf16 backbone + LoRA r=32 (PEFT, unmerged) + mode-B head fp32; calibrated as served; order_average on", "card_uuid": "GPU-fp:33a0b4acb8e8", "num_ctx": 0, "power_cap_w": 300, "seed": 20260921, "stamp": "2026-09-28T13:13:52Z"} {"mode": "kev-noul", "prompt_sha256": "f7e2a4ea6ecd51b7f9f233118addb1fdd88d344a05bdcd77161f004a73b3bab1", "n_options": 2, "kev_response": {"model": "Qwen/Qwen3.5-4B", "answers": {"q": {"type": "noul", "noul": 0.9414339317673652, "probabilities": {"0": 0.043392033247098276, "1": 0.003178886551542986, "2": 0.003278882000734984, "3": 0.0035458161880174313, "4": 0.005685247876115304, "5": 0.006099388016288303, "6": 0.00544309150854987, "7": 0.009812362508567892, "8": 0.9195642921030849}, "confidence": 0.8286739967579979}}, "usage": {"input_tokens": 9962, "output_tokens": 0, "cached_input_tokens": 9866}}, "kev_noul": 0.9414339317673652, "choice": "yes", "probs": {"yes": 0.9414339317673652, "no": 0.05856606823263477}, "confidence": 0.8286739967579979, "seconds": 2.822944233000271, "prompt_tokens": 9962, "eval_tokens": 0, "server_latency_ms": null, "prompt_eval_ms": null, "total_ms": 2822.944233000271, "done_reason": null, "id": "on-17", "task": "ka", "slug": "how-a-vision-model-sees", "label": "yes", "kind": "on-page", "state_sha256": "6a885683413930cdad4c104dc4ede2330b45bb9903d54621ace0d4f518a89fa9", "prompt_sha256_matches_kev9b": null, "kev9b_refused": "kev-422", "card": "RTX 3090 (fp 33a0b4acb8e8)", "box": "benchbox", "os": {"pretty_name": "Ubuntu 26.04 LTS", "id": "ubuntu", "version_id": "26.04", "kernel": "7.0.0-31-generic", "arch": "x86_64", "source": "/etc/os-release"}, "dry": false, "runtime": "lev serve 0.1.1 (git 0ea955f5aee0) over HTTP on 127.0.0.1, one question per request", "lev_revision": "f8ef71157ec06a7d3b6435bc0756f9d735c33748", "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "lev_commit": "0ea955f5aee0d56d906121d1779af20fa8b922ed", "torch": "2.14.0+cu130", "torch_cuda": "13.0", "transformers": "5.17.0", "peft": "0.21.0", "dtype": "bfloat16", "kernels": "reference (no flash-linear-attention)", "arm": "lev-4b", "model": "interfaze-ai/lev@f8ef7115 on Qwen/Qwen3.5-4B@851bf6e8", "quant": "bf16 backbone + LoRA r=32 (PEFT, unmerged) + mode-B head fp32; calibrated as served; order_average on", "card_uuid": "GPU-fp:33a0b4acb8e8", "num_ctx": 0, "power_cap_w": 300, "seed": 20260921, "stamp": "2026-09-28T13:13:55Z"} {"mode": "kev-noul", "prompt_sha256": "9ed4b6bd4df4dbfce12a9a35ffc8a8e6cdc67795590fc668d5806c02f879bfb4", "n_options": 2, "kev_response": {"model": "Qwen/Qwen3.5-4B", "answers": {"q": {"type": "noul", "noul": 0.9149656213630695, "probabilities": {"0": 0.060113312003529276, "1": 0.004871262510136779, "2": 0.004995138733314728, "3": 0.005866054399727645, "4": 0.008563739782110788, "5": 0.010124422631935888, "6": 0.011421395343191833, "7": 0.0184975733870912, "8": 0.8755471012089618}, "confidence": 0.7422938627326301}}, "usage": {"input_tokens": 5123, "output_tokens": 0, "cached_input_tokens": 5037}}, "kev_noul": 0.9149656213630695, "choice": "yes", "probs": {"yes": 0.9149656213630695, "no": 0.08503437863693053}, "confidence": 0.7422938627326301, "seconds": 1.4284351160004007, "prompt_tokens": 5123, "eval_tokens": 0, "server_latency_ms": null, "prompt_eval_ms": null, "total_ms": 1428.4351160004007, "done_reason": null, "id": "on-18", "task": "ka", "slug": "fifty-seven-milliseconds", "label": "yes", "kind": "on-page", "state_sha256": "bec5d08ff12a12a4d87ca9d1d31e3bec848dfe329b6a47e9b8616e071da4ba5c", "prompt_sha256_matches_kev9b": true, "card": "RTX 3090 (fp 33a0b4acb8e8)", "box": "benchbox", "os": {"pretty_name": "Ubuntu 26.04 LTS", "id": "ubuntu", "version_id": "26.04", "kernel": "7.0.0-31-generic", "arch": "x86_64", "source": "/etc/os-release"}, "dry": false, "runtime": "lev serve 0.1.1 (git 0ea955f5aee0) over HTTP on 127.0.0.1, one question per request", "lev_revision": "f8ef71157ec06a7d3b6435bc0756f9d735c33748", "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "lev_commit": "0ea955f5aee0d56d906121d1779af20fa8b922ed", "torch": "2.14.0+cu130", "torch_cuda": "13.0", "transformers": "5.17.0", "peft": "0.21.0", "dtype": "bfloat16", "kernels": "reference (no flash-linear-attention)", "arm": "lev-4b", "model": "interfaze-ai/lev@f8ef7115 on Qwen/Qwen3.5-4B@851bf6e8", "quant": "bf16 backbone + LoRA r=32 (PEFT, unmerged) + mode-B head fp32; calibrated as served; order_average on", "card_uuid": "GPU-fp:33a0b4acb8e8", "num_ctx": 0, "power_cap_w": 300, "seed": 20260921, "stamp": "2026-09-28T13:13:56Z"} {"mode": "kev-noul", "prompt_sha256": "edb28ef33c63af998e62baa172266a3b22df050244e0d4c044638c421e7785c0", "n_options": 2, "kev_response": {"model": "Qwen/Qwen3.5-4B", "answers": {"q": {"type": "noul", "noul": 0.8837826099722073, "probabilities": {"0": 0.07298618445641725, "1": 0.008364011503971628, "2": 0.008007756387534872, "3": 0.009530724774684937, "4": 0.013774660263705173, "5": 0.0188699038425875, "6": 0.024666104829060058, "7": 0.03056083960386499, "8": 0.8132398143381736}, "confidence": 0.6276239768045857}}, "usage": {"input_tokens": 5831, "output_tokens": 0, "cached_input_tokens": 5745}}, "kev_noul": 0.8837826099722073, "choice": "yes", "probs": {"yes": 0.8837826099722073, "no": 0.11621739002779274}, "confidence": 0.6276239768045857, "seconds": 1.6413923839991185, "prompt_tokens": 5831, "eval_tokens": 0, "server_latency_ms": null, "prompt_eval_ms": null, "total_ms": 1641.3923839991185, "done_reason": null, "id": "on-19", "task": "ka", "slug": "the-dog-the-dice-and-the-painter", "label": "yes", "kind": "on-page", "state_sha256": "a136b61fadbc1ff5eded7327a0b3f4818be7eb2a2122cfd25b64de203f98cb0f", "prompt_sha256_matches_kev9b": true, "card": "RTX 3090 (fp 33a0b4acb8e8)", "box": "benchbox", "os": {"pretty_name": "Ubuntu 26.04 LTS", "id": "ubuntu", "version_id": "26.04", "kernel": "7.0.0-31-generic", "arch": "x86_64", "source": "/etc/os-release"}, "dry": false, "runtime": "lev serve 0.1.1 (git 0ea955f5aee0) over HTTP on 127.0.0.1, one question per request", "lev_revision": "f8ef71157ec06a7d3b6435bc0756f9d735c33748", "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "lev_commit": "0ea955f5aee0d56d906121d1779af20fa8b922ed", "torch": "2.14.0+cu130", "torch_cuda": "13.0", "transformers": "5.17.0", "peft": "0.21.0", "dtype": "bfloat16", "kernels": "reference (no flash-linear-attention)", "arm": "lev-4b", "model": "interfaze-ai/lev@f8ef7115 on Qwen/Qwen3.5-4B@851bf6e8", "quant": "bf16 backbone + LoRA r=32 (PEFT, unmerged) + mode-B head fp32; calibrated as served; order_average on", "card_uuid": "GPU-fp:33a0b4acb8e8", "num_ctx": 0, "power_cap_w": 300, "seed": 20260921, "stamp": "2026-09-28T13:13:58Z"} {"mode": "kev-noul", "prompt_sha256": "1d5ab0e80fcbfc522cbbe4dba6522120f03b714d21695880515d101660dfdb04", "n_options": 2, "kev_response": {"model": "Qwen/Qwen3.5-4B", "answers": {"q": {"type": "noul", "noul": 0.9491327740182995, "probabilities": {"0": 0.03536110485255173, "1": 0.003314756438062423, "2": 0.003733132919095278, "3": 0.003683468066003874, "4": 0.00587636672912761, "5": 0.006220556003970719, "6": 0.004510577419598748, "7": 0.008841246354542589, "8": 0.9284587912170471}, "confidence": 0.8464334123146776}}, "usage": {"input_tokens": 5256, "output_tokens": 0, "cached_input_tokens": 5167}}, "kev_noul": 0.9491327740182995, "choice": "yes", "probs": {"yes": 0.9491327740182995, "no": 0.05086722598170046}, "confidence": 0.8464334123146776, "seconds": 1.508013201999347, "prompt_tokens": 5256, "eval_tokens": 0, "server_latency_ms": null, "prompt_eval_ms": null, "total_ms": 1508.013201999347, "done_reason": null, "id": "on-20", "task": "ka", "slug": "where-new-knowledge-comes-from", "label": "yes", "kind": "on-page", "state_sha256": "26b4da101ba323b5f8e9999ee8b445f148b5cf9757bf21acaaa991f748bd4225", "prompt_sha256_matches_kev9b": true, "card": "RTX 3090 (fp 33a0b4acb8e8)", "box": "benchbox", "os": {"pretty_name": "Ubuntu 26.04 LTS", "id": "ubuntu", "version_id": "26.04", "kernel": "7.0.0-31-generic", "arch": "x86_64", "source": "/etc/os-release"}, "dry": false, "runtime": "lev serve 0.1.1 (git 0ea955f5aee0) over HTTP on 127.0.0.1, one question per request", "lev_revision": "f8ef71157ec06a7d3b6435bc0756f9d735c33748", "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "lev_commit": "0ea955f5aee0d56d906121d1779af20fa8b922ed", "torch": "2.14.0+cu130", "torch_cuda": "13.0", "transformers": "5.17.0", "peft": "0.21.0", "dtype": "bfloat16", "kernels": "reference (no flash-linear-attention)", "arm": "lev-4b", "model": "interfaze-ai/lev@f8ef7115 on Qwen/Qwen3.5-4B@851bf6e8", "quant": "bf16 backbone + LoRA r=32 (PEFT, unmerged) + mode-B head fp32; calibrated as served; order_average on", "card_uuid": "GPU-fp:33a0b4acb8e8", "num_ctx": 0, "power_cap_w": 300, "seed": 20260921, "stamp": "2026-09-28T13:13:59Z"} {"mode": "kev-noul", "prompt_sha256": "cd046120ec7849c3fe98cc45aa86a5fe535bea84f2adb6e91abe6fedcd6a34ba", "n_options": 2, "kev_response": {"model": "Qwen/Qwen3.5-4B", "answers": {"q": {"type": "noul", "noul": 0.9016915192153259, "probabilities": {"0": 0.06299794271605479, "1": 0.007365888572128545, "2": 0.007490241580911678, "3": 0.007797321796615616, "4": 0.0124601855068118, "5": 0.014632651847319711, "6": 0.012713037398261873, "7": 0.027830253709776404, "8": 0.8467124768721196}, "confidence": 0.6876633616776566}}, "usage": {"input_tokens": 5204, "output_tokens": 0, "cached_input_tokens": 5112}}, "kev_noul": 0.9016915192153259, "choice": "yes", "probs": {"yes": 0.9016915192153259, "no": 0.09830848078467413}, "confidence": 0.6876633616776566, "seconds": 1.4452587780015165, "prompt_tokens": 5204, "eval_tokens": 0, "server_latency_ms": null, "prompt_eval_ms": null, "total_ms": 1445.2587780015165, "done_reason": null, "id": "on-21", "task": "ka", "slug": "compression-is-the-objective", "label": "yes", "kind": "on-page", "state_sha256": "68bc7c0890d5d05afefc1bbd6178315fddd18fb35775c738f50544096a982f7f", "prompt_sha256_matches_kev9b": true, "card": "RTX 3090 (fp 33a0b4acb8e8)", "box": "benchbox", "os": {"pretty_name": "Ubuntu 26.04 LTS", "id": "ubuntu", "version_id": "26.04", "kernel": "7.0.0-31-generic", "arch": "x86_64", "source": "/etc/os-release"}, "dry": false, "runtime": "lev serve 0.1.1 (git 0ea955f5aee0) over HTTP on 127.0.0.1, one question per request", "lev_revision": "f8ef71157ec06a7d3b6435bc0756f9d735c33748", "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "lev_commit": "0ea955f5aee0d56d906121d1779af20fa8b922ed", "torch": "2.14.0+cu130", "torch_cuda": "13.0", "transformers": "5.17.0", "peft": "0.21.0", "dtype": "bfloat16", "kernels": "reference (no flash-linear-attention)", "arm": "lev-4b", "model": "interfaze-ai/lev@f8ef7115 on Qwen/Qwen3.5-4B@851bf6e8", "quant": "bf16 backbone + LoRA r=32 (PEFT, unmerged) + mode-B head fp32; calibrated as served; order_average on", "card_uuid": "GPU-fp:33a0b4acb8e8", "num_ctx": 0, "power_cap_w": 300, "seed": 20260921, "stamp": "2026-09-28T13:14:01Z"} {"mode": "kev-noul", "prompt_sha256": "284c6b4eaa96f868169bb2e6f9693a52a1b617f6046fa04c5f1cbc436ef1f25c", "n_options": 2, "kev_response": {"model": "Qwen/Qwen3.5-4B", "answers": {"q": {"type": "noul", "noul": 0.9755469882626333, "probabilities": {"0": 0.017617219608451154, "1": 0.001536735908040048, "2": 0.0016024184629005243, "3": 0.00158906111358955, "4": 0.0026039082088326233, "5": 0.0026745995013297374, "6": 0.001791875088041775, "7": 0.004346187814290353, "8": 0.9662379942945243}, "confidence": 0.92571593100371}}, "usage": {"input_tokens": 7145, "output_tokens": 0, "cached_input_tokens": 7055}}, "kev_noul": 0.9755469882626333, "choice": "yes", "probs": {"yes": 0.9755469882626333, "no": 0.024453011737366692}, "confidence": 0.92571593100371, "seconds": 1.9992739659992367, "prompt_tokens": 7145, "eval_tokens": 0, "server_latency_ms": null, "prompt_eval_ms": null, "total_ms": 1999.2739659992367, "done_reason": null, "id": "on-22", "task": "ka", "slug": "the-free-speed-wasnt-free", "label": "yes", "kind": "on-page", "state_sha256": "3ae132432376b190cec501235414ae89382270e55515119fc3f0b4d51c4d6de3", "prompt_sha256_matches_kev9b": true, "card": "RTX 3090 (fp 33a0b4acb8e8)", "box": "benchbox", "os": {"pretty_name": "Ubuntu 26.04 LTS", "id": "ubuntu", "version_id": "26.04", "kernel": "7.0.0-31-generic", "arch": "x86_64", "source": "/etc/os-release"}, "dry": false, "runtime": "lev serve 0.1.1 (git 0ea955f5aee0) over HTTP on 127.0.0.1, one question per request", "lev_revision": "f8ef71157ec06a7d3b6435bc0756f9d735c33748", "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "lev_commit": "0ea955f5aee0d56d906121d1779af20fa8b922ed", "torch": "2.14.0+cu130", "torch_cuda": "13.0", "transformers": "5.17.0", "peft": "0.21.0", "dtype": "bfloat16", "kernels": "reference (no flash-linear-attention)", "arm": "lev-4b", "model": "interfaze-ai/lev@f8ef7115 on Qwen/Qwen3.5-4B@851bf6e8", "quant": "bf16 backbone + LoRA r=32 (PEFT, unmerged) + mode-B head fp32; calibrated as served; order_average on", "card_uuid": "GPU-fp:33a0b4acb8e8", "num_ctx": 0, "power_cap_w": 300, "seed": 20260921, "stamp": "2026-09-28T13:14:03Z"} {"mode": "kev-noul", "prompt_sha256": "330898f43055b8d1b8e0c8f31843f1ba31ca73a1132e96da4f147f9c99ab302c", "n_options": 2, "kev_response": {"model": "Qwen/Qwen3.5-4B", "answers": {"q": {"type": "noul", "noul": 0.9141964729587239, "probabilities": {"0": 0.059326630628470385, "1": 0.005664651925975785, "2": 0.0057314262089763505, "3": 0.0063582986595046714, "4": 0.0089767047057152, "5": 0.010974003727596797, "6": 0.009925240092917367, "7": 0.017303247077748603, "8": 0.8757397969730949}, "confidence": 0.7425371454955771}}, "usage": {"input_tokens": 8190, "output_tokens": 0, "cached_input_tokens": 8104}}, "kev_noul": 0.9141964729587239, "choice": "yes", "probs": {"yes": 0.9141964729587239, "no": 0.08580352704127614}, "confidence": 0.7425371454955771, "seconds": 2.2997985210004117, "prompt_tokens": 8190, "eval_tokens": 0, "server_latency_ms": null, "prompt_eval_ms": null, "total_ms": 2299.7985210004117, "done_reason": null, "id": "on-23", "task": "ka", "slug": "the-compressed-photograph", "label": "yes", "kind": "on-page", "state_sha256": "30efd8ed8a4dcfd536f819d785eb18027a4f37e47597ae0e18294a5ede65b01a", "prompt_sha256_matches_kev9b": true, "card": "RTX 3090 (fp 33a0b4acb8e8)", "box": "benchbox", "os": {"pretty_name": "Ubuntu 26.04 LTS", "id": "ubuntu", "version_id": "26.04", "kernel": "7.0.0-31-generic", "arch": "x86_64", "source": "/etc/os-release"}, "dry": false, "runtime": "lev serve 0.1.1 (git 0ea955f5aee0) over HTTP on 127.0.0.1, one question per request", "lev_revision": "f8ef71157ec06a7d3b6435bc0756f9d735c33748", "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "lev_commit": "0ea955f5aee0d56d906121d1779af20fa8b922ed", "torch": "2.14.0+cu130", "torch_cuda": "13.0", "transformers": "5.17.0", "peft": "0.21.0", "dtype": "bfloat16", "kernels": "reference (no flash-linear-attention)", "arm": "lev-4b", "model": "interfaze-ai/lev@f8ef7115 on Qwen/Qwen3.5-4B@851bf6e8", "quant": "bf16 backbone + LoRA r=32 (PEFT, unmerged) + mode-B head fp32; calibrated as served; order_average on", "card_uuid": "GPU-fp:33a0b4acb8e8", "num_ctx": 0, "power_cap_w": 300, "seed": 20260921, "stamp": "2026-09-28T13:14:05Z"} {"mode": "kev-noul", "prompt_sha256": "dc069ad09e93b282d9e3b744ce733f164c25b6c16e916e03f86f23dcb553c53c", "n_options": 2, "kev_response": {"model": "Qwen/Qwen3.5-4B", "answers": {"q": {"type": "noul", "noul": 0.9648306061336881, "probabilities": {"0": 0.026313803231466808, "1": 0.0019676834584375757, "2": 0.001954550800197162, "3": 0.0021394819975090024, "4": 0.0035117272306745173, "5": 0.003424638657284999, "6": 0.0024718280407597607, "7": 0.005381745104896597, "8": 0.9528345414787736}, "confidence": 0.897239667922888}}, "usage": {"input_tokens": 7301, "output_tokens": 0, "cached_input_tokens": 7202}}, "kev_noul": 0.9648306061336881, "choice": "yes", "probs": {"yes": 0.9648306061336881, "no": 0.035169393866311904}, "confidence": 0.897239667922888, "seconds": 2.0868387449991133, "prompt_tokens": 7301, "eval_tokens": 0, "server_latency_ms": null, "prompt_eval_ms": null, "total_ms": 2086.8387449991133, "done_reason": null, "id": "on-24", "task": "ka", "slug": "three-at-the-table", "label": "yes", "kind": "on-page", "state_sha256": "5ee526a9690f5511741a41a6eb1e91418702e2cd8e49d1c2db9e03802a7bde5c", "prompt_sha256_matches_kev9b": true, "card": "RTX 3090 (fp 33a0b4acb8e8)", "box": "benchbox", "os": {"pretty_name": "Ubuntu 26.04 LTS", "id": "ubuntu", "version_id": "26.04", "kernel": "7.0.0-31-generic", "arch": "x86_64", "source": "/etc/os-release"}, "dry": false, "runtime": "lev serve 0.1.1 (git 0ea955f5aee0) over HTTP on 127.0.0.1, one question per request", "lev_revision": "f8ef71157ec06a7d3b6435bc0756f9d735c33748", "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "lev_commit": "0ea955f5aee0d56d906121d1779af20fa8b922ed", "torch": "2.14.0+cu130", "torch_cuda": "13.0", "transformers": "5.17.0", "peft": "0.21.0", "dtype": "bfloat16", "kernels": "reference (no flash-linear-attention)", "arm": "lev-4b", "model": "interfaze-ai/lev@f8ef7115 on Qwen/Qwen3.5-4B@851bf6e8", "quant": "bf16 backbone + LoRA r=32 (PEFT, unmerged) + mode-B head fp32; calibrated as served; order_average on", "card_uuid": "GPU-fp:33a0b4acb8e8", "num_ctx": 0, "power_cap_w": 300, "seed": 20260921, "stamp": "2026-09-28T13:14:07Z"} {"mode": "kev-noul", "prompt_sha256": "f1ea881a09acefb481e491b99d0fa7892b9a7e558ca155a01fa58e2619b119b2", "n_options": 2, "kev_response": {"model": "Qwen/Qwen3.5-4B", "answers": {"q": {"type": "noul", "noul": 0.9525813015942802, "probabilities": {"0": 0.03278711468226612, "1": 0.003347414294482978, "2": 0.0034210651970150176, "3": 0.0033699056249278515, "4": 0.0055036229518292505, "5": 0.005671995763334989, "6": 0.004224467014534299, "7": 0.008765437293128087, "8": 0.9329089771784813}, "confidence": 0.8555337551539923}}, "usage": {"input_tokens": 5218, "output_tokens": 0, "cached_input_tokens": 5129}}, "kev_noul": 0.9525813015942802, "choice": "yes", "probs": {"yes": 0.9525813015942802, "no": 0.04741869840571977}, "confidence": 0.8555337551539923, "seconds": 1.4513687679991563, "prompt_tokens": 5218, "eval_tokens": 0, "server_latency_ms": null, "prompt_eval_ms": null, "total_ms": 1451.3687679991563, "done_reason": null, "id": "on-25", "task": "ka", "slug": "reading-is-fast", "label": "yes", "kind": "on-page", "state_sha256": "79b9c9e1be3ac3cdc484fec1f5eb9a83bf909612d6eb1c3c7c83bb469125ff2f", "prompt_sha256_matches_kev9b": true, "card": "RTX 3090 (fp 33a0b4acb8e8)", "box": "benchbox", "os": {"pretty_name": "Ubuntu 26.04 LTS", "id": "ubuntu", "version_id": "26.04", "kernel": "7.0.0-31-generic", "arch": "x86_64", "source": "/etc/os-release"}, "dry": false, "runtime": "lev serve 0.1.1 (git 0ea955f5aee0) over HTTP on 127.0.0.1, one question per request", "lev_revision": "f8ef71157ec06a7d3b6435bc0756f9d735c33748", "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "lev_commit": "0ea955f5aee0d56d906121d1779af20fa8b922ed", "torch": "2.14.0+cu130", "torch_cuda": "13.0", "transformers": "5.17.0", "peft": "0.21.0", "dtype": "bfloat16", "kernels": "reference (no flash-linear-attention)", "arm": "lev-4b", "model": "interfaze-ai/lev@f8ef7115 on Qwen/Qwen3.5-4B@851bf6e8", "quant": "bf16 backbone + LoRA r=32 (PEFT, unmerged) + mode-B head fp32; calibrated as served; order_average on", "card_uuid": "GPU-fp:33a0b4acb8e8", "num_ctx": 0, "power_cap_w": 300, "seed": 20260921, "stamp": "2026-09-28T13:14:09Z"} {"mode": "kev-noul", "prompt_sha256": "db3855f1aa358bda6553b365078e77ce8df6b68b5e99a51f512c13e97d5e9ce3", "n_options": 2, "kev_response": {"model": "Qwen/Qwen3.5-4B", "answers": {"q": {"type": "noul", "noul": 0.9625190286634486, "probabilities": {"0": 0.028430368184637737, "1": 0.0019179594089207812, "2": 0.0019683801022018317, "3": 0.0022326509739362234, "4": 0.003530658484116459, "5": 0.003675406201151706, "6": 0.0025978884149493423, "7": 0.005660944500153238, "8": 0.9499857437299328}, "confidence": 0.891278319723676}}, "usage": {"input_tokens": 5605, "output_tokens": 0, "cached_input_tokens": 5512}}, "kev_noul": 0.9625190286634486, "choice": "yes", "probs": {"yes": 0.9625190286634486, "no": 0.03748097133655137}, "confidence": 0.891278319723676, "seconds": 1.5902993439995043, "prompt_tokens": 5605, "eval_tokens": 0, "server_latency_ms": null, "prompt_eval_ms": null, "total_ms": 1590.2993439995043, "done_reason": null, "id": "on-26", "task": "ka", "slug": "three-librarians", "label": "yes", "kind": "on-page", "state_sha256": "8883de675623f6941e01030f0e305b0a8f73197ce266d599ff002b148a135167", "prompt_sha256_matches_kev9b": true, "card": "RTX 3090 (fp 33a0b4acb8e8)", "box": "benchbox", "os": {"pretty_name": "Ubuntu 26.04 LTS", "id": "ubuntu", "version_id": "26.04", "kernel": "7.0.0-31-generic", "arch": "x86_64", "source": "/etc/os-release"}, "dry": false, "runtime": "lev serve 0.1.1 (git 0ea955f5aee0) over HTTP on 127.0.0.1, one question per request", "lev_revision": "f8ef71157ec06a7d3b6435bc0756f9d735c33748", "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "lev_commit": "0ea955f5aee0d56d906121d1779af20fa8b922ed", "torch": "2.14.0+cu130", "torch_cuda": "13.0", "transformers": "5.17.0", "peft": "0.21.0", "dtype": "bfloat16", "kernels": "reference (no flash-linear-attention)", "arm": "lev-4b", "model": "interfaze-ai/lev@f8ef7115 on Qwen/Qwen3.5-4B@851bf6e8", "quant": "bf16 backbone + LoRA r=32 (PEFT, unmerged) + mode-B head fp32; calibrated as served; order_average on", "card_uuid": "GPU-fp:33a0b4acb8e8", "num_ctx": 0, "power_cap_w": 300, "seed": 20260921, "stamp": "2026-09-28T13:14:10Z"} {"mode": "kev-noul", "prompt_sha256": "f1eeaf9128a6de26f5c7be831b1b8f7c4d6fd25c99f9c01f5dde951ec2e73c10", "n_options": 2, "kev_response": {"model": "Qwen/Qwen3.5-4B", "answers": {"q": {"type": "noul", "noul": 0.9663748949500237, "probabilities": {"0": 0.02397903922146531, "1": 0.0021886178049151184, "2": 0.0023864385566064717, "3": 0.0022914947006824034, "4": 0.003677185826346105, "5": 0.003473723640293037, "6": 0.002572092314586688, "7": 0.005797998295194278, "8": 0.9536334096399106}, "confidence": 0.8988323851932696}}, "usage": {"input_tokens": 14590, "output_tokens": 0, "cached_input_tokens": 14501}}, "kev_noul": 0.9663748949500237, "choice": "yes", "probs": {"yes": 0.9663748949500237, "no": 0.03362510504997629}, "confidence": 0.8988323851932696, "seconds": 4.217633428001136, "prompt_tokens": 14590, "eval_tokens": 0, "server_latency_ms": null, "prompt_eval_ms": null, "total_ms": 4217.633428001136, "done_reason": null, "id": "on-27", "task": "ka", "slug": "the-new-kid", "label": "yes", "kind": "on-page", "state_sha256": "676acfa90c7de336adc467067853954373df2cc40b45fb728501238f7b963b79", "prompt_sha256_matches_kev9b": null, "kev9b_refused": "kev-422", "card": "RTX 3090 (fp 33a0b4acb8e8)", "box": "benchbox", "os": {"pretty_name": "Ubuntu 26.04 LTS", "id": "ubuntu", "version_id": "26.04", "kernel": "7.0.0-31-generic", "arch": "x86_64", "source": "/etc/os-release"}, "dry": false, "runtime": "lev serve 0.1.1 (git 0ea955f5aee0) over HTTP on 127.0.0.1, one question per request", "lev_revision": "f8ef71157ec06a7d3b6435bc0756f9d735c33748", "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "lev_commit": "0ea955f5aee0d56d906121d1779af20fa8b922ed", "torch": "2.14.0+cu130", "torch_cuda": "13.0", "transformers": "5.17.0", "peft": "0.21.0", "dtype": "bfloat16", "kernels": "reference (no flash-linear-attention)", "arm": "lev-4b", "model": "interfaze-ai/lev@f8ef7115 on Qwen/Qwen3.5-4B@851bf6e8", "quant": "bf16 backbone + LoRA r=32 (PEFT, unmerged) + mode-B head fp32; calibrated as served; order_average on", "card_uuid": "GPU-fp:33a0b4acb8e8", "num_ctx": 0, "power_cap_w": 300, "seed": 20260921, "stamp": "2026-09-28T13:14:15Z"} {"mode": "kev-noul", "prompt_sha256": "8da3fb4865799703851d3c144a54788423c49155fd44bff34831e0882ad2f2a7", "n_options": 2, "kev_response": {"model": "Qwen/Qwen3.5-4B", "answers": {"q": {"type": "noul", "noul": 0.962814100734887, "probabilities": {"0": 0.027678770740378104, "1": 0.002136646634325716, "2": 0.0021964904817871915, "3": 0.002256121136427186, "4": 0.0035902425879926012, "5": 0.0036938899354673283, "6": 0.0027305376573948416, "7": 0.00573723771157693, "8": 0.94998006311465}, "confidence": 0.8912233235957783}}, "usage": {"input_tokens": 12013, "output_tokens": 0, "cached_input_tokens": 11914}}, "kev_noul": 0.962814100734887, "choice": "yes", "probs": {"yes": 0.962814100734887, "no": 0.037185899265112954}, "confidence": 0.8912233235957783, "seconds": 3.4429194669992285, "prompt_tokens": 12013, "eval_tokens": 0, "server_latency_ms": null, "prompt_eval_ms": null, "total_ms": 3442.9194669992285, "done_reason": null, "id": "on-28", "task": "ka", "slug": "llms-txt", "label": "yes", "kind": "on-page", "state_sha256": "96cab65488f086e43a1b03248954f7a7abadb4ae49c8591b267586825532ff41", "prompt_sha256_matches_kev9b": null, "kev9b_refused": "kev-422", "card": "RTX 3090 (fp 33a0b4acb8e8)", "box": "benchbox", "os": {"pretty_name": "Ubuntu 26.04 LTS", "id": "ubuntu", "version_id": "26.04", "kernel": "7.0.0-31-generic", "arch": "x86_64", "source": "/etc/os-release"}, "dry": false, "runtime": "lev serve 0.1.1 (git 0ea955f5aee0) over HTTP on 127.0.0.1, one question per request", "lev_revision": "f8ef71157ec06a7d3b6435bc0756f9d735c33748", "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "lev_commit": "0ea955f5aee0d56d906121d1779af20fa8b922ed", "torch": "2.14.0+cu130", "torch_cuda": "13.0", "transformers": "5.17.0", "peft": "0.21.0", "dtype": "bfloat16", "kernels": "reference (no flash-linear-attention)", "arm": "lev-4b", "model": "interfaze-ai/lev@f8ef7115 on Qwen/Qwen3.5-4B@851bf6e8", "quant": "bf16 backbone + LoRA r=32 (PEFT, unmerged) + mode-B head fp32; calibrated as served; order_average on", "card_uuid": "GPU-fp:33a0b4acb8e8", "num_ctx": 0, "power_cap_w": 300, "seed": 20260921, "stamp": "2026-09-28T13:14:18Z"} {"refused": "overlong", "choice": null, "probs": null, "seconds": null, "why": "inherited from arm 2's own refusal rows so the denominator matches (GAPS-CARD.md, arm 12)", "id": "on-29", "task": "ka", "slug": "the-open-call", "label": "yes", "kind": "on-page", "state_sha256": "dd50fd254da5ef89e4284d245102a3ca9f9f7dd4e41f90c7687adb01223798a4", "prompt_sha256_matches_kev9b": null, "card": "RTX 3090 (fp 33a0b4acb8e8)", "box": "benchbox", "os": {"pretty_name": "Ubuntu 26.04 LTS", "id": "ubuntu", "version_id": "26.04", "kernel": "7.0.0-31-generic", "arch": "x86_64", "source": "/etc/os-release"}, "dry": false, "runtime": "lev serve 0.1.1 (git 0ea955f5aee0) over HTTP on 127.0.0.1, one question per request", "lev_revision": "f8ef71157ec06a7d3b6435bc0756f9d735c33748", "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "lev_commit": "0ea955f5aee0d56d906121d1779af20fa8b922ed", "torch": "2.14.0+cu130", "torch_cuda": "13.0", "transformers": "5.17.0", "peft": "0.21.0", "dtype": "bfloat16", "kernels": "reference (no flash-linear-attention)", "arm": "lev-4b", "model": "interfaze-ai/lev@f8ef7115 on Qwen/Qwen3.5-4B@851bf6e8", "quant": "bf16 backbone + LoRA r=32 (PEFT, unmerged) + mode-B head fp32; calibrated as served; order_average on", "card_uuid": "GPU-fp:33a0b4acb8e8", "num_ctx": 0, "power_cap_w": 300, "seed": 20260921, "stamp": "2026-09-28T13:14:18Z"} {"mode": "kev-noul", "prompt_sha256": "c0ddf9d73aa3cefab86b3e69e03b8e812a3a5fa698bdee051b5d28197024b673", "n_options": 2, "kev_response": {"model": "Qwen/Qwen3.5-4B", "answers": {"q": {"type": "noul", "noul": 0.9149478058782077, "probabilities": {"0": 0.05612769418288624, "1": 0.006076183051116371, "2": 0.006584610871210603, "3": 0.006993660787918908, "4": 0.010592962551920635, "5": 0.011712281964244274, "6": 0.009080864128558453, "7": 0.018716324629042883, "8": 0.8741154178331015}, "confidence": 0.7390443491258669}}, "usage": {"input_tokens": 4704, "output_tokens": 0, "cached_input_tokens": 4618}}, "kev_noul": 0.9149478058782077, "choice": "yes", "probs": {"yes": 0.9149478058782077, "no": 0.08505219412179232}, "confidence": 0.7390443491258669, "seconds": 1.3360196580015327, "prompt_tokens": 4704, "eval_tokens": 0, "server_latency_ms": null, "prompt_eval_ms": null, "total_ms": 1336.0196580015327, "done_reason": null, "id": "on-30", "task": "ka", "slug": "where-your-question-goes", "label": "yes", "kind": "on-page", "state_sha256": "cdc24e58a19354d77487d8df5e8782b2c15b4483afe7405f34478bf769d44b11", "prompt_sha256_matches_kev9b": true, "card": "RTX 3090 (fp 33a0b4acb8e8)", "box": "benchbox", "os": {"pretty_name": "Ubuntu 26.04 LTS", "id": "ubuntu", "version_id": "26.04", "kernel": "7.0.0-31-generic", "arch": "x86_64", "source": "/etc/os-release"}, "dry": false, "runtime": "lev serve 0.1.1 (git 0ea955f5aee0) over HTTP on 127.0.0.1, one question per request", "lev_revision": "f8ef71157ec06a7d3b6435bc0756f9d735c33748", "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "lev_commit": "0ea955f5aee0d56d906121d1779af20fa8b922ed", "torch": "2.14.0+cu130", "torch_cuda": "13.0", "transformers": "5.17.0", "peft": "0.21.0", "dtype": "bfloat16", "kernels": "reference (no flash-linear-attention)", "arm": "lev-4b", "model": "interfaze-ai/lev@f8ef7115 on Qwen/Qwen3.5-4B@851bf6e8", "quant": "bf16 backbone + LoRA r=32 (PEFT, unmerged) + mode-B head fp32; calibrated as served; order_average on", "card_uuid": "GPU-fp:33a0b4acb8e8", "num_ctx": 0, "power_cap_w": 300, "seed": 20260921, "stamp": "2026-09-28T13:14:19Z"} {"mode": "kev-noul", "prompt_sha256": "7a56fea6c96e13f42ed4a44e374ab48cb8f33b63e0a508f72d864481b95a69d3", "n_options": 2, "kev_response": {"model": "Qwen/Qwen3.5-4B", "answers": {"q": {"type": "noul", "noul": 0.9269274908826455, "probabilities": {"0": 0.04549281306817151, "1": 0.0057066130803623905, "2": 0.007070386193478229, "3": 0.006204857747739237, "4": 0.010566702125212478, "5": 0.010566702125212478, "6": 0.008528541286705354, "7": 0.01622067348146394, "8": 0.8896427108916544}, "confidence": 0.7684907070128816}}, "usage": {"input_tokens": 4925, "output_tokens": 0, "cached_input_tokens": 4836}}, "kev_noul": 0.9269274908826455, "choice": "yes", "probs": {"yes": 0.9269274908826455, "no": 0.07307250911735452}, "confidence": 0.7684907070128816, "seconds": 1.3935553940009413, "prompt_tokens": 4925, "eval_tokens": 0, "server_latency_ms": null, "prompt_eval_ms": null, "total_ms": 1393.5553940009413, "done_reason": null, "id": "on-31", "task": "ka", "slug": "how-we-work", "label": "yes", "kind": "on-page", "state_sha256": "77d23d89a5adf6c0d561223d2c37308b5118a80f1f07f35f78bb1b0d08276e99", "prompt_sha256_matches_kev9b": true, "card": "RTX 3090 (fp 33a0b4acb8e8)", "box": "benchbox", "os": {"pretty_name": "Ubuntu 26.04 LTS", "id": "ubuntu", "version_id": "26.04", "kernel": "7.0.0-31-generic", "arch": "x86_64", "source": "/etc/os-release"}, "dry": false, "runtime": "lev serve 0.1.1 (git 0ea955f5aee0) over HTTP on 127.0.0.1, one question per request", "lev_revision": "f8ef71157ec06a7d3b6435bc0756f9d735c33748", "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "lev_commit": "0ea955f5aee0d56d906121d1779af20fa8b922ed", "torch": "2.14.0+cu130", "torch_cuda": "13.0", "transformers": "5.17.0", "peft": "0.21.0", "dtype": "bfloat16", "kernels": "reference (no flash-linear-attention)", "arm": "lev-4b", "model": "interfaze-ai/lev@f8ef7115 on Qwen/Qwen3.5-4B@851bf6e8", "quant": "bf16 backbone + LoRA r=32 (PEFT, unmerged) + mode-B head fp32; calibrated as served; order_average on", "card_uuid": "GPU-fp:33a0b4acb8e8", "num_ctx": 0, "power_cap_w": 300, "seed": 20260921, "stamp": "2026-09-28T13:14:21Z"} {"mode": "kev-noul", "prompt_sha256": "2b8132cb0cfe51926d74fdca96be58607c86cbb4cf971432cd0fd6a73e19f2a5", "n_options": 2, "kev_response": {"model": "Qwen/Qwen3.5-4B", "answers": {"q": {"type": "noul", "noul": 0.953533309547285, "probabilities": {"0": 0.034263784093075025, "1": 0.0027304581347566814, "2": 0.0025896555079632723, "3": 0.0029416435713394873, "4": 0.004519423098461472, "5": 0.004930495182701733, "6": 0.0039003269489831897, "7": 0.007594061189427519, "8": 0.9365301522732915}, "confidence": 0.863203560945777}}, "usage": {"input_tokens": 5829, "output_tokens": 0, "cached_input_tokens": 5741}}, "kev_noul": 0.953533309547285, "choice": "yes", "probs": {"yes": 0.953533309547285, "no": 0.04646669045271501}, "confidence": 0.863203560945777, "seconds": 1.649257637998744, "prompt_tokens": 5829, "eval_tokens": 0, "server_latency_ms": null, "prompt_eval_ms": null, "total_ms": 1649.257637998744, "done_reason": null, "id": "on-32", "task": "ka", "slug": "the-move", "label": "yes", "kind": "on-page", "state_sha256": "3d1d2fe049acbdd5e286fbc87c126a77f0410d05b6a51694ef22f6658b409e9f", "prompt_sha256_matches_kev9b": true, "card": "RTX 3090 (fp 33a0b4acb8e8)", "box": "benchbox", "os": {"pretty_name": "Ubuntu 26.04 LTS", "id": "ubuntu", "version_id": "26.04", "kernel": "7.0.0-31-generic", "arch": "x86_64", "source": "/etc/os-release"}, "dry": false, "runtime": "lev serve 0.1.1 (git 0ea955f5aee0) over HTTP on 127.0.0.1, one question per request", "lev_revision": "f8ef71157ec06a7d3b6435bc0756f9d735c33748", "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "lev_commit": "0ea955f5aee0d56d906121d1779af20fa8b922ed", "torch": "2.14.0+cu130", "torch_cuda": "13.0", "transformers": "5.17.0", "peft": "0.21.0", "dtype": "bfloat16", "kernels": "reference (no flash-linear-attention)", "arm": "lev-4b", "model": "interfaze-ai/lev@f8ef7115 on Qwen/Qwen3.5-4B@851bf6e8", "quant": "bf16 backbone + LoRA r=32 (PEFT, unmerged) + mode-B head fp32; calibrated as served; order_average on", "card_uuid": "GPU-fp:33a0b4acb8e8", "num_ctx": 0, "power_cap_w": 300, "seed": 20260921, "stamp": "2026-09-28T13:14:22Z"} {"refused": "overlong", "choice": null, "probs": null, "seconds": null, "why": "inherited from arm 2's own refusal rows so the denominator matches (GAPS-CARD.md, arm 12)", "id": "on-33", "task": "ka", "slug": "chair-trials", "label": "yes", "kind": "on-page", "state_sha256": "33688798032704bdb7db628ffa9c2c557245b03bcc2611f357853ee101698402", "prompt_sha256_matches_kev9b": null, "card": "RTX 3090 (fp 33a0b4acb8e8)", "box": "benchbox", "os": {"pretty_name": "Ubuntu 26.04 LTS", "id": "ubuntu", "version_id": "26.04", "kernel": "7.0.0-31-generic", "arch": "x86_64", "source": "/etc/os-release"}, "dry": false, "runtime": "lev serve 0.1.1 (git 0ea955f5aee0) over HTTP on 127.0.0.1, one question per request", "lev_revision": "f8ef71157ec06a7d3b6435bc0756f9d735c33748", "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "lev_commit": "0ea955f5aee0d56d906121d1779af20fa8b922ed", "torch": "2.14.0+cu130", "torch_cuda": "13.0", "transformers": "5.17.0", "peft": "0.21.0", "dtype": "bfloat16", "kernels": "reference (no flash-linear-attention)", "arm": "lev-4b", "model": "interfaze-ai/lev@f8ef7115 on Qwen/Qwen3.5-4B@851bf6e8", "quant": "bf16 backbone + LoRA r=32 (PEFT, unmerged) + mode-B head fp32; calibrated as served; order_average on", "card_uuid": "GPU-fp:33a0b4acb8e8", "num_ctx": 0, "power_cap_w": 300, "seed": 20260921, "stamp": "2026-09-28T13:14:22Z"} {"refused": "overlong", "choice": null, "probs": null, "seconds": null, "why": "inherited from arm 2's own refusal rows so the denominator matches (GAPS-CARD.md, arm 12)", "id": "on-34", "task": "ka", "slug": "cove-voice-head-to-head", "label": "yes", "kind": "on-page", "state_sha256": "53932c8b9bd51ad2a1b2e91fef30dd9dfc61b5c1e531fb3ee4818db6fbf2f5ee", "prompt_sha256_matches_kev9b": null, "card": "RTX 3090 (fp 33a0b4acb8e8)", "box": "benchbox", "os": {"pretty_name": "Ubuntu 26.04 LTS", "id": "ubuntu", "version_id": "26.04", "kernel": "7.0.0-31-generic", "arch": "x86_64", "source": "/etc/os-release"}, "dry": false, "runtime": "lev serve 0.1.1 (git 0ea955f5aee0) over HTTP on 127.0.0.1, one question per request", "lev_revision": "f8ef71157ec06a7d3b6435bc0756f9d735c33748", "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "lev_commit": "0ea955f5aee0d56d906121d1779af20fa8b922ed", "torch": "2.14.0+cu130", "torch_cuda": "13.0", "transformers": "5.17.0", "peft": "0.21.0", "dtype": "bfloat16", "kernels": "reference (no flash-linear-attention)", "arm": "lev-4b", "model": "interfaze-ai/lev@f8ef7115 on Qwen/Qwen3.5-4B@851bf6e8", "quant": "bf16 backbone + LoRA r=32 (PEFT, unmerged) + mode-B head fp32; calibrated as served; order_average on", "card_uuid": "GPU-fp:33a0b4acb8e8", "num_ctx": 0, "power_cap_w": 300, "seed": 20260921, "stamp": "2026-09-28T13:14:22Z"} {"mode": "kev-noul", "prompt_sha256": "75cf89ce8145ee0fe72c39c2235cdad4cfb0757e13eddc7b65203c9c6fd8980a", "n_options": 2, "kev_response": {"model": "Qwen/Qwen3.5-4B", "answers": {"q": {"type": "noul", "noul": 0.9455910740832275, "probabilities": {"0": 0.039709273113041864, "1": 0.0030749714963849077, "2": 0.0031604347874666272, "3": 0.0035638052650156377, "4": 0.005379881403856535, "5": 0.005869217988077114, "6": 0.005150731757776212, "7": 0.009862143810062712, "8": 0.9242295403783184}, "confidence": 0.8379959465650174}}, "usage": {"input_tokens": 10061, "output_tokens": 0, "cached_input_tokens": 9968}}, "kev_noul": 0.9455910740832275, "choice": "yes", "probs": {"yes": 0.9455910740832275, "no": 0.05440892591677249}, "confidence": 0.8379959465650174, "seconds": 2.861443056001008, "prompt_tokens": 10061, "eval_tokens": 0, "server_latency_ms": null, "prompt_eval_ms": null, "total_ms": 2861.443056001008, "done_reason": null, "id": "on-35", "task": "ka", "slug": "outside-judges", "label": "yes", "kind": "on-page", "state_sha256": "fc7235d0fabb2221f559392fc0236d981f9898d96425df0d7355f2d6f11d17d6", "prompt_sha256_matches_kev9b": null, "kev9b_refused": "kev-422", "card": "RTX 3090 (fp 33a0b4acb8e8)", "box": "benchbox", "os": {"pretty_name": "Ubuntu 26.04 LTS", "id": "ubuntu", "version_id": "26.04", "kernel": "7.0.0-31-generic", "arch": "x86_64", "source": "/etc/os-release"}, "dry": false, "runtime": "lev serve 0.1.1 (git 0ea955f5aee0) over HTTP on 127.0.0.1, one question per request", "lev_revision": "f8ef71157ec06a7d3b6435bc0756f9d735c33748", "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "lev_commit": "0ea955f5aee0d56d906121d1779af20fa8b922ed", "torch": "2.14.0+cu130", "torch_cuda": "13.0", "transformers": "5.17.0", "peft": "0.21.0", "dtype": "bfloat16", "kernels": "reference (no flash-linear-attention)", "arm": "lev-4b", "model": "interfaze-ai/lev@f8ef7115 on Qwen/Qwen3.5-4B@851bf6e8", "quant": "bf16 backbone + LoRA r=32 (PEFT, unmerged) + mode-B head fp32; calibrated as served; order_average on", "card_uuid": "GPU-fp:33a0b4acb8e8", "num_ctx": 0, "power_cap_w": 300, "seed": 20260921, "stamp": "2026-09-28T13:14:25Z"} {"mode": "kev-noul", "prompt_sha256": "caa9227270616296639b0112ed1ca9b6cd49e8bb08ad5a39b6a5ec589f57fbad", "n_options": 2, "kev_response": {"model": "Qwen/Qwen3.5-4B", "answers": {"q": {"type": "noul", "noul": 0.9584757468278602, "probabilities": {"0": 0.030182973564172867, "1": 0.002524917507227017, "2": 0.0025589614176987035, "3": 0.0027545758241444727, "4": 0.004299865085436147, "5": 0.004461185952011727, "6": 0.0033505931151902356, "7": 0.00664496225807132, "8": 0.9432219652760475}, "confidence": 0.8770295933535222}}, "usage": {"input_tokens": 10996, "output_tokens": 0, "cached_input_tokens": 10907}}, "kev_noul": 0.9584757468278602, "choice": "yes", "probs": {"yes": 0.9584757468278602, "no": 0.04152425317213981}, "confidence": 0.8770295933535222, "seconds": 3.1407335359999706, "prompt_tokens": 10996, "eval_tokens": 0, "server_latency_ms": null, "prompt_eval_ms": null, "total_ms": 3140.7335359999706, "done_reason": null, "id": "on-36", "task": "ka", "slug": "august-arrivals", "label": "yes", "kind": "on-page", "state_sha256": "4e79d2cabc185d243c590fe6100c9ba3cd0a34262da8edab751ca5859dfa0d9b", "prompt_sha256_matches_kev9b": null, "kev9b_refused": "kev-422", "card": "RTX 3090 (fp 33a0b4acb8e8)", "box": "benchbox", "os": {"pretty_name": "Ubuntu 26.04 LTS", "id": "ubuntu", "version_id": "26.04", "kernel": "7.0.0-31-generic", "arch": "x86_64", "source": "/etc/os-release"}, "dry": false, "runtime": "lev serve 0.1.1 (git 0ea955f5aee0) over HTTP on 127.0.0.1, one question per request", "lev_revision": "f8ef71157ec06a7d3b6435bc0756f9d735c33748", "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "lev_commit": "0ea955f5aee0d56d906121d1779af20fa8b922ed", "torch": "2.14.0+cu130", "torch_cuda": "13.0", "transformers": "5.17.0", "peft": "0.21.0", "dtype": "bfloat16", "kernels": "reference (no flash-linear-attention)", "arm": "lev-4b", "model": "interfaze-ai/lev@f8ef7115 on Qwen/Qwen3.5-4B@851bf6e8", "quant": "bf16 backbone + LoRA r=32 (PEFT, unmerged) + mode-B head fp32; calibrated as served; order_average on", "card_uuid": "GPU-fp:33a0b4acb8e8", "num_ctx": 0, "power_cap_w": 300, "seed": 20260921, "stamp": "2026-09-28T13:14:28Z"} {"mode": "kev-noul", "prompt_sha256": "d521a044697a9469f7dee687aa84b4ab159e1eb698c4756fb4c4c4bea9aef755", "n_options": 2, "kev_response": {"model": "Qwen/Qwen3.5-4B", "answers": {"q": {"type": "noul", "noul": 0.9508592110462791, "probabilities": {"0": 0.03551290467517707, "1": 0.00298449463834923, "2": 0.003099058924361914, "3": 0.0032328704511928132, "4": 0.005025401086243102, "5": 0.005688185670905093, "6": 0.00419418305566001, "7": 0.007818378488763161, "8": 0.9324445230093477}, "confidence": 0.8547391517954698}}, "usage": {"input_tokens": 7007, "output_tokens": 0, "cached_input_tokens": 6919}}, "kev_noul": 0.9508592110462791, "choice": "yes", "probs": {"yes": 0.9508592110462791, "no": 0.04914078895372087}, "confidence": 0.8547391517954698, "seconds": 1.980337402999794, "prompt_tokens": 7007, "eval_tokens": 0, "server_latency_ms": null, "prompt_eval_ms": null, "total_ms": 1980.337402999794, "done_reason": null, "id": "on-37", "task": "ka", "slug": "seat-trials", "label": "yes", "kind": "on-page", "state_sha256": "c9374a27b5e06de1440ffed7d81385fa68f7572790c0f553f8f2948c0218eea1", "prompt_sha256_matches_kev9b": true, "card": "RTX 3090 (fp 33a0b4acb8e8)", "box": "benchbox", "os": {"pretty_name": "Ubuntu 26.04 LTS", "id": "ubuntu", "version_id": "26.04", "kernel": "7.0.0-31-generic", "arch": "x86_64", "source": "/etc/os-release"}, "dry": false, "runtime": "lev serve 0.1.1 (git 0ea955f5aee0) over HTTP on 127.0.0.1, one question per request", "lev_revision": "f8ef71157ec06a7d3b6435bc0756f9d735c33748", "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "lev_commit": "0ea955f5aee0d56d906121d1779af20fa8b922ed", "torch": "2.14.0+cu130", "torch_cuda": "13.0", "transformers": "5.17.0", "peft": "0.21.0", "dtype": "bfloat16", "kernels": "reference (no flash-linear-attention)", "arm": "lev-4b", "model": "interfaze-ai/lev@f8ef7115 on Qwen/Qwen3.5-4B@851bf6e8", "quant": "bf16 backbone + LoRA r=32 (PEFT, unmerged) + mode-B head fp32; calibrated as served; order_average on", "card_uuid": "GPU-fp:33a0b4acb8e8", "num_ctx": 0, "power_cap_w": 300, "seed": 20260921, "stamp": "2026-09-28T13:14:30Z"} {"mode": "kev-noul", "prompt_sha256": "90c5a1d337de9dfabce1ad0966b52b7bac9b65938417d5534d5f2fcf62e95501", "n_options": 2, "kev_response": {"model": "Qwen/Qwen3.5-4B", "answers": {"q": {"type": "noul", "noul": 0.9510950942149733, "probabilities": {"0": 0.03600996096255001, "1": 0.0028600176397963017, "2": 0.0029034364458101545, "3": 0.0030376879684193615, "4": 0.004631954729773234, "5": 0.005199145849394842, "6": 0.0039641216980986545, "7": 0.008476876720807346, "8": 0.9329167979853501}, "confidence": 0.8557664097992614}}, "usage": {"input_tokens": 10010, "output_tokens": 0, "cached_input_tokens": 9921}}, "kev_noul": 0.9510950942149733, "choice": "yes", "probs": {"yes": 0.9510950942149733, "no": 0.04890490578502671}, "confidence": 0.8557664097992614, "seconds": 2.854291914998612, "prompt_tokens": 10010, "eval_tokens": 0, "server_latency_ms": null, "prompt_eval_ms": null, "total_ms": 2854.291914998612, "done_reason": null, "id": "on-38", "task": "ka", "slug": "voice-trials", "label": "yes", "kind": "on-page", "state_sha256": "34230a4e79273a8665f9112f22ddbdd39df22fdcb0190066ae293f8a9c7c34dc", "prompt_sha256_matches_kev9b": null, "kev9b_refused": "kev-422", "card": "RTX 3090 (fp 33a0b4acb8e8)", "box": "benchbox", "os": {"pretty_name": "Ubuntu 26.04 LTS", "id": "ubuntu", "version_id": "26.04", "kernel": "7.0.0-31-generic", "arch": "x86_64", "source": "/etc/os-release"}, "dry": false, "runtime": "lev serve 0.1.1 (git 0ea955f5aee0) over HTTP on 127.0.0.1, one question per request", "lev_revision": "f8ef71157ec06a7d3b6435bc0756f9d735c33748", "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "lev_commit": "0ea955f5aee0d56d906121d1779af20fa8b922ed", "torch": "2.14.0+cu130", "torch_cuda": "13.0", "transformers": "5.17.0", "peft": "0.21.0", "dtype": "bfloat16", "kernels": "reference (no flash-linear-attention)", "arm": "lev-4b", "model": "interfaze-ai/lev@f8ef7115 on Qwen/Qwen3.5-4B@851bf6e8", "quant": "bf16 backbone + LoRA r=32 (PEFT, unmerged) + mode-B head fp32; calibrated as served; order_average on", "card_uuid": "GPU-fp:33a0b4acb8e8", "num_ctx": 0, "power_cap_w": 300, "seed": 20260921, "stamp": "2026-09-28T13:14:33Z"} {"mode": "kev-noul", "prompt_sha256": "cfd0106359e94c981d662777f27744b65dbd2675d8e3da4f2f252fa40e2fec35", "n_options": 2, "kev_response": {"model": "Qwen/Qwen3.5-4B", "answers": {"q": {"type": "noul", "noul": 0.948270233391358, "probabilities": {"0": 0.03680424022636439, "1": 0.0032387455937093667, "2": 0.0035555834673277023, "3": 0.0033504254803911546, "4": 0.005662859573798588, "5": 0.006240303873988541, "6": 0.004420075858307062, "7": 0.008434862062560028, "8": 0.9282929038635532}, "confidence": 0.8461881075243042}}, "usage": {"input_tokens": 8184, "output_tokens": 0, "cached_input_tokens": 8097}}, "kev_noul": 0.948270233391358, "choice": "yes", "probs": {"yes": 0.948270233391358, "no": 0.051729766608642036}, "confidence": 0.8461881075243042, "seconds": 2.3083435219996318, "prompt_tokens": 8184, "eval_tokens": 0, "server_latency_ms": null, "prompt_eval_ms": null, "total_ms": 2308.343521999632, "done_reason": null, "id": "on-39", "task": "ka", "slug": "licences", "label": "yes", "kind": "on-page", "state_sha256": "cce674dd43eecdff402378e65d4ae1f1f64bb6db5702c2694bed4c5733c37090", "prompt_sha256_matches_kev9b": true, "card": "RTX 3090 (fp 33a0b4acb8e8)", "box": "benchbox", "os": {"pretty_name": "Ubuntu 26.04 LTS", "id": "ubuntu", "version_id": "26.04", "kernel": "7.0.0-31-generic", "arch": "x86_64", "source": "/etc/os-release"}, "dry": false, "runtime": "lev serve 0.1.1 (git 0ea955f5aee0) over HTTP on 127.0.0.1, one question per request", "lev_revision": "f8ef71157ec06a7d3b6435bc0756f9d735c33748", "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "lev_commit": "0ea955f5aee0d56d906121d1779af20fa8b922ed", "torch": "2.14.0+cu130", "torch_cuda": "13.0", "transformers": "5.17.0", "peft": "0.21.0", "dtype": "bfloat16", "kernels": "reference (no flash-linear-attention)", "arm": "lev-4b", "model": "interfaze-ai/lev@f8ef7115 on Qwen/Qwen3.5-4B@851bf6e8", "quant": "bf16 backbone + LoRA r=32 (PEFT, unmerged) + mode-B head fp32; calibrated as served; order_average on", "card_uuid": "GPU-fp:33a0b4acb8e8", "num_ctx": 0, "power_cap_w": 300, "seed": 20260921, "stamp": "2026-09-28T13:14:36Z"} {"refused": "overlong", "choice": null, "probs": null, "seconds": null, "why": "inherited from arm 2's own refusal rows so the denominator matches (GAPS-CARD.md, arm 12)", "id": "on-40", "task": "ka", "slug": "hear-it-for-yourself", "label": "yes", "kind": "on-page", "state_sha256": "679607114cbafcf5a158729b859012472889e4ec49b0c3ae61df66a3fafc6827", "prompt_sha256_matches_kev9b": null, "card": "RTX 3090 (fp 33a0b4acb8e8)", "box": "benchbox", "os": {"pretty_name": "Ubuntu 26.04 LTS", "id": "ubuntu", "version_id": "26.04", "kernel": "7.0.0-31-generic", "arch": "x86_64", "source": "/etc/os-release"}, "dry": false, "runtime": "lev serve 0.1.1 (git 0ea955f5aee0) over HTTP on 127.0.0.1, one question per request", "lev_revision": "f8ef71157ec06a7d3b6435bc0756f9d735c33748", "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "lev_commit": "0ea955f5aee0d56d906121d1779af20fa8b922ed", "torch": "2.14.0+cu130", "torch_cuda": "13.0", "transformers": "5.17.0", "peft": "0.21.0", "dtype": "bfloat16", "kernels": "reference (no flash-linear-attention)", "arm": "lev-4b", "model": "interfaze-ai/lev@f8ef7115 on Qwen/Qwen3.5-4B@851bf6e8", "quant": "bf16 backbone + LoRA r=32 (PEFT, unmerged) + mode-B head fp32; calibrated as served; order_average on", "card_uuid": "GPU-fp:33a0b4acb8e8", "num_ctx": 0, "power_cap_w": 300, "seed": 20260921, "stamp": "2026-09-28T13:14:36Z"} {"mode": "kev-noul", "prompt_sha256": "9b5a7fd2e981cc720b5a06acec3c5f820cd06bb034afab220cbcc050e5a7c0f1", "n_options": 2, "kev_response": {"model": "Qwen/Qwen3.5-4B", "answers": {"q": {"type": "noul", "noul": 0.9559907360553274, "probabilities": {"0": 0.03218411618427866, "1": 0.0026637391818267247, "2": 0.0028351430939950453, "3": 0.0028685634435203894, "4": 0.004478735763424019, "5": 0.0045543470440282076, "6": 0.003220498027877598, "7": 0.006582351787256639, "8": 0.9406125054737927}, "confidence": 0.8716437607385187}}, "usage": {"input_tokens": 9004, "output_tokens": 0, "cached_input_tokens": 8907}}, "kev_noul": 0.9559907360553274, "choice": "yes", "probs": {"yes": 0.9559907360553274, "no": 0.04400926394467264}, "confidence": 0.8716437607385187, "seconds": 2.568019472000742, "prompt_tokens": 9004, "eval_tokens": 0, "server_latency_ms": null, "prompt_eval_ms": null, "total_ms": 2568.019472000742, "done_reason": null, "id": "on-41", "task": "ka", "slug": "the-ceiling-is-not-the-corpus", "label": "yes", "kind": "on-page", "state_sha256": "f4bb84be11307889eba78ab0a1e5af28c73807a48012d5c6850a848cece69cb8", "prompt_sha256_matches_kev9b": null, "kev9b_refused": "kev-422", "card": "RTX 3090 (fp 33a0b4acb8e8)", "box": "benchbox", "os": {"pretty_name": "Ubuntu 26.04 LTS", "id": "ubuntu", "version_id": "26.04", "kernel": "7.0.0-31-generic", "arch": "x86_64", "source": "/etc/os-release"}, "dry": false, "runtime": "lev serve 0.1.1 (git 0ea955f5aee0) over HTTP on 127.0.0.1, one question per request", "lev_revision": "f8ef71157ec06a7d3b6435bc0756f9d735c33748", "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "lev_commit": "0ea955f5aee0d56d906121d1779af20fa8b922ed", "torch": "2.14.0+cu130", "torch_cuda": "13.0", "transformers": "5.17.0", "peft": "0.21.0", "dtype": "bfloat16", "kernels": "reference (no flash-linear-attention)", "arm": "lev-4b", "model": "interfaze-ai/lev@f8ef7115 on Qwen/Qwen3.5-4B@851bf6e8", "quant": "bf16 backbone + LoRA r=32 (PEFT, unmerged) + mode-B head fp32; calibrated as served; order_average on", "card_uuid": "GPU-fp:33a0b4acb8e8", "num_ctx": 0, "power_cap_w": 300, "seed": 20260921, "stamp": "2026-09-28T13:14:38Z"} {"mode": "kev-noul", "prompt_sha256": "95467f4791fac437dacefc0e3b46bc5cb14a90a6a1946e9d21a675d2f575eaf6", "n_options": 2, "kev_response": {"model": "Qwen/Qwen3.5-4B", "answers": {"q": {"type": "noul", "noul": 0.8663948563542655, "probabilities": {"0": 0.08152767102022765, "1": 0.010158565824682851, "2": 0.010862133108746163, "3": 0.011850117548137869, "4": 0.017012547634073342, "5": 0.02079780579541346, "6": 0.02682451386398457, "7": 0.03699379818760518, "8": 0.7839728470171289}, "confidence": 0.5774858479389318}}, "usage": {"input_tokens": 6631, "output_tokens": 0, "cached_input_tokens": 6552}}, "kev_noul": 0.8663948563542655, "choice": "yes", "probs": {"yes": 0.8663948563542655, "no": 0.1336051436457345}, "confidence": 0.5774858479389318, "seconds": 1.8867136330009089, "prompt_tokens": 6631, "eval_tokens": 0, "server_latency_ms": null, "prompt_eval_ms": null, "total_ms": 1886.7136330009089, "done_reason": null, "id": "on-42", "task": "ka", "slug": "the-indifferent-prediction", "label": "yes", "kind": "on-page", "state_sha256": "de97acfa81114f48f10f062ee956f9beb81fce1b56537b3ca2a7ad894da658b6", "prompt_sha256_matches_kev9b": true, "card": "RTX 3090 (fp 33a0b4acb8e8)", "box": "benchbox", "os": {"pretty_name": "Ubuntu 26.04 LTS", "id": "ubuntu", "version_id": "26.04", "kernel": "7.0.0-31-generic", "arch": "x86_64", "source": "/etc/os-release"}, "dry": false, "runtime": "lev serve 0.1.1 (git 0ea955f5aee0) over HTTP on 127.0.0.1, one question per request", "lev_revision": "f8ef71157ec06a7d3b6435bc0756f9d735c33748", "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "lev_commit": "0ea955f5aee0d56d906121d1779af20fa8b922ed", "torch": "2.14.0+cu130", "torch_cuda": "13.0", "transformers": "5.17.0", "peft": "0.21.0", "dtype": "bfloat16", "kernels": "reference (no flash-linear-attention)", "arm": "lev-4b", "model": "interfaze-ai/lev@f8ef7115 on Qwen/Qwen3.5-4B@851bf6e8", "quant": "bf16 backbone + LoRA r=32 (PEFT, unmerged) + mode-B head fp32; calibrated as served; order_average on", "card_uuid": "GPU-fp:33a0b4acb8e8", "num_ctx": 0, "power_cap_w": 300, "seed": 20260921, "stamp": "2026-09-28T13:14:40Z"} {"mode": "kev-noul", "prompt_sha256": "8d600d4f9e1d55c8f6ae4216cc3c3edd312644ed4004b2522ff7dfdc90a4b634", "n_options": 2, "kev_response": {"model": "Qwen/Qwen3.5-4B", "answers": {"q": {"type": "noul", "noul": 0.058566312888400814, "probabilities": {"0": 0.8942037920562541, "1": 0.020748540927635898, "2": 0.015661744389935336, "3": 0.010909213365428539, "4": 0.009414807588135192, "5": 0.007805124387352324, "6": 0.007805124387352324, "7": 0.007397988493251467, "8": 0.026053664404654713}, "confidence": 0.7765066304312285}}, "usage": {"input_tokens": 6953, "output_tokens": 0, "cached_input_tokens": 6867}}, "kev_noul": 0.058566312888400814, "choice": "no", "probs": {"yes": 0.058566312888400814, "no": 0.9414336871115991}, "confidence": 0.7765066304312285, "seconds": 1.9717709430005925, "prompt_tokens": 6953, "eval_tokens": 0, "server_latency_ms": null, "prompt_eval_ms": null, "total_ms": 1971.7709430005925, "done_reason": null, "id": "off-01", "task": "ka", "slug": "what-150-watts-buys", "label": "no", "kind": "off-page", "state_sha256": "0fd39e7ea02081243919348e7a76a2acba2433694142679ddfd8fbaa29a3d00d", "prompt_sha256_matches_kev9b": true, "card": "RTX 3090 (fp 33a0b4acb8e8)", "box": "benchbox", "os": {"pretty_name": "Ubuntu 26.04 LTS", "id": "ubuntu", "version_id": "26.04", "kernel": "7.0.0-31-generic", "arch": "x86_64", "source": "/etc/os-release"}, "dry": false, "runtime": "lev serve 0.1.1 (git 0ea955f5aee0) over HTTP on 127.0.0.1, one question per request", "lev_revision": "f8ef71157ec06a7d3b6435bc0756f9d735c33748", "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "lev_commit": "0ea955f5aee0d56d906121d1779af20fa8b922ed", "torch": "2.14.0+cu130", "torch_cuda": "13.0", "transformers": "5.17.0", "peft": "0.21.0", "dtype": "bfloat16", "kernels": "reference (no flash-linear-attention)", "arm": "lev-4b", "model": "interfaze-ai/lev@f8ef7115 on Qwen/Qwen3.5-4B@851bf6e8", "quant": "bf16 backbone + LoRA r=32 (PEFT, unmerged) + mode-B head fp32; calibrated as served; order_average on", "card_uuid": "GPU-fp:33a0b4acb8e8", "num_ctx": 0, "power_cap_w": 300, "seed": 20260921, "stamp": "2026-09-28T13:14:42Z"} {"mode": "kev-noul", "prompt_sha256": "552e46aa2217dd8fdb55e81b1f8782bdbc1c7fb1ef2a23af452d4bb037cbea32", "n_options": 2, "kev_response": {"model": "Qwen/Qwen3.5-4B", "answers": {"q": {"type": "noul", "noul": 0.05787351329855172, "probabilities": {"0": 0.8995087661491383, "1": 0.018012519644898677, "2": 0.015338254375377653, "3": 0.009470662093325442, "4": 0.008173317209421186, "5": 0.007746975405667252, "6": 0.006959849089422553, "7": 0.005617396951765151, "8": 0.029172259080983885}, "confidence": 0.7871761684233595}}, "usage": {"input_tokens": 4922, "output_tokens": 0, "cached_input_tokens": 4836}}, "kev_noul": 0.05787351329855172, "choice": "no", "probs": {"yes": 0.05787351329855172, "no": 0.9421264867014483}, "confidence": 0.7871761684233595, "seconds": 1.396528359999138, "prompt_tokens": 4922, "eval_tokens": 0, "server_latency_ms": null, "prompt_eval_ms": null, "total_ms": 1396.528359999138, "done_reason": null, "id": "off-02", "task": "ka", "slug": "how-we-work", "label": "no", "kind": "off-page", "state_sha256": "77d23d89a5adf6c0d561223d2c37308b5118a80f1f07f35f78bb1b0d08276e99", "prompt_sha256_matches_kev9b": true, "card": "RTX 3090 (fp 33a0b4acb8e8)", "box": "benchbox", "os": {"pretty_name": "Ubuntu 26.04 LTS", "id": "ubuntu", "version_id": "26.04", "kernel": "7.0.0-31-generic", "arch": "x86_64", "source": "/etc/os-release"}, "dry": false, "runtime": "lev serve 0.1.1 (git 0ea955f5aee0) over HTTP on 127.0.0.1, one question per request", "lev_revision": "f8ef71157ec06a7d3b6435bc0756f9d735c33748", "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "lev_commit": "0ea955f5aee0d56d906121d1779af20fa8b922ed", "torch": "2.14.0+cu130", "torch_cuda": "13.0", "transformers": "5.17.0", "peft": "0.21.0", "dtype": "bfloat16", "kernels": "reference (no flash-linear-attention)", "arm": "lev-4b", "model": "interfaze-ai/lev@f8ef7115 on Qwen/Qwen3.5-4B@851bf6e8", "quant": "bf16 backbone + LoRA r=32 (PEFT, unmerged) + mode-B head fp32; calibrated as served; order_average on", "card_uuid": "GPU-fp:33a0b4acb8e8", "num_ctx": 0, "power_cap_w": 300, "seed": 20260921, "stamp": "2026-09-28T13:14:43Z"} {"mode": "kev-noul", "prompt_sha256": "92846aa94ff8202ba46abcdcff4cc2605ea586391ee0ccb8ca610ce4dfdd29bf", "n_options": 2, "kev_response": {"model": "Qwen/Qwen3.5-4B", "answers": {"q": {"type": "noul", "noul": 0.04397451411071623, "probabilities": {"0": 0.9222114337026209, "1": 0.014127609559188733, "2": 0.01187007896699359, "3": 0.0076297032460445275, "4": 0.0070405790798295015, "5": 0.005995285224128599, "6": 0.005532361923577282, "7": 0.0050372647079399355, "8": 0.02055568358967701}, "confidence": 0.8328662417583257}}, "usage": {"input_tokens": 11998, "output_tokens": 0, "cached_input_tokens": 11914}}, "kev_noul": 0.04397451411071623, "choice": "no", "probs": {"yes": 0.04397451411071623, "no": 0.9560254858892838}, "confidence": 0.8328662417583257, "seconds": 3.4495904729992617, "prompt_tokens": 11998, "eval_tokens": 0, "server_latency_ms": null, "prompt_eval_ms": null, "total_ms": 3449.5904729992617, "done_reason": null, "id": "off-03", "task": "ka", "slug": "llms-txt", "label": "no", "kind": "off-page", "state_sha256": "96cab65488f086e43a1b03248954f7a7abadb4ae49c8591b267586825532ff41", "prompt_sha256_matches_kev9b": null, "kev9b_refused": "kev-422", "card": "RTX 3090 (fp 33a0b4acb8e8)", "box": "benchbox", "os": {"pretty_name": "Ubuntu 26.04 LTS", "id": "ubuntu", "version_id": "26.04", "kernel": "7.0.0-31-generic", "arch": "x86_64", "source": "/etc/os-release"}, "dry": false, "runtime": "lev serve 0.1.1 (git 0ea955f5aee0) over HTTP on 127.0.0.1, one question per request", "lev_revision": "f8ef71157ec06a7d3b6435bc0756f9d735c33748", "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "lev_commit": "0ea955f5aee0d56d906121d1779af20fa8b922ed", "torch": "2.14.0+cu130", "torch_cuda": "13.0", "transformers": "5.17.0", "peft": "0.21.0", "dtype": "bfloat16", "kernels": "reference (no flash-linear-attention)", "arm": "lev-4b", "model": "interfaze-ai/lev@f8ef7115 on Qwen/Qwen3.5-4B@851bf6e8", "quant": "bf16 backbone + LoRA r=32 (PEFT, unmerged) + mode-B head fp32; calibrated as served; order_average on", "card_uuid": "GPU-fp:33a0b4acb8e8", "num_ctx": 0, "power_cap_w": 300, "seed": 20260921, "stamp": "2026-09-28T13:14:47Z"} {"refused": "overlong", "choice": null, "probs": null, "seconds": null, "why": "inherited from arm 2's own refusal rows so the denominator matches (GAPS-CARD.md, arm 12)", "id": "off-04", "task": "ka", "slug": "three-new-voices-at-the-narrators-chair", "label": "no", "kind": "off-page", "state_sha256": "d944ffba7d7a97d70ffb305031ae8069d7e1bf47491eb08abdaea5d81af2ffe1", "prompt_sha256_matches_kev9b": null, "card": "RTX 3090 (fp 33a0b4acb8e8)", "box": "benchbox", "os": {"pretty_name": "Ubuntu 26.04 LTS", "id": "ubuntu", "version_id": "26.04", "kernel": "7.0.0-31-generic", "arch": "x86_64", "source": "/etc/os-release"}, "dry": false, "runtime": "lev serve 0.1.1 (git 0ea955f5aee0) over HTTP on 127.0.0.1, one question per request", "lev_revision": "f8ef71157ec06a7d3b6435bc0756f9d735c33748", "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "lev_commit": "0ea955f5aee0d56d906121d1779af20fa8b922ed", "torch": "2.14.0+cu130", "torch_cuda": "13.0", "transformers": "5.17.0", "peft": "0.21.0", "dtype": "bfloat16", "kernels": "reference (no flash-linear-attention)", "arm": "lev-4b", "model": "interfaze-ai/lev@f8ef7115 on Qwen/Qwen3.5-4B@851bf6e8", "quant": "bf16 backbone + LoRA r=32 (PEFT, unmerged) + mode-B head fp32; calibrated as served; order_average on", "card_uuid": "GPU-fp:33a0b4acb8e8", "num_ctx": 0, "power_cap_w": 300, "seed": 20260921, "stamp": "2026-09-28T13:14:47Z"} {"mode": "kev-noul", "prompt_sha256": "20e3610dcf34195439f1acc62acc6349f52980cd8e9b8c61155d5632aa3e2edf", "n_options": 2, "kev_response": {"model": "Qwen/Qwen3.5-4B", "answers": {"q": {"type": "noul", "noul": 0.09249941417456849, "probabilities": {"0": 0.8507053948711487, "1": 0.027222455996113736, "2": 0.021679339750965, "3": 0.011246965976088201, "4": 0.009199987070232134, "5": 0.0069444808860158825, "6": 0.0068520929023906905, "7": 0.006155890531478338, "8": 0.05999339201556739}, "confidence": 0.694960905043926}}, "usage": {"input_tokens": 10708, "output_tokens": 0, "cached_input_tokens": 10621}}, "kev_noul": 0.09249941417456849, "choice": "no", "probs": {"yes": 0.09249941417456849, "no": 0.9075005858254315}, "confidence": 0.694960905043926, "seconds": 3.072214110999994, "prompt_tokens": 10708, "eval_tokens": 0, "server_latency_ms": null, "prompt_eval_ms": null, "total_ms": 3072.214110999994, "done_reason": null, "id": "off-05", "task": "ka", "slug": "two-hours-on-battery", "label": "no", "kind": "off-page", "state_sha256": "c4c4db33a200bbafecb6bc46874d79161edbca9f5ee4e5451aac9225077b64f1", "prompt_sha256_matches_kev9b": null, "kev9b_refused": "kev-422", "card": "RTX 3090 (fp 33a0b4acb8e8)", "box": "benchbox", "os": {"pretty_name": "Ubuntu 26.04 LTS", "id": "ubuntu", "version_id": "26.04", "kernel": "7.0.0-31-generic", "arch": "x86_64", "source": "/etc/os-release"}, "dry": false, "runtime": "lev serve 0.1.1 (git 0ea955f5aee0) over HTTP on 127.0.0.1, one question per request", "lev_revision": "f8ef71157ec06a7d3b6435bc0756f9d735c33748", "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "lev_commit": "0ea955f5aee0d56d906121d1779af20fa8b922ed", "torch": "2.14.0+cu130", "torch_cuda": "13.0", "transformers": "5.17.0", "peft": "0.21.0", "dtype": "bfloat16", "kernels": "reference (no flash-linear-attention)", "arm": "lev-4b", "model": "interfaze-ai/lev@f8ef7115 on Qwen/Qwen3.5-4B@851bf6e8", "quant": "bf16 backbone + LoRA r=32 (PEFT, unmerged) + mode-B head fp32; calibrated as served; order_average on", "card_uuid": "GPU-fp:33a0b4acb8e8", "num_ctx": 0, "power_cap_w": 300, "seed": 20260921, "stamp": "2026-09-28T13:14:50Z"} {"mode": "kev-noul", "prompt_sha256": "9b62a1b35b3173c70b56c9acb97460e2a7e69d86a505e5032218cb9966259975", "n_options": 2, "kev_response": {"model": "Qwen/Qwen3.5-4B", "answers": {"q": {"type": "noul", "noul": 0.05488566391310394, "probabilities": {"0": 0.8800212718021353, "1": 0.030928374650949636, "2": 0.02125660280100625, "3": 0.013847278921486482, "4": 0.011176343251754865, "5": 0.010452423797935536, "6": 0.010880945698040694, "7": 0.009645344511060616, "8": 0.011791414565630539}, "confidence": 0.7487000001979507}}, "usage": {"input_tokens": 5216, "output_tokens": 0, "cached_input_tokens": 5129}}, "kev_noul": 0.05488566391310394, "choice": "no", "probs": {"yes": 0.05488566391310394, "no": 0.9451143360868961}, "confidence": 0.7487000001979507, "seconds": 1.4588845870002842, "prompt_tokens": 5216, "eval_tokens": 0, "server_latency_ms": null, "prompt_eval_ms": null, "total_ms": 1458.8845870002842, "done_reason": null, "id": "off-06", "task": "ka", "slug": "reading-is-fast", "label": "no", "kind": "off-page", "state_sha256": "79b9c9e1be3ac3cdc484fec1f5eb9a83bf909612d6eb1c3c7c83bb469125ff2f", "prompt_sha256_matches_kev9b": true, "card": "RTX 3090 (fp 33a0b4acb8e8)", "box": "benchbox", "os": {"pretty_name": "Ubuntu 26.04 LTS", "id": "ubuntu", "version_id": "26.04", "kernel": "7.0.0-31-generic", "arch": "x86_64", "source": "/etc/os-release"}, "dry": false, "runtime": "lev serve 0.1.1 (git 0ea955f5aee0) over HTTP on 127.0.0.1, one question per request", "lev_revision": "f8ef71157ec06a7d3b6435bc0756f9d735c33748", "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "lev_commit": "0ea955f5aee0d56d906121d1779af20fa8b922ed", "torch": "2.14.0+cu130", "torch_cuda": "13.0", "transformers": "5.17.0", "peft": "0.21.0", "dtype": "bfloat16", "kernels": "reference (no flash-linear-attention)", "arm": "lev-4b", "model": "interfaze-ai/lev@f8ef7115 on Qwen/Qwen3.5-4B@851bf6e8", "quant": "bf16 backbone + LoRA r=32 (PEFT, unmerged) + mode-B head fp32; calibrated as served; order_average on", "card_uuid": "GPU-fp:33a0b4acb8e8", "num_ctx": 0, "power_cap_w": 300, "seed": 20260921, "stamp": "2026-09-28T13:14:51Z"} {"mode": "kev-noul", "prompt_sha256": "84543fe40f7be55f38cedd0c0f4e77d5730b7aeb7551a6fd8294abf7816ebddf", "n_options": 2, "kev_response": {"model": "Qwen/Qwen3.5-4B", "answers": {"q": {"type": "noul", "noul": 0.05738139900226249, "probabilities": {"0": 0.9072175206043288, "1": 0.01567828335048461, "2": 0.011521737870626967, "3": 0.008025485133825939, "4": 0.00701949472012077, "5": 0.006139604684846456, "6": 0.006057924613088901, "7": 0.005590164639785921, "8": 0.03274978438289167}, "confidence": 0.8028033196776987}}, "usage": {"input_tokens": 4440, "output_tokens": 0, "cached_input_tokens": 4354}}, "kev_noul": 0.05738139900226249, "choice": "no", "probs": {"yes": 0.05738139900226249, "no": 0.9426186009977375}, "confidence": 0.8028033196776987, "seconds": 1.2720765550002398, "prompt_tokens": 4440, "eval_tokens": 0, "server_latency_ms": null, "prompt_eval_ms": null, "total_ms": 1272.0765550002398, "done_reason": null, "id": "off-07", "task": "ka", "slug": "a-rig-your-friend-already-owns", "label": "no", "kind": "off-page", "state_sha256": "9d22ebe41036f74e0c8501e4631982f4d610eebb419288b4535c8a5e9f9b65d4", "prompt_sha256_matches_kev9b": true, "card": "RTX 3090 (fp 33a0b4acb8e8)", "box": "benchbox", "os": {"pretty_name": "Ubuntu 26.04 LTS", "id": "ubuntu", "version_id": "26.04", "kernel": "7.0.0-31-generic", "arch": "x86_64", "source": "/etc/os-release"}, "dry": false, "runtime": "lev serve 0.1.1 (git 0ea955f5aee0) over HTTP on 127.0.0.1, one question per request", "lev_revision": "f8ef71157ec06a7d3b6435bc0756f9d735c33748", "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "lev_commit": "0ea955f5aee0d56d906121d1779af20fa8b922ed", "torch": "2.14.0+cu130", "torch_cuda": "13.0", "transformers": "5.17.0", "peft": "0.21.0", "dtype": "bfloat16", "kernels": "reference (no flash-linear-attention)", "arm": "lev-4b", "model": "interfaze-ai/lev@f8ef7115 on Qwen/Qwen3.5-4B@851bf6e8", "quant": "bf16 backbone + LoRA r=32 (PEFT, unmerged) + mode-B head fp32; calibrated as served; order_average on", "card_uuid": "GPU-fp:33a0b4acb8e8", "num_ctx": 0, "power_cap_w": 300, "seed": 20260921, "stamp": "2026-09-28T13:14:53Z"} {"mode": "kev-noul", "prompt_sha256": "713661cd57354fb28ebff6d4bfea3fe674b6a1f304124e9af12717cf96e1acbe", "n_options": 2, "kev_response": {"model": "Qwen/Qwen3.5-4B", "answers": {"q": {"type": "noul", "noul": 0.0805034761810293, "probabilities": {"0": 0.8721254639077016, "1": 0.02282858116335067, "2": 0.015901285102126718, "3": 0.010084856397197264, "4": 0.008249384647968198, "5": 0.006838957959974595, "6": 0.006838957959974595, "7": 0.006144090156205769, "8": 0.050988422705500594}, "confidence": 0.7348123992739102}}, "usage": {"input_tokens": 5828, "output_tokens": 0, "cached_input_tokens": 5741}}, "kev_noul": 0.0805034761810293, "choice": "no", "probs": {"yes": 0.0805034761810293, "no": 0.9194965238189707}, "confidence": 0.7348123992739102, "seconds": 1.6549108520011941, "prompt_tokens": 5828, "eval_tokens": 0, "server_latency_ms": null, "prompt_eval_ms": null, "total_ms": 1654.9108520011941, "done_reason": null, "id": "off-08", "task": "ka", "slug": "the-move", "label": "no", "kind": "off-page", "state_sha256": "3d1d2fe049acbdd5e286fbc87c126a77f0410d05b6a51694ef22f6658b409e9f", "prompt_sha256_matches_kev9b": true, "card": "RTX 3090 (fp 33a0b4acb8e8)", "box": "benchbox", "os": {"pretty_name": "Ubuntu 26.04 LTS", "id": "ubuntu", "version_id": "26.04", "kernel": "7.0.0-31-generic", "arch": "x86_64", "source": "/etc/os-release"}, "dry": false, "runtime": "lev serve 0.1.1 (git 0ea955f5aee0) over HTTP on 127.0.0.1, one question per request", "lev_revision": "f8ef71157ec06a7d3b6435bc0756f9d735c33748", "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "lev_commit": "0ea955f5aee0d56d906121d1779af20fa8b922ed", "torch": "2.14.0+cu130", "torch_cuda": "13.0", "transformers": "5.17.0", "peft": "0.21.0", "dtype": "bfloat16", "kernels": "reference (no flash-linear-attention)", "arm": "lev-4b", "model": "interfaze-ai/lev@f8ef7115 on Qwen/Qwen3.5-4B@851bf6e8", "quant": "bf16 backbone + LoRA r=32 (PEFT, unmerged) + mode-B head fp32; calibrated as served; order_average on", "card_uuid": "GPU-fp:33a0b4acb8e8", "num_ctx": 0, "power_cap_w": 300, "seed": 20260921, "stamp": "2026-09-28T13:14:54Z"} {"mode": "kev-noul", "prompt_sha256": "2a2695e3876d1b582ebab9d92a35db85ae1f131c90e682872ae735ead287cd2b", "n_options": 2, "kev_response": {"model": "Qwen/Qwen3.5-4B", "answers": {"q": {"type": "noul", "noul": 0.032007362499352196, "probabilities": {"0": 0.9438446931671115, "1": 0.009937461972529671, "2": 0.007913966851320155, "3": 0.005973746605512287, "4": 0.005086842197301682, "5": 0.004569996957654861, "6": 0.003997150200929148, "7": 0.0042171266610703416, "8": 0.0144590153865703}, "confidence": 0.8777456422960281}}, "usage": {"input_tokens": 5123, "output_tokens": 0, "cached_input_tokens": 5037}}, "kev_noul": 0.032007362499352196, "choice": "no", "probs": {"yes": 0.032007362499352196, "no": 0.9679926375006478}, "confidence": 0.8777456422960281, "seconds": 1.446449388999099, "prompt_tokens": 5123, "eval_tokens": 0, "server_latency_ms": null, "prompt_eval_ms": null, "total_ms": 1446.449388999099, "done_reason": null, "id": "off-09", "task": "ka", "slug": "fifty-seven-milliseconds", "label": "no", "kind": "off-page", "state_sha256": "bec5d08ff12a12a4d87ca9d1d31e3bec848dfe329b6a47e9b8616e071da4ba5c", "prompt_sha256_matches_kev9b": true, "card": "RTX 3090 (fp 33a0b4acb8e8)", "box": "benchbox", "os": {"pretty_name": "Ubuntu 26.04 LTS", "id": "ubuntu", "version_id": "26.04", "kernel": "7.0.0-31-generic", "arch": "x86_64", "source": "/etc/os-release"}, "dry": false, "runtime": "lev serve 0.1.1 (git 0ea955f5aee0) over HTTP on 127.0.0.1, one question per request", "lev_revision": "f8ef71157ec06a7d3b6435bc0756f9d735c33748", "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "lev_commit": "0ea955f5aee0d56d906121d1779af20fa8b922ed", "torch": "2.14.0+cu130", "torch_cuda": "13.0", "transformers": "5.17.0", "peft": "0.21.0", "dtype": "bfloat16", "kernels": "reference (no flash-linear-attention)", "arm": "lev-4b", "model": "interfaze-ai/lev@f8ef7115 on Qwen/Qwen3.5-4B@851bf6e8", "quant": "bf16 backbone + LoRA r=32 (PEFT, unmerged) + mode-B head fp32; calibrated as served; order_average on", "card_uuid": "GPU-fp:33a0b4acb8e8", "num_ctx": 0, "power_cap_w": 300, "seed": 20260921, "stamp": "2026-09-28T13:14:56Z"} {"mode": "kev-noul", "prompt_sha256": "9b0b371518ff7abb1bcdef463b5d83200d64ad49e561aa94ba78125be9665ff7", "n_options": 2, "kev_response": {"model": "Qwen/Qwen3.5-4B", "answers": {"q": {"type": "noul", "noul": 0.04183845145367711, "probabilities": {"0": 0.9228305010693414, "1": 0.014327706156467637, "2": 0.012038200997884678, "3": 0.00849828788061717, "4": 0.007737766655596235, "5": 0.006588963432134972, "6": 0.00576304027129774, "7": 0.005389753868799172, "8": 0.016825779667860988}, "confidence": 0.83404811401869}}, "usage": {"input_tokens": 2910, "output_tokens": 0, "cached_input_tokens": 2823}}, "kev_noul": 0.04183845145367711, "choice": "no", "probs": {"yes": 0.04183845145367711, "no": 0.9581615485463228}, "confidence": 0.83404811401869, "seconds": 0.8336731849994976, "prompt_tokens": 2910, "eval_tokens": 0, "server_latency_ms": null, "prompt_eval_ms": null, "total_ms": 833.6731849994976, "done_reason": null, "id": "off-10", "task": "ka", "slug": "six-worlds-painted-large", "label": "no", "kind": "off-page", "state_sha256": "4902422ba42eec05800d9a72fd69c58559d0cb098878daec40505c90f50184d4", "prompt_sha256_matches_kev9b": true, "card": "RTX 3090 (fp 33a0b4acb8e8)", "box": "benchbox", "os": {"pretty_name": "Ubuntu 26.04 LTS", "id": "ubuntu", "version_id": "26.04", "kernel": "7.0.0-31-generic", "arch": "x86_64", "source": "/etc/os-release"}, "dry": false, "runtime": "lev serve 0.1.1 (git 0ea955f5aee0) over HTTP on 127.0.0.1, one question per request", "lev_revision": "f8ef71157ec06a7d3b6435bc0756f9d735c33748", "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "lev_commit": "0ea955f5aee0d56d906121d1779af20fa8b922ed", "torch": "2.14.0+cu130", "torch_cuda": "13.0", "transformers": "5.17.0", "peft": "0.21.0", "dtype": "bfloat16", "kernels": "reference (no flash-linear-attention)", "arm": "lev-4b", "model": "interfaze-ai/lev@f8ef7115 on Qwen/Qwen3.5-4B@851bf6e8", "quant": "bf16 backbone + LoRA r=32 (PEFT, unmerged) + mode-B head fp32; calibrated as served; order_average on", "card_uuid": "GPU-fp:33a0b4acb8e8", "num_ctx": 0, "power_cap_w": 300, "seed": 20260921, "stamp": "2026-09-28T13:14:57Z"} {"mode": "kev-noul", "prompt_sha256": "19d9e4f9f0d5b96afe38a6413aa876999067343b558556fca9a50d2df81ec057", "n_options": 2, "kev_response": {"model": "Qwen/Qwen3.5-4B", "answers": {"q": {"type": "noul", "noul": 0.03231985024039509, "probabilities": {"0": 0.9418294316713463, "1": 0.010603028303120137, "2": 0.00867324783425179, "3": 0.006041364795603358, "4": 0.005574883481293295, "5": 0.004876074718186066, "6": 0.0046840413360446495, "7": 0.004042394848879165, "8": 0.013675533011275174}, "confidence": 0.8734903555612797}}, "usage": {"input_tokens": 10053, "output_tokens": 0, "cached_input_tokens": 9968}}, "kev_noul": 0.03231985024039509, "choice": "no", "probs": {"yes": 0.03231985024039509, "no": 0.9676801497596049}, "confidence": 0.8734903555612797, "seconds": 2.8747200569996494, "prompt_tokens": 10053, "eval_tokens": 0, "server_latency_ms": null, "prompt_eval_ms": null, "total_ms": 2874.7200569996494, "done_reason": null, "id": "off-11", "task": "ka", "slug": "outside-judges", "label": "no", "kind": "off-page", "state_sha256": "fc7235d0fabb2221f559392fc0236d981f9898d96425df0d7355f2d6f11d17d6", "prompt_sha256_matches_kev9b": null, "kev9b_refused": "kev-422", "card": "RTX 3090 (fp 33a0b4acb8e8)", "box": "benchbox", "os": {"pretty_name": "Ubuntu 26.04 LTS", "id": "ubuntu", "version_id": "26.04", "kernel": "7.0.0-31-generic", "arch": "x86_64", "source": "/etc/os-release"}, "dry": false, "runtime": "lev serve 0.1.1 (git 0ea955f5aee0) over HTTP on 127.0.0.1, one question per request", "lev_revision": "f8ef71157ec06a7d3b6435bc0756f9d735c33748", "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "lev_commit": "0ea955f5aee0d56d906121d1779af20fa8b922ed", "torch": "2.14.0+cu130", "torch_cuda": "13.0", "transformers": "5.17.0", "peft": "0.21.0", "dtype": "bfloat16", "kernels": "reference (no flash-linear-attention)", "arm": "lev-4b", "model": "interfaze-ai/lev@f8ef7115 on Qwen/Qwen3.5-4B@851bf6e8", "quant": "bf16 backbone + LoRA r=32 (PEFT, unmerged) + mode-B head fp32; calibrated as served; order_average on", "card_uuid": "GPU-fp:33a0b4acb8e8", "num_ctx": 0, "power_cap_w": 300, "seed": 20260921, "stamp": "2026-09-28T13:14:59Z"} {"mode": "kev-noul", "prompt_sha256": "f509dfbc3ebb640055c777670874e04b146aa31d974e7416edf8428ae2de8b83", "n_options": 2, "kev_response": {"model": "Qwen/Qwen3.5-4B", "answers": {"q": {"type": "noul", "noul": 0.46499202974928217, "probabilities": {"0": 0.13435299555587332, "1": 0.11751191700074433, "2": 0.12070214823409507, "3": 0.11751191700074433, "4": 0.1084382862328265, "5": 0.11138218517057244, "6": 0.10278185893850846, "7": 0.09742048582521295, "8": 0.08989820604142254}, "confidence": 0.0016066443326137048}}, "usage": {"input_tokens": 5225, "output_tokens": 0, "cached_input_tokens": 5143}}, "kev_noul": 0.46499202974928217, "choice": "no", "probs": {"yes": 0.46499202974928217, "no": 0.5350079702507178}, "confidence": 0.0016066443326137048, "seconds": 1.4613661259991204, "prompt_tokens": 5225, "eval_tokens": 0, "server_latency_ms": null, "prompt_eval_ms": null, "total_ms": 1461.3661259991204, "done_reason": null, "id": "off-12", "task": "ka", "slug": "the-typist-and-the-developer", "label": "no", "kind": "off-page", "state_sha256": "2d33747167f415355e5637f99962534c215404a547456ccde9c80e2d7591cf5d", "prompt_sha256_matches_kev9b": true, "card": "RTX 3090 (fp 33a0b4acb8e8)", "box": "benchbox", "os": {"pretty_name": "Ubuntu 26.04 LTS", "id": "ubuntu", "version_id": "26.04", "kernel": "7.0.0-31-generic", "arch": "x86_64", "source": "/etc/os-release"}, "dry": false, "runtime": "lev serve 0.1.1 (git 0ea955f5aee0) over HTTP on 127.0.0.1, one question per request", "lev_revision": "f8ef71157ec06a7d3b6435bc0756f9d735c33748", "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "lev_commit": "0ea955f5aee0d56d906121d1779af20fa8b922ed", "torch": "2.14.0+cu130", "torch_cuda": "13.0", "transformers": "5.17.0", "peft": "0.21.0", "dtype": "bfloat16", "kernels": "reference (no flash-linear-attention)", "arm": "lev-4b", "model": "interfaze-ai/lev@f8ef7115 on Qwen/Qwen3.5-4B@851bf6e8", "quant": "bf16 backbone + LoRA r=32 (PEFT, unmerged) + mode-B head fp32; calibrated as served; order_average on", "card_uuid": "GPU-fp:33a0b4acb8e8", "num_ctx": 0, "power_cap_w": 300, "seed": 20260921, "stamp": "2026-09-28T13:15:01Z"} {"mode": "kev-noul", "prompt_sha256": "e9a0fe7c450713b43dc95f2c9b5af7257cf5c7de9b680dbaa37f90bbc64fe0f8", "n_options": 2, "kev_response": {"model": "Qwen/Qwen3.5-4B", "answers": {"q": {"type": "noul", "noul": 0.1619655636476634, "probabilities": {"0": 0.6734510913287429, "1": 0.07290540198487643, "2": 0.060440505626370794, "3": 0.03833236218967113, "4": 0.027795117867638306, "5": 0.024311014208309802, "6": 0.022433849984034488, "7": 0.023042887524481703, "8": 0.05728776928587434}, "confidence": 0.40336044574521124}}, "usage": {"input_tokens": 5195, "output_tokens": 0, "cached_input_tokens": 5112}}, "kev_noul": 0.1619655636476634, "choice": "no", "probs": {"yes": 0.1619655636476634, "no": 0.8380344363523367}, "confidence": 0.40336044574521124, "seconds": 1.4556425030004903, "prompt_tokens": 5195, "eval_tokens": 0, "server_latency_ms": null, "prompt_eval_ms": null, "total_ms": 1455.6425030004903, "done_reason": null, "id": "off-13", "task": "ka", "slug": "compression-is-the-objective", "label": "no", "kind": "off-page", "state_sha256": "68bc7c0890d5d05afefc1bbd6178315fddd18fb35775c738f50544096a982f7f", "prompt_sha256_matches_kev9b": true, "card": "RTX 3090 (fp 33a0b4acb8e8)", "box": "benchbox", "os": {"pretty_name": "Ubuntu 26.04 LTS", "id": "ubuntu", "version_id": "26.04", "kernel": "7.0.0-31-generic", "arch": "x86_64", "source": "/etc/os-release"}, "dry": false, "runtime": "lev serve 0.1.1 (git 0ea955f5aee0) over HTTP on 127.0.0.1, one question per request", "lev_revision": "f8ef71157ec06a7d3b6435bc0756f9d735c33748", "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "lev_commit": "0ea955f5aee0d56d906121d1779af20fa8b922ed", "torch": "2.14.0+cu130", "torch_cuda": "13.0", "transformers": "5.17.0", "peft": "0.21.0", "dtype": "bfloat16", "kernels": "reference (no flash-linear-attention)", "arm": "lev-4b", "model": "interfaze-ai/lev@f8ef7115 on Qwen/Qwen3.5-4B@851bf6e8", "quant": "bf16 backbone + LoRA r=32 (PEFT, unmerged) + mode-B head fp32; calibrated as served; order_average on", "card_uuid": "GPU-fp:33a0b4acb8e8", "num_ctx": 0, "power_cap_w": 300, "seed": 20260921, "stamp": "2026-09-28T13:15:02Z"} {"refused": "overlong", "choice": null, "probs": null, "seconds": null, "why": "inherited from arm 2's own refusal rows so the denominator matches (GAPS-CARD.md, arm 12)", "id": "off-14", "task": "ka", "slug": "the-open-call", "label": "no", "kind": "off-page", "state_sha256": "dd50fd254da5ef89e4284d245102a3ca9f9f7dd4e41f90c7687adb01223798a4", "prompt_sha256_matches_kev9b": null, "card": "RTX 3090 (fp 33a0b4acb8e8)", "box": "benchbox", "os": {"pretty_name": "Ubuntu 26.04 LTS", "id": "ubuntu", "version_id": "26.04", "kernel": "7.0.0-31-generic", "arch": "x86_64", "source": "/etc/os-release"}, "dry": false, "runtime": "lev serve 0.1.1 (git 0ea955f5aee0) over HTTP on 127.0.0.1, one question per request", "lev_revision": "f8ef71157ec06a7d3b6435bc0756f9d735c33748", "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "lev_commit": "0ea955f5aee0d56d906121d1779af20fa8b922ed", "torch": "2.14.0+cu130", "torch_cuda": "13.0", "transformers": "5.17.0", "peft": "0.21.0", "dtype": "bfloat16", "kernels": "reference (no flash-linear-attention)", "arm": "lev-4b", "model": "interfaze-ai/lev@f8ef7115 on Qwen/Qwen3.5-4B@851bf6e8", "quant": "bf16 backbone + LoRA r=32 (PEFT, unmerged) + mode-B head fp32; calibrated as served; order_average on", "card_uuid": "GPU-fp:33a0b4acb8e8", "num_ctx": 0, "power_cap_w": 300, "seed": 20260921, "stamp": "2026-09-28T13:15:02Z"} {"mode": "kev-noul", "prompt_sha256": "afd2a4577ff2913d759649def5f3b2644e5e08e6a1c1eb230628a8f33099cf68", "n_options": 2, "kev_response": {"model": "Qwen/Qwen3.5-4B", "answers": {"q": {"type": "noul", "noul": 0.07306330352654541, "probabilities": {"0": 0.8740866778576108, "1": 0.024139075408302724, "2": 0.020281756095135193, "3": 0.011250649808699548, "4": 0.008492394821354991, "5": 0.007231554196415443, "6": 0.006946755482134613, "7": 0.006325082615386514, "8": 0.04124605371496013}, "confidence": 0.7379448167079077}}, "usage": {"input_tokens": 14590, "output_tokens": 0, "cached_input_tokens": 14501}}, "kev_noul": 0.07306330352654541, "choice": "no", "probs": {"yes": 0.07306330352654541, "no": 0.9269366964734546}, "confidence": 0.7379448167079077, "seconds": 4.246197283000583, "prompt_tokens": 14590, "eval_tokens": 0, "server_latency_ms": null, "prompt_eval_ms": null, "total_ms": 4246.197283000583, "done_reason": null, "id": "off-15", "task": "ka", "slug": "the-new-kid", "label": "no", "kind": "off-page", "state_sha256": "676acfa90c7de336adc467067853954373df2cc40b45fb728501238f7b963b79", "prompt_sha256_matches_kev9b": null, "kev9b_refused": "kev-422", "card": "RTX 3090 (fp 33a0b4acb8e8)", "box": "benchbox", "os": {"pretty_name": "Ubuntu 26.04 LTS", "id": "ubuntu", "version_id": "26.04", "kernel": "7.0.0-31-generic", "arch": "x86_64", "source": "/etc/os-release"}, "dry": false, "runtime": "lev serve 0.1.1 (git 0ea955f5aee0) over HTTP on 127.0.0.1, one question per request", "lev_revision": "f8ef71157ec06a7d3b6435bc0756f9d735c33748", "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "lev_commit": "0ea955f5aee0d56d906121d1779af20fa8b922ed", "torch": "2.14.0+cu130", "torch_cuda": "13.0", "transformers": "5.17.0", "peft": "0.21.0", "dtype": "bfloat16", "kernels": "reference (no flash-linear-attention)", "arm": "lev-4b", "model": "interfaze-ai/lev@f8ef7115 on Qwen/Qwen3.5-4B@851bf6e8", "quant": "bf16 backbone + LoRA r=32 (PEFT, unmerged) + mode-B head fp32; calibrated as served; order_average on", "card_uuid": "GPU-fp:33a0b4acb8e8", "num_ctx": 0, "power_cap_w": 300, "seed": 20260921, "stamp": "2026-09-28T13:15:07Z"} {"mode": "kev-noul", "prompt_sha256": "b20457a7455573562b8b154b5d9aca5a954b6a2434349474b2a591f6214d0f38", "n_options": 2, "kev_response": {"model": "Qwen/Qwen3.5-4B", "answers": {"q": {"type": "noul", "noul": 0.0779986064487923, "probabilities": {"0": 0.8615557479467664, "1": 0.02869994963433132, "2": 0.021955849682796198, "3": 0.012509944662900827, "4": 0.010096961027923715, "5": 0.008370647622847805, "6": 0.007828460029039365, "7": 0.007223988945613364, "8": 0.04175845044778107}, "confidence": 0.7139910332462783}}, "usage": {"input_tokens": 4702, "output_tokens": 0, "cached_input_tokens": 4618}}, "kev_noul": 0.0779986064487923, "choice": "no", "probs": {"yes": 0.0779986064487923, "no": 0.9220013935512077}, "confidence": 0.7139910332462783, "seconds": 1.3433081080002012, "prompt_tokens": 4702, "eval_tokens": 0, "server_latency_ms": null, "prompt_eval_ms": null, "total_ms": 1343.3081080002012, "done_reason": null, "id": "off-16", "task": "ka", "slug": "where-your-question-goes", "label": "no", "kind": "off-page", "state_sha256": "cdc24e58a19354d77487d8df5e8782b2c15b4483afe7405f34478bf769d44b11", "prompt_sha256_matches_kev9b": true, "card": "RTX 3090 (fp 33a0b4acb8e8)", "box": "benchbox", "os": {"pretty_name": "Ubuntu 26.04 LTS", "id": "ubuntu", "version_id": "26.04", "kernel": "7.0.0-31-generic", "arch": "x86_64", "source": "/etc/os-release"}, "dry": false, "runtime": "lev serve 0.1.1 (git 0ea955f5aee0) over HTTP on 127.0.0.1, one question per request", "lev_revision": "f8ef71157ec06a7d3b6435bc0756f9d735c33748", "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "lev_commit": "0ea955f5aee0d56d906121d1779af20fa8b922ed", "torch": "2.14.0+cu130", "torch_cuda": "13.0", "transformers": "5.17.0", "peft": "0.21.0", "dtype": "bfloat16", "kernels": "reference (no flash-linear-attention)", "arm": "lev-4b", "model": "interfaze-ai/lev@f8ef7115 on Qwen/Qwen3.5-4B@851bf6e8", "quant": "bf16 backbone + LoRA r=32 (PEFT, unmerged) + mode-B head fp32; calibrated as served; order_average on", "card_uuid": "GPU-fp:33a0b4acb8e8", "num_ctx": 0, "power_cap_w": 300, "seed": 20260921, "stamp": "2026-09-28T13:15:08Z"} {"mode": "kev-noul", "prompt_sha256": "8f17af6313832b7a0ccf7340d2d3e7ccbbf39825e0127d3fca2cf9fda26773bf", "n_options": 2, "kev_response": {"model": "Qwen/Qwen3.5-4B", "answers": {"q": {"type": "noul", "noul": 0.5044575807055498, "probabilities": {"0": 0.10751716212147841, "1": 0.11043605424302375, "2": 0.11043605424302375, "3": 0.11043605424302375, "4": 0.11043605424302375, "5": 0.11343418888780092, "6": 0.11043605424302375, "7": 0.11343418888780092, "8": 0.11343418888780092}, "confidence": 3.530818044616968e-05}}, "usage": {"input_tokens": 1822, "output_tokens": 0, "cached_input_tokens": 1738}}, "kev_noul": 0.5044575807055498, "choice": "yes", "probs": {"yes": 0.5044575807055498, "no": 0.4955424192944502}, "confidence": 3.530818044616968e-05, "seconds": 0.5371726659996057, "prompt_tokens": 1822, "eval_tokens": 0, "server_latency_ms": null, "prompt_eval_ms": null, "total_ms": 537.1726659996057, "done_reason": null, "id": "off-17", "task": "ka", "slug": "nine-worlds-one-dog", "label": "no", "kind": "off-page", "state_sha256": "84c842a71a59b037055f5448244f732c94578d47c68a907c19c9a4a0324a0095", "prompt_sha256_matches_kev9b": true, "card": "RTX 3090 (fp 33a0b4acb8e8)", "box": "benchbox", "os": {"pretty_name": "Ubuntu 26.04 LTS", "id": "ubuntu", "version_id": "26.04", "kernel": "7.0.0-31-generic", "arch": "x86_64", "source": "/etc/os-release"}, "dry": false, "runtime": "lev serve 0.1.1 (git 0ea955f5aee0) over HTTP on 127.0.0.1, one question per request", "lev_revision": "f8ef71157ec06a7d3b6435bc0756f9d735c33748", "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "lev_commit": "0ea955f5aee0d56d906121d1779af20fa8b922ed", "torch": "2.14.0+cu130", "torch_cuda": "13.0", "transformers": "5.17.0", "peft": "0.21.0", "dtype": "bfloat16", "kernels": "reference (no flash-linear-attention)", "arm": "lev-4b", "model": "interfaze-ai/lev@f8ef7115 on Qwen/Qwen3.5-4B@851bf6e8", "quant": "bf16 backbone + LoRA r=32 (PEFT, unmerged) + mode-B head fp32; calibrated as served; order_average on", "card_uuid": "GPU-fp:33a0b4acb8e8", "num_ctx": 0, "power_cap_w": 300, "seed": 20260921, "stamp": "2026-09-28T13:15:08Z"} {"refused": "overlong", "choice": null, "probs": null, "seconds": null, "why": "inherited from arm 2's own refusal rows so the denominator matches (GAPS-CARD.md, arm 12)", "id": "off-18", "task": "ka", "slug": "how-the-beat-lab-works", "label": "no", "kind": "off-page", "state_sha256": "fdb6d4bf89ebedc44a2d887309bf63ba59598c898a6eca00760115d49bc2133d", "prompt_sha256_matches_kev9b": null, "card": "RTX 3090 (fp 33a0b4acb8e8)", "box": "benchbox", "os": {"pretty_name": "Ubuntu 26.04 LTS", "id": "ubuntu", "version_id": "26.04", "kernel": "7.0.0-31-generic", "arch": "x86_64", "source": "/etc/os-release"}, "dry": false, "runtime": "lev serve 0.1.1 (git 0ea955f5aee0) over HTTP on 127.0.0.1, one question per request", "lev_revision": "f8ef71157ec06a7d3b6435bc0756f9d735c33748", "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "lev_commit": "0ea955f5aee0d56d906121d1779af20fa8b922ed", "torch": "2.14.0+cu130", "torch_cuda": "13.0", "transformers": "5.17.0", "peft": "0.21.0", "dtype": "bfloat16", "kernels": "reference (no flash-linear-attention)", "arm": "lev-4b", "model": "interfaze-ai/lev@f8ef7115 on Qwen/Qwen3.5-4B@851bf6e8", "quant": "bf16 backbone + LoRA r=32 (PEFT, unmerged) + mode-B head fp32; calibrated as served; order_average on", "card_uuid": "GPU-fp:33a0b4acb8e8", "num_ctx": 0, "power_cap_w": 300, "seed": 20260921, "stamp": "2026-09-28T13:15:08Z"} {"mode": "kev-noul", "prompt_sha256": "aca3d93ad50e4f4eaf9e80c4f149926870065ff34d47a39b4f05dd7116ffb923", "n_options": 2, "kev_response": {"model": "Qwen/Qwen3.5-4B", "answers": {"q": {"type": "noul", "noul": 0.07480065984730383, "probabilities": {"0": 0.8645612273940948, "1": 0.028038864173484085, "2": 0.02006073354909795, "3": 0.012894391377107482, "4": 0.011740457965604236, "5": 0.00973315003288059, "6": 0.009603662153135671, "7": 0.00862788840696805, "8": 0.034739624947627244}, "confidence": 0.7192304616995401}}, "usage": {"input_tokens": 7288, "output_tokens": 0, "cached_input_tokens": 7202}}, "kev_noul": 0.07480065984730383, "choice": "no", "probs": {"yes": 0.07480065984730383, "no": 0.9251993401526961}, "confidence": 0.7192304616995401, "seconds": 2.0654823310014763, "prompt_tokens": 7288, "eval_tokens": 0, "server_latency_ms": null, "prompt_eval_ms": null, "total_ms": 2065.4823310014763, "done_reason": null, "id": "off-19", "task": "ka", "slug": "three-at-the-table", "label": "no", "kind": "off-page", "state_sha256": "5ee526a9690f5511741a41a6eb1e91418702e2cd8e49d1c2db9e03802a7bde5c", "prompt_sha256_matches_kev9b": true, "card": "RTX 3090 (fp 33a0b4acb8e8)", "box": "benchbox", "os": {"pretty_name": "Ubuntu 26.04 LTS", "id": "ubuntu", "version_id": "26.04", "kernel": "7.0.0-31-generic", "arch": "x86_64", "source": "/etc/os-release"}, "dry": false, "runtime": "lev serve 0.1.1 (git 0ea955f5aee0) over HTTP on 127.0.0.1, one question per request", "lev_revision": "f8ef71157ec06a7d3b6435bc0756f9d735c33748", "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "lev_commit": "0ea955f5aee0d56d906121d1779af20fa8b922ed", "torch": "2.14.0+cu130", "torch_cuda": "13.0", "transformers": "5.17.0", "peft": "0.21.0", "dtype": "bfloat16", "kernels": "reference (no flash-linear-attention)", "arm": "lev-4b", "model": "interfaze-ai/lev@f8ef7115 on Qwen/Qwen3.5-4B@851bf6e8", "quant": "bf16 backbone + LoRA r=32 (PEFT, unmerged) + mode-B head fp32; calibrated as served; order_average on", "card_uuid": "GPU-fp:33a0b4acb8e8", "num_ctx": 0, "power_cap_w": 300, "seed": 20260921, "stamp": "2026-09-28T13:15:11Z"} {"mode": "kev-noul", "prompt_sha256": "9b582945b20a9d0a581a3907b8efa5f8c9be40c164523b0b971b01b4a8fc46f5", "n_options": 2, "kev_response": {"model": "Qwen/Qwen3.5-4B", "answers": {"q": {"type": "noul", "noul": 0.07451840006257521, "probabilities": {"0": 0.8932038688295876, "1": 0.015855164938029305, "2": 0.011343763329029435, "3": 0.00620887138957909, "4": 0.005430592569832868, "5": 0.004412561633830054, "6": 0.004718169143596099, "7": 0.003732363906406855, "8": 0.05509464426010873}, "confidence": 0.7765213848851461}}, "usage": {"input_tokens": 6248, "output_tokens": 0, "cached_input_tokens": 6159}}, "kev_noul": 0.07451840006257521, "choice": "no", "probs": {"yes": 0.07451840006257521, "no": 0.9254815999374248}, "confidence": 0.7765213848851461, "seconds": 1.7625168359991221, "prompt_tokens": 6248, "eval_tokens": 0, "server_latency_ms": null, "prompt_eval_ms": null, "total_ms": 1762.5168359991221, "done_reason": null, "id": "off-20", "task": "ka", "slug": "the-same-sixteen", "label": "no", "kind": "off-page", "state_sha256": "830caf9405ece8c6ff598d3adfbdaf8da136fd260537d81f576a5059e5a17953", "prompt_sha256_matches_kev9b": true, "card": "RTX 3090 (fp 33a0b4acb8e8)", "box": "benchbox", "os": {"pretty_name": "Ubuntu 26.04 LTS", "id": "ubuntu", "version_id": "26.04", "kernel": "7.0.0-31-generic", "arch": "x86_64", "source": "/etc/os-release"}, "dry": false, "runtime": "lev serve 0.1.1 (git 0ea955f5aee0) over HTTP on 127.0.0.1, one question per request", "lev_revision": "f8ef71157ec06a7d3b6435bc0756f9d735c33748", "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "lev_commit": "0ea955f5aee0d56d906121d1779af20fa8b922ed", "torch": "2.14.0+cu130", "torch_cuda": "13.0", "transformers": "5.17.0", "peft": "0.21.0", "dtype": "bfloat16", "kernels": "reference (no flash-linear-attention)", "arm": "lev-4b", "model": "interfaze-ai/lev@f8ef7115 on Qwen/Qwen3.5-4B@851bf6e8", "quant": "bf16 backbone + LoRA r=32 (PEFT, unmerged) + mode-B head fp32; calibrated as served; order_average on", "card_uuid": "GPU-fp:33a0b4acb8e8", "num_ctx": 0, "power_cap_w": 300, "seed": 20260921, "stamp": "2026-09-28T13:15:12Z"} {"mode": "kev-noul", "prompt_sha256": "5da80e7cb08a3aa664717d70834de08a1c2bb7cf6e2d33e45b04ab19c2672493", "n_options": 2, "kev_response": {"model": "Qwen/Qwen3.5-4B", "answers": {"q": {"type": "noul", "noul": 0.08401632466117108, "probabilities": {"0": 0.8638370266481296, "1": 0.024177672464286752, "2": 0.02031418550936159, "3": 0.010398535998274628, "4": 0.008505973694315695, "5": 0.007146756075023879, "6": 0.00625091389468508, "7": 0.006085698436339349, "8": 0.05328323727958334}, "confidence": 0.7191532054829589}}, "usage": {"input_tokens": 7003, "output_tokens": 0, "cached_input_tokens": 6919}}, "kev_noul": 0.08401632466117108, "choice": "no", "probs": {"yes": 0.08401632466117108, "no": 0.915983675338829}, "confidence": 0.7191532054829589, "seconds": 1.987926328998583, "prompt_tokens": 7003, "eval_tokens": 0, "server_latency_ms": null, "prompt_eval_ms": null, "total_ms": 1987.926328998583, "done_reason": null, "id": "off-21", "task": "ka", "slug": "seat-trials", "label": "no", "kind": "off-page", "state_sha256": "c9374a27b5e06de1440ffed7d81385fa68f7572790c0f553f8f2948c0218eea1", "prompt_sha256_matches_kev9b": true, "card": "RTX 3090 (fp 33a0b4acb8e8)", "box": "benchbox", "os": {"pretty_name": "Ubuntu 26.04 LTS", "id": "ubuntu", "version_id": "26.04", "kernel": "7.0.0-31-generic", "arch": "x86_64", "source": "/etc/os-release"}, "dry": false, "runtime": "lev serve 0.1.1 (git 0ea955f5aee0) over HTTP on 127.0.0.1, one question per request", "lev_revision": "f8ef71157ec06a7d3b6435bc0756f9d735c33748", "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", "lev_commit": "0ea955f5aee0d56d906121d1779af20fa8b922ed", "torch": "2.14.0+cu130", "torch_cuda": "13.0", "transformers": "5.17.0", "peft": "0.21.0", "dtype": "bfloat16", "kernels": "reference (no flash-linear-attention)", "arm": "lev-4b", "model": "interfaze-ai/lev@f8ef7115 on Qwen/Qwen3.5-4B@851bf6e8", "quant": "bf16 backbone + LoRA r=32 (PEFT, unmerged) + mode-B head fp32; calibrated as served; order_average on", "card_uuid": "GPU-fp:33a0b4acb8e8", "num_ctx": 0, "power_cap_w": 300, "seed": 20260921, "stamp": "2026-09-28T13:15:14Z"}