census: longest page 82500 tokens, num_ctx 16384, fits=False, overlong pages 7
ANNOUNCE  box benchbox · cards GPU-46890836-d8f, GPU-7aa0be10-974 · arm gemma4-fp8-readout · task a · start 2026-09-21T22:04:37Z · expect ~20 min
  on-01                        yes    -> REFUSED: overlong                       — (82500 tokens)
  on-02                        yes    -> REFUSED: overlong                       — (26700 tokens)
  on-03                        yes    -> yes                                  1.14s
  on-04                        yes    -> yes                                  1.91s
  on-05                        yes    -> yes                                  5.10s
  on-06                        yes    -> yes                                  5.73s
  on-07                        yes    -> yes                                  2.06s
  on-08                        yes    -> yes                                  1.36s
  on-09                        yes    -> yes                                  2.39s
  on-10                        yes    -> yes                                  4.19s
  on-11                        yes    -> yes                                  3.27s
  on-12                        yes    -> yes                                  0.68s
  on-13                        yes    -> yes                                  2.66s
  on-14                        yes    -> REFUSED: overlong                       — (31403 tokens)
  on-15                        yes    -> yes                                  2.00s
  on-16                        yes    -> yes                                  1.71s
  on-17                        yes    -> yes                                  3.89s
  on-18                        yes    -> yes                                  1.97s
  on-19                        yes    -> yes                                  2.23s
  on-20                        yes    -> yes                                  1.98s
  on-21                        yes    -> yes                                  1.97s
  on-22                        yes    -> yes                                  2.76s
  on-23                        yes    -> yes                                  3.22s
  on-24                        yes    -> yes                                  2.82s
  on-25                        yes    -> yes                                  2.01s
  on-26                        yes    -> yes                                  2.30s
  on-27                        yes    -> yes                                  5.85s
  on-28                        yes    -> yes                                  4.78s
  on-29                        yes    -> REFUSED: overlong                       — (63199 tokens)
  on-30                        yes    -> yes                                  1.79s
  on-31                        yes    -> yes                                  1.87s
  on-32                        yes    -> yes                                  2.23s
  on-33                        yes    -> REFUSED: overlong                       — (30028 tokens)
  on-34                        yes    -> REFUSED: overlong                       — (16374 tokens)
  on-35                        yes    -> yes                                  3.91s
  on-36                        yes    -> yes                                  4.27s
  on-37                        yes    -> yes                                  2.66s
  on-38                        yes    -> yes                                  3.92s
  on-39                        yes    -> yes                                  3.21s
  on-40                        yes    -> REFUSED: overlong                       — (18777 tokens)
  on-41                        yes    -> yes                                  3.47s
  on-42                        yes    -> yes                                  2.52s
  off-01                       no     -> no                                   2.67s
  off-02                       no     -> no                                   1.87s
  off-03                       no     -> no                                   4.78s
  off-04                       no     -> REFUSED: overlong                       — (82500 tokens)
  off-05                       no     -> no                                   4.22s
  off-06                       no     -> no                                   2.02s
  off-07                       no     -> no                                   1.71s
  off-08                       no     -> no                                   2.23s
  off-09                       no     -> no                                   1.97s
  off-10                       no     -> no                                   1.09s
  off-11                       no     -> no                                   3.92s
  off-12                       no     -> no                                   2.01s
  off-13                       no     -> no                                   1.97s
  off-14                       no     -> REFUSED: overlong                       — (63199 tokens)
  off-15                       no     -> no                                   5.85s
  off-16                       no     -> no                                   1.79s
  off-17                       no     -> no                                   0.68s
  off-18                       no     -> REFUSED: overlong                       — (31403 tokens)
  off-19                       no     -> no                                   2.83s
  off-20                       no     -> no                                   2.40s
  off-21                       no     -> no                                   2.66s
{"n": 63, "n_reached": 53, "n_labelled": 53, "refused": 10, "refused_by_reason": {"overlong": 10}, "correct": 53, "errors": 0, "brier": 0.00013928258326234135, "reliability": [{"lo": 0.0, "hi": 0.1, "n": 0, "sum_p": 0.0, "correct": 0, "mean_p": null, "accuracy": null, "gap": null}, {"lo": 0.1, "hi": 0.2, "n": 0, "sum_p": 0.0, "correct": 0, "mean_p": null, "accuracy": null, "gap": null}, {"lo": 0.2, "hi": 0.3, "n": 0, "sum_p": 0.0, "correct": 0, "mean_p": null, "accuracy": null, "gap": null}, {"lo": 0.3, "hi": 0.4, "n": 0, "sum_p": 0.0, "correct": 0, "mean_p": null, "accuracy": null, "gap": nul
ANNOUNCE  box benchbox · cards GPU-46890836-d8f, GPU-7aa0be10-974 · arm gemma4-fp8-readout · task b · start 2026-09-21T22:07:15Z · expect ~15 min
  on-01                        -      -> REFUSED: overlong                       — (82500 tokens)
  on-02                        -      -> REFUSED: overlong                       — (26700 tokens)
  on-03                        -      -> for-the-technically-curious          1.14s
  on-04                        -      -> where-these-answers-come-from        2.02s
  on-05                        -      -> top                                  5.26s
  on-06                        -      -> how-to-listen                        5.76s
  on-07                        -      -> the-bill                             2.21s
  on-08                        -      -> five-minutes-of-training-honestly-   1.46s
  on-09                        -      -> sixteen-of-nineteen                  2.51s
  on-10                        -      -> two-hours-at-12-tok-s-the-test-tha   0.24s
  on-11                        -      -> what-the-battery-was-for-the-instr   3.40s
  on-12                        -      -> top                                  0.68s
  on-13                        -      -> the-answer-in-one-paragraph          0.20s
  on-14                        -      -> REFUSED: overlong                       — (31403 tokens)
  on-15                        -      -> the-developer                        2.06s
  on-16                        -      -> the-table-that-answers-the-questio   0.14s
  on-17                        -      -> finding-one-a-picture-is-199-token   4.01s
  on-18                        -      -> finding-four-the-argument-with-our   0.17s
  on-19                        -      -> the-world-is-a-file                  2.28s
  on-20                        -      -> two-kinds-of-there-all-along         2.05s
  on-21                        -      -> the-punchline                        0.13s
  on-22                        -      -> the-day-the-box-got-faster           2.95s
  on-23                        -      -> the-same-question-down-the-whole-l   3.32s
  on-24                        -      -> the-payroll-and-the-meeting          0.21s
  on-25                        -      -> a-measurement-from-our-own-card      2.11s
  on-26                        -      -> the-third-librarian-never-left-you   2.28s
  on-27                        -      -> what-to-take-with-you                6.05s
  on-28                        -      -> where-the-file-came-from             4.90s
  on-29                        -      -> REFUSED: overlong                       — (63199 tokens)
  on-30                        -      -> the-first-promise-your-visit-stays   1.84s
  on-31                        -      -> the-thorough-walk                    1.95s
  on-32                        -      -> receipts                             2.28s
  on-33                        -      -> REFUSED: overlong                       — (30028 tokens)
  on-34                        -      -> REFUSED: overlong                       — (16374 tokens)
  on-35                        -      -> vendors                              3.98s
  on-36                        -      -> what-to-take-with-you                4.36s
  on-37                        -      -> the-instrument                       2.75s
  on-38                        -      -> provenance-4cae43                    4.05s
  on-39                        -      -> top                                  3.34s
  on-40                        -      -> REFUSED: overlong                       — (18777 tokens)
  on-41                        -      -> the-blind-sitting                    3.60s
  on-42                        ['no template tail — no closing sentence of the shape "the article does not specify what was left out, moved or not measured" where the article carries no such statement', "an operator, never the definite one — the hub's voice law (an operator 2026-09-03, re-stated 2026-09-04)", "the prediction's own subject kept — the page predicts that a few COMPANIES will sell intelligence as a utility; the commodity sentence is about SOLD INTELLIGENCE and may not be re-subjected onto the prediction"] -> what-an-operator-saw                 2.63s
{"n": 42, "n_reached": 35, "n_labelled": 1, "refused": 7, "refused_by_reason": {"overlong": 7}, "correct": 0, "errors": 0, "brier": 0.9999626109234422, "reliability": [{"lo": 0.0, "hi": 0.1, "n": 0, "sum_p": 0.0, "correct": 0, "mean_p": null, "accuracy": null, "gap": null}, {"lo": 0.1, "hi": 0.2, "n": 0, "sum_p": 0.0, "correct": 0, "mean_p": null, "accuracy": null, "gap": null}, {"lo": 0.2, "hi": 0.3, "n": 0, "sum_p": 0.0, "correct": 0, "mean_p": null, "accuracy": null, "gap": null}, {"lo": 0.3, "hi": 0.4, "n": 0, "sum_p": 0.0, "correct": 0, "mean_p": null, "accuracy": null, "gap": null}, {"lo
census: reusing rows-addenda/census.gemma4-fp8-vllm.json
census: longest page 82500 tokens, num_ctx 16384, fits=False, overlong pages 7
ANNOUNCE  box benchbox · cards GPU-46890836-d8f, GPU-7aa0be10-974 · arm gemma4-fp8-generate · task a · start 2026-09-21T22:08:54Z · expect ~20 min
  on-01                        yes    -> REFUSED: overlong                       — (82500 tokens)
  on-02                        yes    -> REFUSED: overlong                       — (26700 tokens)
  on-03                        yes    -> yes                                  1.11s
  on-04                        yes    -> yes                                  1.79s
  on-05                        yes    -> yes                                  5.07s
  on-06                        yes    -> yes                                  5.61s
  on-07                        yes    -> yes                                  2.10s
  on-08                        yes    -> yes                                  1.39s
  on-09                        yes    -> yes                                  2.42s
  on-10                        yes    -> yes                                  4.24s
  on-11                        yes    -> yes                                  3.29s
  on-12                        yes    -> yes                                  0.69s
  on-13                        yes    -> yes                                  2.68s
  on-14                        yes    -> REFUSED: overlong                       — (31403 tokens)
  on-15                        yes    -> yes                                  2.01s
  on-16                        yes    -> yes                                  1.71s
  on-17                        yes    -> yes                                  3.90s
  on-18                        yes    -> yes                                  1.97s
  on-19                        yes    -> yes                                  2.24s
  on-20                        yes    -> yes                                  1.99s
  on-21                        yes    -> yes                                  1.98s
  on-22                        yes    -> yes                                  2.77s
  on-23                        yes    -> yes                                  3.23s
  on-24                        yes    -> yes                                  2.84s
  on-25                        yes    -> yes                                  2.02s
  on-26                        yes    -> yes                                  2.16s
  on-27                        yes    -> yes                                  5.86s
  on-28                        yes    -> yes                                  4.80s
  on-29                        yes    -> REFUSED: overlong                       — (63199 tokens)
  on-30                        yes    -> yes                                  1.79s
  on-31                        yes    -> yes                                  1.87s
  on-32                        yes    -> yes                                  2.23s
  on-33                        yes    -> REFUSED: overlong                       — (30028 tokens)
  on-34                        yes    -> REFUSED: overlong                       — (16374 tokens)
  on-35                        yes    -> yes                                  3.93s
  on-36                        yes    -> yes                                  4.29s
  on-37                        yes    -> yes                                  2.67s
  on-38                        yes    -> yes                                  3.93s
  on-39                        yes    -> yes                                  3.21s
  on-40                        yes    -> REFUSED: overlong                       — (18777 tokens)
  on-41                        yes    -> yes                                  3.48s
  on-42                        yes    -> yes                                  2.52s
  off-01                       no     -> no                                   2.67s
  off-02                       no     -> no                                   1.87s
  off-03                       no     -> no                                   4.79s
  off-04                       no     -> REFUSED: overlong                       — (82500 tokens)
  off-05                       no     -> no                                   4.22s
  off-06                       no     -> no                                   2.01s
  off-07                       no     -> no                                   1.71s
  off-08                       no     -> no                                   2.23s
  off-09                       no     -> no                                   1.97s
  off-10                       no     -> no                                   1.10s
  off-11                       no     -> no                                   3.92s
  off-12                       no     -> no                                   2.01s
  off-13                       no     -> no                                   1.98s
  off-14                       no     -> REFUSED: overlong                       — (63199 tokens)
  off-15                       no     -> no                                   5.87s
  off-16                       no     -> no                                   1.79s
  off-17                       no     -> no                                   0.70s
  off-18                       no     -> REFUSED: overlong                       — (31403 tokens)
  off-19                       no     -> no                                   2.83s
  off-20                       no     -> no                                   2.40s
  off-21                       no     -> no                                   2.67s
{"n": 63, "n_reached": 53, "n_labelled": 53, "refused": 10, "refused_by_reason": {"overlong": 10}, "correct": 53, "errors": 0, "brier": 8.349285404259997e-05, "reliability": [{"lo": 0.0, "hi": 0.1, "n": 0, "sum_p": 0.0, "correct": 0, "mean_p": null, "accuracy": null, "gap": null}, {"lo": 0.1, "hi": 0.2, "n": 0, "sum_p": 0.0, "correct": 0, "mean_p": null, "accuracy": null, "gap": null}, {"lo": 0.2, "hi": 0.3, "n": 0, "sum_p": 0.0, "correct": 0, "mean_p": null, "accuracy": null, "gap": null}, {"lo": 0.3, "hi": 0.4, "n": 0, "sum_p": 0.0, "correct": 0, "mean_p": null, "accuracy": null, "gap": null
ANNOUNCE  box benchbox · cards GPU-46890836-d8f, GPU-7aa0be10-974 · arm gemma4-fp8-generate · task b · start 2026-09-21T22:11:31Z · expect ~15 min
  on-01                        -      -> REFUSED: overlong                       — (82500 tokens)
  on-02                        -      -> REFUSED: overlong                       — (26700 tokens)
  on-03                        -      -> for-the-technically-curious          1.14s
  on-04                        -      -> where-these-answers-come-from        2.04s
  on-05                        -      -> top                                  5.19s
  on-06                        -      -> how-to-listen                        5.67s
  on-07                        -      -> the-bill                             2.19s
  on-08                        -      -> five-minutes-of-training-honestly-   1.44s
  on-09                        -      -> sixteen-of-nineteen                  2.50s
  on-10                        -      -> two-hours-at-12-tok-s-the-test-tha   0.24s
  on-11                        -      -> what-the-battery-was-for-the-instr   3.41s
  on-12                        -      -> top                                  0.69s
  on-13                        -      -> the-answer-in-one-paragraph          0.20s
  on-14                        -      -> REFUSED: overlong                       — (31403 tokens)
  on-15                        -      -> the-developer                        2.07s
  on-16                        -      -> the-table-that-answers-the-questio   0.15s
  on-17                        -      -> finding-one-a-picture-is-199-token   4.01s
  on-18                        -      -> finding-four-the-argument-with-our   0.18s
  on-19                        -      -> the-world-is-a-file                  2.28s
  on-20                        -      -> two-kinds-of-there-all-along         2.05s
  on-21                        -      -> what-we-can-show-you                 0.13s
  on-22                        -      -> the-day-the-box-got-faster           2.96s
  on-23                        -      -> the-same-question-down-the-whole-l   3.36s
  on-24                        -      -> the-payroll-and-the-meeting          0.23s
  on-25                        -      -> a-measurement-from-our-own-card      2.12s
  on-26                        -      -> the-third-librarian-never-left-you   2.30s
  on-27                        -      -> the-seat-question                    6.05s
  on-28                        -      -> where-the-file-came-from             4.92s
  on-29                        -      -> REFUSED: overlong                       — (63199 tokens)
  on-30                        -      -> the-first-promise-your-visit-stays   1.90s
  on-31                        -      -> the-thorough-walk                    1.96s
  on-32                        -      -> receipts                             2.31s
  on-33                        -      -> REFUSED: overlong                       — (30028 tokens)
  on-34                        -      -> REFUSED: overlong                       — (16374 tokens)
  on-35                        -      -> vendors                              3.99s
  on-36                        -      -> what-to-take-with-you                4.39s
  on-37                        -      -> the-instrument                       2.81s
  on-38                        -      -> provenance-4cae43                    4.07s
  on-39                        -      -> provenance-063a00                    3.40s
  on-40                        -      -> REFUSED: overlong                       — (18777 tokens)
  on-41                        -      -> the-blind-sitting                    3.64s
  on-42                        ['no template tail — no closing sentence of the shape "the article does not specify what was left out, moved or not measured" where the article carries no such statement', "an operator, never the definite one — the hub's voice law (an operator 2026-09-03, re-stated 2026-09-04)", "the prediction's own subject kept — the page predicts that a few COMPANIES will sell intelligence as a utility; the commodity sentence is about SOLD INTELLIGENCE and may not be re-subjected onto the prediction"] -> what-an-operator-saw                 2.68s
{"n": 42, "n_reached": 35, "n_labelled": 1, "refused": 7, "refused_by_reason": {"overlong": 7}, "correct": 0, "errors": 0, "brier": 1.0, "reliability": [{"lo": 0.0, "hi": 0.1, "n": 0, "sum_p": 0.0, "correct": 0, "mean_p": null, "accuracy": null, "gap": null}, {"lo": 0.1, "hi": 0.2, "n": 0, "sum_p": 0.0, "correct": 0, "mean_p": null, "accuracy": null, "gap": null}, {"lo": 0.2, "hi": 0.3, "n": 0, "sum_p": 0.0, "correct": 0, "mean_p": null, "accuracy": null, "gap": null}, {"lo": 0.3, "hi": 0.4, "n": 0, "sum_p": 0.0, "correct": 0, "mean_p": null, "accuracy": null, "gap": null}, {"lo": 0.4, "hi": 0