# IdleBench v0.1, suite s2: long-tail multi-model serving (measured, one A10 per arm)

Per cell: mean over seeds of the per-run value, 95% bootstrap CI over seeds (B=2000). Latency: arrival to the last of 64 greedy tokens. Goodput@15: requests within 15 s per GPU-hour held (dedicated: K GPUs over the trace). Waited: share of requests whose model was not serving when they arrived.

| K | Trace | Arm | Class | Record | Seeds | Requests | p50 s | p95 s | p99 s | Attain@5 | Attain@15 | Goodput@15 per GPU-h | Waited | Swaps | Swap p95 s |
|---:|---|---|---|---|---|---:|---|---|---|---|---|---|---|---:|---|
| 2 | st-uniform-k2-r16 | nodus-swap | nodus | imported | 1 | 136 | 4.74 | 8.71 | 9.81 | 0.588 | 1.000 | 1020 | 0.84 | 60 | 4.62 |
| 2 | st-uniform-k2-r2 | nodus-agent-cold | baseline | imported | 1 | 29 | 64.92 | 128.46 | 138.63 | 0.103 | 0.172 | 22 | 0.93 | 0 | n/a |
| 2 | st-uniform-k2-r4 | nodus-swap | nodus | imported | 1 | 96 | 4.72 | 6.92 | 9.32 | 0.708 | 1.000 | 288 | 0.71 | 53 | 4.62 |
| 2 | st-uniform-k2-r60 | nodus-swap | nodus | imported | 1 | 245 | 4.39 | 9.20 | 9.72 | 0.576 | 1.000 | 3599 | 0.93 | 47 | 4.66 |
| 4 | pr754-uniform-k4-r32 | nodus-pool-one-resident | nodus | imported | 1 | 146 | 6.70 | 12.96 | 14.37 | 0.370 | 0.993 | 2106 | 0.92 | 55 | 3.77 |
| 4 | pr754-uniform-k4-r32 | nodus-pool-two-resident | nodus | imported | 1 | 146 | 3.60 | 9.21 | 10.27 | 0.671 | 1.000 | 2163 | 0.70 | 54 | 3.87 |
| 4 | sp-uniform-k4-r16 | dedicated | reference | new | 1 | 19 | 0.51 | 0.99 | 0.99 | 1.000 | 1.000 | 258 | 0.00 | 0 | n/a |
| 4 | sp-uniform-k4-r16 | nodus-lru1 | nodus | imported | 1 | 86 | 4.62 | 10.93 | 15.82 | 0.523 | 0.988 | 1004 | 0.92 | 56 | 3.99 |
| 4 | sp-uniform-k4-r16 | nodus-lru2 | nodus | imported | 1 | 86 | 3.68 | 7.37 | 10.57 | 0.791 | 1.000 | 1018 | 0.74 | 45 | 3.84 |
| 4 | sp-uniform-k4-r16 | nodus-pred1 | nodus | imported | 1 | 86 | 4.46 | 10.69 | 15.51 | 0.535 | 0.988 | 1004 | 0.92 | 56 | 3.94 |
| 4 | sp-uniform-k4-r16 | nodus-pred2 | nodus | imported | 1 | 86 | 3.57 | 7.48 | 10.59 | 0.837 | 1.000 | 1018 | 0.72 | 44 | 3.87 |
| 4 | sp-uniform-k4-r16 | sleep1 | baseline | new | 1 | 86 | 1.11 | 2.41 | 3.32 | 1.000 | 1.000 | 1032 | 0.83 | 66 | 0.75 |
| 4 | sp-uniform-k4-r32 | dedicated | reference | new | 1 | 26 | 0.57 | 0.99 | 1.00 | 1.000 | 1.000 | 415 | 0.00 | 0 | n/a |
| 4 | sp-uniform-k4-r32 | nodus-lru1 | nodus | imported | 1 | 83 | 7.01 | 13.09 | 14.04 | 0.361 | 1.000 | 1598 | 0.95 | 40 | 3.86 |
| 4 | sp-uniform-k4-r32 | nodus-lru2 | nodus | imported | 1 | 83 | 4.16 | 8.46 | 10.85 | 0.602 | 1.000 | 1659 | 0.61 | 36 | 3.90 |
| 4 | sp-uniform-k4-r32 | nodus-pred1 | nodus | imported | 1 | 83 | 6.99 | 13.30 | 14.02 | 0.361 | 1.000 | 1599 | 0.95 | 40 | 3.92 |
| 4 | sp-uniform-k4-r32 | nodus-pred2 | nodus | imported | 1 | 83 | 4.76 | 9.24 | 11.20 | 0.554 | 1.000 | 1629 | 0.66 | 39 | 3.92 |
| 4 | sp-uniform-k4-r32 | sleep1 | baseline | new | 1 | 83 | 1.15 | 2.42 | 3.88 | 1.000 | 1.000 | 1653 | 0.84 | 64 | 0.75 |
| 4 | sp-uniform-k4-r64 | dedicated | reference | new | 1 | 55 | 0.57 | 0.99 | 1.00 | 1.000 | 1.000 | 818 | 0.00 | 0 | n/a |
| 4 | sp-uniform-k4-r64 | nodus-lru1 | nodus | imported | 1 | 109 | 8.03 | 15.36 | 16.96 | 0.294 | 0.927 | 2784 | 0.95 | 30 | 3.91 |
| 4 | sp-uniform-k4-r64 | nodus-lru2 | nodus | imported | 1 | 109 | 3.33 | 9.59 | 10.88 | 0.651 | 1.000 | 3062 | 0.66 | 32 | 3.92 |
| 4 | sp-uniform-k4-r64 | nodus-pred1 | nodus | imported | 1 | 109 | 7.84 | 15.87 | 17.62 | 0.294 | 0.927 | 2772 | 0.94 | 30 | 3.86 |
| 4 | sp-uniform-k4-r64 | nodus-pred2 | nodus | imported | 1 | 109 | 4.29 | 10.00 | 10.89 | 0.578 | 1.000 | 3040 | 0.78 | 34 | 3.90 |
| 4 | sp-uniform-k4-r64 | sleep1 | baseline | new | 1 | 109 | 1.33 | 3.26 | 3.60 | 1.000 | 1.000 | 3226 | 0.85 | 70 | 0.74 |
| 4 | sp-uniform-k4-r8 | dedicated | reference | new | 1 | 5 | 0.57 | 0.99 | 0.99 | 1.000 | 1.000 | 132 | 0.00 | 0 | n/a |
| 4 | sp-uniform-k4-r8 | nodus-lru1 | nodus | imported | 1 | 88 | 4.36 | 8.81 | 11.47 | 0.659 | 1.000 | 528 | 0.88 | 67 | 4.48 |
| 4 | sp-uniform-k4-r8 | nodus-lru2 | nodus | imported | 1 | 88 | 3.45 | 5.85 | 7.46 | 0.932 | 1.000 | 528 | 0.64 | 49 | 3.65 |
| 4 | sp-uniform-k4-r8 | nodus-pred1 | nodus | imported | 1 | 88 | 4.35 | 8.83 | 11.38 | 0.705 | 1.000 | 528 | 0.88 | 70 | 4.01 |
| 4 | sp-uniform-k4-r8 | nodus-pred2 | nodus | imported | 1 | 88 | 2.32 | 7.03 | 7.54 | 0.920 | 1.000 | 528 | 0.57 | 46 | 3.71 |
| 4 | sp-uniform-k4-r8 | sleep1 | baseline | new | 1 | 88 | 1.12 | 2.23 | 3.20 | 1.000 | 1.000 | 528 | 0.80 | 69 | 0.75 |
| 4 | sp-zipf-k4-r16 | dedicated | reference | new | 1 | 17 | 0.25 | 1.19 | 1.19 | 1.000 | 1.000 | 228 | 0.00 | 0 | n/a |
| 4 | sp-zipf-k4-r16 | nodus-lru1 | nodus | imported | 1 | 76 | 4.46 | 10.73 | 13.28 | 0.658 | 1.000 | 900 | 0.92 | 47 | 3.92 |
| 4 | sp-zipf-k4-r16 | nodus-lru2 | nodus | imported | 1 | 76 | 0.98 | 6.01 | 7.12 | 0.895 | 1.000 | 912 | 0.50 | 29 | 3.81 |
| 4 | sp-zipf-k4-r16 | nodus-pred1 | nodus | imported | 1 | 76 | 4.30 | 10.65 | 13.13 | 0.658 | 1.000 | 900 | 0.91 | 52 | 3.99 |
| 4 | sp-zipf-k4-r16 | nodus-pred2 | nodus | imported | 1 | 76 | 0.52 | 7.15 | 8.94 | 0.855 | 1.000 | 912 | 0.45 | 38 | 3.83 |
| 4 | sp-zipf-k4-r16 | sleep1 | baseline | new | 1 | 76 | 0.93 | 1.90 | 2.02 | 1.000 | 1.000 | 912 | 0.74 | 50 | 0.74 |
| 4 | sp-zipf-k4-r16 | sleep2 | baseline | new | 1 | 76 | 1.18 | 2.89 | 3.47 | 1.000 | 1.000 | 912 | 0.76 | 49 | 1.66 |
| 4 | sp-zipf-k4-r8 | cold | baseline | new | 1 | 83 | 47.19 | 116.09 | 123.21 | 0.012 | 0.120 | 51 | 1.00 | 23 | 31.98 |
| 4 | sp-zipf-k4-r8 | dedicated | reference | new | 1 | 5 | 0.51 | 0.57 | 0.57 | 1.000 | 1.000 | 124 | 0.00 | 0 | n/a |
| 4 | sp-zipf-k4-r8 | nodus-lru1 | nodus | imported | 1 | 83 | 4.11 | 8.67 | 15.11 | 0.771 | 0.988 | 492 | 0.80 | 52 | 3.86 |
| 4 | sp-zipf-k4-r8 | nodus-lru2 | nodus | imported | 1 | 83 | 0.57 | 5.61 | 11.02 | 0.928 | 1.000 | 498 | 0.47 | 31 | 3.87 |
| 4 | sp-zipf-k4-r8 | nodus-pred1 | nodus | imported | 1 | 83 | 4.18 | 8.59 | 15.20 | 0.747 | 0.988 | 492 | 0.75 | 69 | 3.86 |
| 4 | sp-zipf-k4-r8 | nodus-pred2 | nodus | imported | 1 | 83 | 0.51 | 4.78 | 10.47 | 0.952 | 1.000 | 498 | 0.28 | 28 | 3.82 |
| 4 | sp-zipf-k4-r8 | sleep1 | baseline | new | 1 | 83 | 0.95 | 1.75 | 3.47 | 1.000 | 1.000 | 498 | 0.69 | 55 | 0.74 |
| 4 | sp-zipf-k4-r8 | sleep2 | baseline | new | 1 | 83 | 1.59 | 3.21 | 5.57 | 0.976 | 1.000 | 498 | 0.71 | 55 | 1.58 |
| 4 | st-uniform-k4-r16 | nodus-swap | nodus | imported | 1 | 144 | 8.41 | 17.25 | 19.46 | 0.229 | 0.882 | 944 | 0.97 | 82 | 4.90 |
| 4 | st-uniform-k4-r32 | nodus-swap | nodus | imported | 1 | 212 | 10.36 | 19.92 | 21.20 | 0.255 | 0.750 | 1553 | 0.96 | 68 | 4.93 |
| 4 | st-uniform-k4-r4 | nodus-swap | nodus | imported | 1 | 98 | 5.12 | 10.33 | 19.04 | 0.429 | 0.980 | 192 | 0.76 | 69 | 4.93 |
| 4 | st-uniform-k4-r8 | nodus-swap | nodus | imported | 1 | 115 | 5.26 | 11.47 | 14.50 | 0.383 | 0.991 | 453 | 0.79 | 79 | 4.90 |
| 6 | pr754-uniform-k6-r24 | nodus-pool-one-resident | nodus | imported | 1 | 115 | 7.71 | 17.69 | 22.47 | 0.322 | 0.913 | 1250 | 0.97 | 69 | 3.76 |
| 6 | pr754-uniform-k6-r24 | nodus-pool-two-resident | nodus | imported | 1 | 115 | 3.99 | 11.41 | 14.94 | 0.626 | 0.991 | 1358 | 0.83 | 68 | 3.73 |
| 6 | pr754-uniform-k6-r48 | nodus-pool-one-resident | nodus | imported | 1 | 204 | 11.16 | 21.46 | 22.87 | 0.245 | 0.725 | 2116 | 0.97 | 61 | 3.75 |
| 6 | pr754-uniform-k6-r48 | nodus-pool-two-resident | nodus | imported | 1 | 204 | 6.74 | 14.61 | 16.36 | 0.446 | 0.956 | 2875 | 0.81 | 67 | 3.76 |
| 6 | sp-uniform-k6-r12 | dedicated | reference | new | 1 | 14 | 0.52 | 1.00 | 1.00 | 1.000 | 1.000 | 120 | 0.00 | 0 | n/a |
| 6 | sp-uniform-k6-r12 | nodus-lru1 | nodus | imported | 1 | 84 | 4.87 | 10.10 | 13.10 | 0.536 | 1.000 | 720 | 0.92 | 68 | 3.91 |
| 6 | sp-uniform-k6-r12 | nodus-lru2 | nodus | imported | 1 | 84 | 3.93 | 7.21 | 10.22 | 0.857 | 1.000 | 720 | 0.76 | 59 | 3.81 |
| 6 | sp-uniform-k6-r12 | nodus-pred1 | nodus | imported | 1 | 84 | 5.05 | 10.45 | 15.82 | 0.488 | 0.988 | 711 | 1.00 | 75 | 3.99 |
| 6 | sp-uniform-k6-r12 | nodus-pred2 | nodus | imported | 1 | 84 | 4.03 | 6.91 | 10.33 | 0.821 | 1.000 | 720 | 0.71 | 64 | 3.82 |
| 6 | sp-uniform-k6-r12 | sleep1 | baseline | new | 1 | 84 | 1.20 | 1.83 | 2.57 | 1.000 | 1.000 | 720 | 0.88 | 74 | 0.74 |
| 6 | sp-uniform-k6-r24 | dedicated | reference | new | 1 | 19 | 0.52 | 0.99 | 0.99 | 1.000 | 1.000 | 215 | 0.00 | 0 | n/a |
| 6 | sp-uniform-k6-r24 | nodus-lru1 | nodus | imported | 1 | 86 | 6.34 | 13.72 | 17.06 | 0.384 | 0.988 | 1234 | 0.98 | 51 | 3.89 |
| 6 | sp-uniform-k6-r24 | nodus-lru2 | nodus | imported | 1 | 86 | 4.16 | 9.26 | 12.49 | 0.628 | 1.000 | 1271 | 0.83 | 48 | 3.86 |
| 6 | sp-uniform-k6-r24 | nodus-pred1 | nodus | imported | 1 | 86 | 6.27 | 13.60 | 16.93 | 0.407 | 0.977 | 1213 | 0.99 | 51 | 3.89 |
| 6 | sp-uniform-k6-r24 | nodus-pred2 | nodus | imported | 1 | 86 | 3.94 | 9.60 | 12.76 | 0.709 | 1.000 | 1271 | 0.76 | 45 | 3.82 |
| 6 | sp-uniform-k6-r24 | sleep1 | baseline | new | 1 | 86 | 1.21 | 2.63 | 4.25 | 1.000 | 1.000 | 1290 | 0.91 | 68 | 0.74 |
| 6 | sp-uniform-k6-r48 | dedicated | reference | new | 1 | 35 | 0.52 | 0.63 | 0.99 | 1.000 | 1.000 | 477 | 0.00 | 0 | n/a |
| 6 | sp-uniform-k6-r48 | nodus-lru1 | nodus | imported | 1 | 143 | 10.10 | 20.45 | 24.03 | 0.217 | 0.671 | 1744 | 0.99 | 48 | 3.86 |
| 6 | sp-uniform-k6-r48 | nodus-lru2 | nodus | imported | 1 | 143 | 6.45 | 13.88 | 15.11 | 0.406 | 0.986 | 2588 | 0.81 | 54 | 3.88 |
| 6 | sp-uniform-k6-r48 | nodus-pred1 | nodus | imported | 1 | 143 | 9.75 | 21.02 | 24.11 | 0.224 | 0.678 | 1762 | 0.99 | 48 | 3.95 |
| 6 | sp-uniform-k6-r48 | nodus-pred2 | nodus | imported | 1 | 143 | 6.79 | 14.09 | 15.70 | 0.364 | 0.986 | 2578 | 0.83 | 53 | 3.88 |
| 6 | sp-uniform-k6-r48 | sleep1 | baseline | new | 1 | 143 | 1.28 | 3.30 | 4.11 | 1.000 | 1.000 | 2813 | 0.89 | 107 | 0.74 |
| 6 | sp-uniform-k6-r6 | dedicated | reference | new | 1 | 5 | 0.52 | 1.00 | 1.00 | 1.000 | 1.000 | 60 | 0.00 | 0 | n/a |
| 6 | sp-uniform-k6-r6 | nodus-lru1 | nodus | imported | 1 | 72 | 4.15 | 11.12 | 15.36 | 0.736 | 0.986 | 355 | 0.89 | 60 | 3.80 |
| 6 | sp-uniform-k6-r6 | nodus-lru2 | nodus | imported | 1 | 72 | 3.87 | 7.04 | 11.26 | 0.792 | 1.000 | 360 | 0.79 | 55 | 3.64 |
| 6 | sp-uniform-k6-r6 | nodus-pred1 | nodus | imported | 1 | 72 | 4.13 | 11.01 | 15.19 | 0.722 | 0.986 | 355 | 0.86 | 68 | 3.78 |
| 6 | sp-uniform-k6-r6 | nodus-pred1-kill | nodus | imported | 1 | 72 | 4.19 | 24.18 | 28.45 | 0.625 | 0.833 | 300 | 0.89 | 61 | 3.79 |
| 6 | sp-uniform-k6-r6 | nodus-pred2 | nodus | imported | 1 | 72 | 3.85 | 8.67 | 11.27 | 0.792 | 1.000 | 360 | 0.78 | 61 | 3.64 |
| 6 | sp-uniform-k6-r6 | nodus-pred2-kill | nodus | imported | 1 | 72 | 3.83 | 20.58 | 24.03 | 0.708 | 0.875 | 315 | 0.76 | 52 | 3.75 |
| 6 | sp-uniform-k6-r6 | sleep1 | baseline | new | 1 | 72 | 1.18 | 1.84 | 2.04 | 1.000 | 1.000 | 360 | 0.85 | 61 | 0.74 |
| 6 | sp-zipf-k6-r12 | dedicated | reference | new | 1 | 16 | 0.51 | 0.99 | 0.99 | 1.000 | 1.000 | 123 | 0.00 | 0 | n/a |
| 6 | sp-zipf-k6-r12 | nodus-lru1 | nodus | imported | 1 | 86 | 4.31 | 11.03 | 16.07 | 0.581 | 0.988 | 729 | 0.87 | 59 | 3.92 |
| 6 | sp-zipf-k6-r12 | nodus-lru2 | nodus | imported | 1 | 86 | 3.35 | 8.72 | 10.94 | 0.826 | 1.000 | 737 | 0.57 | 41 | 3.82 |
| 6 | sp-zipf-k6-r12 | nodus-pred1 | nodus | imported | 1 | 86 | 4.58 | 10.72 | 13.93 | 0.570 | 1.000 | 737 | 0.90 | 67 | 3.88 |
| 6 | sp-zipf-k6-r12 | nodus-pred2 | nodus | imported | 1 | 86 | 0.53 | 7.02 | 10.81 | 0.826 | 1.000 | 737 | 0.47 | 51 | 3.72 |
| 6 | sp-zipf-k6-r12 | sleep1 | baseline | new | 1 | 86 | 1.17 | 2.07 | 4.19 | 1.000 | 1.000 | 737 | 0.83 | 67 | 0.72 |
| 6 | sp-zipf-k6-r12 | sleep2 | baseline | new | 1 | 86 | 1.58 | 3.52 | 5.84 | 0.977 | 1.000 | 737 | 0.84 | 67 | 1.55 |
| 6 | sp-zipf-k6-r6 | cold | baseline | new | 1 | 66 | 48.33 | 92.10 | 108.82 | 0.030 | 0.061 | 18 | 0.98 | 29 | 28.12 |
| 6 | sp-zipf-k6-r6 | dedicated | reference | new | 1 | 4 | 0.35 | 0.82 | 0.82 | 1.000 | 1.000 | 55 | 0.00 | 0 | n/a |
| 6 | sp-zipf-k6-r6 | nodus-lru1 | nodus | imported | 1 | 66 | 4.21 | 7.71 | 9.51 | 0.742 | 1.000 | 330 | 0.91 | 51 | 3.84 |
| 6 | sp-zipf-k6-r6 | nodus-lru2 | nodus | imported | 1 | 66 | 3.43 | 5.45 | 7.07 | 0.924 | 1.000 | 330 | 0.55 | 33 | 3.67 |
| 6 | sp-zipf-k6-r6 | nodus-pred1 | nodus | imported | 1 | 66 | 4.22 | 7.84 | 9.56 | 0.742 | 1.000 | 330 | 0.85 | 68 | 3.89 |
| 6 | sp-zipf-k6-r6 | nodus-pred2 | nodus | imported | 1 | 66 | 3.41 | 6.36 | 7.07 | 0.894 | 1.000 | 330 | 0.56 | 47 | 3.70 |
| 6 | sp-zipf-k6-r6 | sleep1 | baseline | new | 1 | 66 | 1.10 | 1.73 | 1.99 | 1.000 | 1.000 | 330 | 0.80 | 51 | 0.73 |
| 6 | sp-zipf-k6-r6 | sleep2 | baseline | new | 1 | 66 | 1.57 | 2.55 | 2.96 | 1.000 | 1.000 | 330 | 0.86 | 51 | 1.59 |
| 6 | st-uniform-k6-r12 | nodus-swap | nodus | imported | 1 | 132 | 6.01 | 16.23 | 18.35 | 0.273 | 0.932 | 615 | 0.93 | 96 | 4.93 |
| 6 | st-uniform-k6-r24 | nodus-swap | nodus | imported | 1 | 191 | 13.98 | 24.09 | 26.13 | 0.147 | 0.545 | 772 | 0.99 | 93 | 4.99 |
| 6 | st-uniform-k6-r3 | nodus-swap | nodus | imported | 1 | 114 | 5.15 | 9.25 | 11.17 | 0.333 | 1.000 | 171 | 0.95 | 101 | 4.86 |
| 6 | st-uniform-k6-r6 | nodus-swap | nodus | imported | 1 | 133 | 5.34 | 11.31 | 13.31 | 0.323 | 0.992 | 393 | 0.92 | 114 | 5.03 |
| 6 | st-uniform-k6-r72 | nodus-agent-dedicated | reference | imported | 1 | 157 | 0.51 | 0.99 | 1.62 | 1.000 | 1.000 | 784 | 0.00 | 0 | n/a |

**Headline: SLO-goodput at 15 s per GPU-hour, best Nodus arm ÷ best baseline arm (geometric mean of per-seed ratios, CI over seeds), and best Nodus arm ÷ dedicated.**

| K | Trace | Best Nodus | Goodput@15 | Best baseline | Goodput@15 | Nodus ÷ baseline @15 | Dedicated goodput@15 | Nodus ÷ dedicated @15 | Best Nodus @5 | Best baseline @5 | Nodus ÷ baseline @5 |
|---:|---|---|---:|---|---:|---|---:|---|---|---|---:|
| 2 | st-uniform-k2-r16 | nodus-swap | 1020 | n/a | nan | n/a | nan | n/a |  0 |  0 | nanx |
| 2 | st-uniform-k2-r4 | nodus-swap | 288 | n/a | nan | n/a | nan | n/a |  0 |  0 | nanx |
| 2 | st-uniform-k2-r60 | nodus-swap | 3599 | n/a | nan | n/a | nan | n/a |  0 |  0 | nanx |
| 4 | pr754-uniform-k4-r32 | nodus-pool-two-resident | 2163 | n/a | nan | n/a | nan | n/a |  0 |  0 | nanx |
| 4 | sp-uniform-k4-r16 | nodus-lru2 | 1018 | sleep1 | 1032 | 0.99x | 258 | 3.95x | nodus-pred2 853 | sleep1 1032 | 0.83x |
| 4 | sp-uniform-k4-r32 | nodus-lru2 | 1659 | sleep1 | 1653 | 1.00x | 415 | 4.00x | nodus-lru2 999 | sleep1 1653 | 0.60x |
| 4 | sp-uniform-k4-r64 | nodus-lru2 | 3062 | sleep1 | 3226 | 0.95x | 818 | 3.75x | nodus-lru2 1995 | sleep1 3226 | 0.62x |
| 4 | sp-uniform-k4-r8 | nodus-lru1 | 528 | sleep1 | 528 | 1.00x | 132 | 4.00x | nodus-lru2 492 | sleep1 528 | 0.93x |
| 4 | sp-zipf-k4-r16 | nodus-lru2 | 912 | sleep1 | 912 | 1.00x | 228 | 4.00x | nodus-lru2 816 | sleep1 912 | 0.89x |
| 4 | sp-zipf-k4-r8 | nodus-lru2 | 498 | sleep1 | 498 | 1.00x | 124 | 4.00x | nodus-pred2 474 | sleep1 498 | 0.95x |
| 4 | st-uniform-k4-r16 | nodus-swap | 944 | n/a | nan | n/a | nan | n/a |  0 |  0 | nanx |
| 4 | st-uniform-k4-r32 | nodus-swap | 1553 | n/a | nan | n/a | nan | n/a |  0 |  0 | nanx |
| 4 | st-uniform-k4-r4 | nodus-swap | 192 | n/a | nan | n/a | nan | n/a |  0 |  0 | nanx |
| 4 | st-uniform-k4-r8 | nodus-swap | 453 | n/a | nan | n/a | nan | n/a |  0 |  0 | nanx |
| 6 | pr754-uniform-k6-r24 | nodus-pool-two-resident | 1358 | n/a | nan | n/a | nan | n/a |  0 |  0 | nanx |
| 6 | pr754-uniform-k6-r48 | nodus-pool-two-resident | 2875 | n/a | nan | n/a | nan | n/a |  0 |  0 | nanx |
| 6 | sp-uniform-k6-r12 | nodus-lru1 | 720 | sleep1 | 720 | 1.00x | 120 | 6.00x | nodus-lru2 617 | sleep1 720 | 0.86x |
| 6 | sp-uniform-k6-r24 | nodus-lru2 | 1271 | sleep1 | 1290 | 0.99x | 215 | 5.91x | nodus-pred2 901 | sleep1 1290 | 0.70x |
| 6 | sp-uniform-k6-r48 | nodus-lru2 | 2588 | sleep1 | 2813 | 0.92x | 477 | 5.43x | nodus-lru2 1064 | sleep1 2813 | 0.38x |
| 6 | sp-uniform-k6-r6 | nodus-lru2 | 360 | sleep1 | 360 | 1.00x | 60 | 6.00x | nodus-lru2 285 | sleep1 360 | 0.79x |
| 6 | sp-zipf-k6-r12 | nodus-lru2 | 737 | sleep1 | 737 | 1.00x | 123 | 6.00x | nodus-lru2 609 | sleep1 737 | 0.83x |
| 6 | sp-zipf-k6-r6 | nodus-lru1 | 330 | sleep1 | 330 | 1.00x | 55 | 6.00x | nodus-lru2 305 | sleep1 330 | 0.92x |
| 6 | st-uniform-k6-r12 | nodus-swap | 615 | n/a | nan | n/a | nan | n/a |  0 |  0 | nanx |
| 6 | st-uniform-k6-r24 | nodus-swap | 772 | n/a | nan | n/a | nan | n/a |  0 |  0 | nanx |
| 6 | st-uniform-k6-r3 | nodus-swap | 171 | n/a | nan | n/a | nan | n/a |  0 |  0 | nanx |
| 6 | st-uniform-k6-r6 | nodus-swap | 393 | n/a | nan | n/a | nan | n/a |  0 |  0 | nanx |

Geometric mean over cells of Nodus ÷ best baseline: **0.99**

**Identity.** 3535 first decodes (batch 1) over 77 (model, prompt) keys; 0 keys with more than one text.

**Paired per request (same arrivals): new baseline arm minus imported Nodus arm.** Negative means the baseline answered faster. 95% CI from 2000 paired bootstrap resamples of the requests.

| Trace | Baseline (new) | Nodus (imported) | Paired n | Mean Δ s | Δp50 s | Δp95 s | Baseline p95 | Nodus p95 |
|---|---|---|---:|---|---|---|---:|---:|
| sp-uniform-k4-r16 | sleep1 | nodus-lru1 | 86/86 | -3.99 [-4.56, -3.41] | -3.51 [-4.34, -3.13] | -8.51 [-9.72, -6.05] | 2.41 | 10.93 |
| sp-uniform-k4-r16 | sleep1 | nodus-lru2 | 86/86 | -2.16 [-2.66, -1.68] | -2.56 [-3.02, -1.70] | -4.96 [-6.75, -3.71] | 2.41 | 7.37 |
| sp-uniform-k4-r32 | sleep1 | nodus-lru1 | 83/83 | -5.47 [-6.29, -4.69] | -5.86 [-7.14, -4.73] | -10.66 [-11.30, -8.70] | 2.42 | 13.09 |
| sp-uniform-k4-r32 | sleep1 | nodus-lru2 | 83/83 | -2.56 [-3.22, -1.88] | -3.01 [-4.16, -0.60] | -6.03 [-7.99, -4.88] | 2.42 | 8.46 |
| sp-uniform-k4-r64 | sleep1 | nodus-lru1 | 109/109 | -6.57 [-7.49, -5.66] | -6.70 [-8.35, -5.97] | -12.10 [-13.90, -11.36] | 3.26 | 15.36 |
| sp-uniform-k4-r64 | sleep1 | nodus-lru2 | 109/109 | -2.26 [-2.81, -1.72] | -2.00 [-3.23, -0.43] | -6.34 [-7.81, -5.40] | 3.26 | 9.59 |
| sp-uniform-k4-r8 | sleep1 | nodus-lru1 | 88/88 | -3.36 [-3.79, -2.93] | -3.24 [-3.71, -3.07] | -6.58 [-7.88, -5.17] | 2.23 | 8.81 |
| sp-uniform-k4-r8 | sleep1 | nodus-lru2 | 88/88 | -1.60 [-2.00, -1.21] | -2.33 [-2.73, -0.18] | -3.62 [-5.17, -2.15] | 2.23 | 5.85 |
| sp-uniform-k6-r12 | sleep1 | nodus-lru1 | 84/84 | -3.93 [-4.51, -3.36] | -3.67 [-4.36, -3.06] | -8.27 [-10.50, -6.51] | 1.83 | 10.10 |
| sp-uniform-k6-r12 | sleep1 | nodus-lru2 | 84/84 | -2.33 [-2.74, -1.90] | -2.74 [-2.89, -2.37] | -5.38 [-6.77, -3.66] | 1.83 | 7.21 |
| sp-uniform-k6-r24 | sleep1 | nodus-lru1 | 86/86 | -5.34 [-6.05, -4.63] | -5.13 [-6.41, -3.79] | -11.09 [-12.01, -8.01] | 2.63 | 13.72 |
| sp-uniform-k6-r24 | sleep1 | nodus-lru2 | 86/86 | -3.00 [-3.59, -2.43] | -2.95 [-3.52, -2.52] | -6.64 [-9.74, -5.23] | 2.63 | 9.26 |
| sp-uniform-k6-r48 | sleep1 | nodus-lru1 | 143/143 | -9.19 [-10.25, -8.15] | -8.82 [-10.79, -6.66] | -17.15 [-20.28, -16.15] | 3.30 | 20.45 |
| sp-uniform-k6-r48 | sleep1 | nodus-lru2 | 143/143 | -4.83 [-5.55, -4.08] | -5.17 [-6.52, -3.88] | -10.58 [-11.00, -9.15] | 3.30 | 13.88 |
| sp-uniform-k6-r6 | sleep1 | nodus-lru1 | 72/72 | -3.62 [-4.28, -3.04] | -2.97 [-3.12, -2.83] | -9.28 [-11.29, -5.93] | 1.84 | 11.12 |
| sp-uniform-k6-r6 | sleep1 | nodus-lru2 | 72/72 | -2.57 [-3.03, -2.15] | -2.69 [-2.81, -2.31] | -5.19 [-7.33, -4.28] | 1.84 | 7.04 |
| sp-zipf-k4-r16 | sleep1 | nodus-lru1 | 76/76 | -3.87 [-4.45, -3.27] | -3.53 [-3.94, -3.15] | -8.82 [-10.52, -6.00] | 1.90 | 10.73 |
| sp-zipf-k4-r16 | sleep1 | nodus-lru2 | 76/76 | -1.36 [-1.80, -0.92] | -0.05 [-2.59, 0.47] | -4.11 [-5.05, -3.01] | 1.90 | 6.01 |
| sp-zipf-k4-r16 | sleep2 | nodus-lru1 | 76/76 | -3.41 [-3.97, -2.83] | -3.28 [-3.62, -2.68] | -7.84 [-9.37, -4.89] | 2.89 | 10.73 |
| sp-zipf-k4-r16 | sleep2 | nodus-lru2 | 76/76 | -0.91 [-1.31, -0.50] | 0.20 [-2.11, 1.00] | -3.13 [-3.89, -2.02] | 2.89 | 6.01 |
| sp-zipf-k4-r8 | cold | nodus-lru1 | 83/83 | 45.84 [38.92, 52.87] | 43.08 [32.71, 50.70] | 107.43 [80.75, 113.99] | 116.09 | 8.67 |
| sp-zipf-k4-r8 | cold | nodus-lru2 | 83/83 | 47.73 [40.91, 54.71] | 46.62 [35.13, 53.43] | 110.48 [83.91, 116.80] | 116.09 | 5.61 |
| sp-zipf-k4-r8 | sleep1 | nodus-lru1 | 83/83 | -3.09 [-3.60, -2.59] | -3.16 [-3.29, -2.85] | -6.92 [-9.99, -5.51] | 1.75 | 8.67 |
| sp-zipf-k4-r8 | sleep1 | nodus-lru2 | 83/83 | -1.20 [-1.62, -0.78] | 0.38 [-2.41, 0.53] | -3.87 [-5.52, -2.44] | 1.75 | 5.61 |
| sp-zipf-k4-r8 | sleep2 | nodus-lru1 | 83/83 | -2.57 [-3.01, -2.16] | -2.52 [-3.01, -2.44] | -5.45 [-7.61, -4.44] | 3.21 | 8.67 |
| sp-zipf-k4-r8 | sleep2 | nodus-lru2 | 83/83 | -0.68 [-1.08, -0.27] | 1.02 [-1.79, 1.14] | -2.40 [-3.97, -0.37] | 3.21 | 5.61 |
| sp-zipf-k6-r12 | sleep1 | nodus-lru1 | 86/86 | -4.11 [-4.74, -3.52] | -3.14 [-4.36, -2.97] | -8.96 [-12.48, -7.32] | 2.07 | 11.03 |
| sp-zipf-k6-r12 | sleep1 | nodus-lru2 | 86/86 | -1.84 [-2.37, -1.31] | -2.17 [-2.63, 0.60] | -6.65 [-7.36, -4.82] | 2.07 | 8.72 |
| sp-zipf-k6-r12 | sleep2 | nodus-lru1 | 86/86 | -3.63 [-4.20, -3.09] | -2.74 [-3.94, -2.54] | -7.51 [-11.45, -6.34] | 3.52 | 11.03 |
| sp-zipf-k6-r12 | sleep2 | nodus-lru2 | 86/86 | -1.36 [-1.85, -0.87] | -1.77 [-2.22, 1.00] | -5.20 [-6.09, -3.63] | 3.52 | 8.72 |
| sp-zipf-k6-r6 | cold | nodus-lru1 | 66/66 | 46.54 [40.89, 52.29] | 44.12 [39.52, 52.48] | 84.39 [71.12, 90.80] | 92.10 | 7.71 |
| sp-zipf-k6-r6 | cold | nodus-lru2 | 66/66 | 48.46 [42.99, 54.31] | 44.90 [40.79, 53.36] | 86.65 [73.02, 93.31] | 92.10 | 5.45 |
| sp-zipf-k6-r6 | sleep1 | nodus-lru1 | 66/66 | -3.40 [-3.79, -3.01] | -3.11 [-3.36, -2.94] | -5.98 [-6.25, -5.00] | 1.73 | 7.71 |
| sp-zipf-k6-r6 | sleep1 | nodus-lru2 | 66/66 | -1.48 [-1.97, -1.01] | -2.33 [-2.61, 0.57] | -3.72 [-5.10, -2.93] | 1.73 | 5.45 |
| sp-zipf-k6-r6 | sleep2 | nodus-lru1 | 66/66 | -3.00 [-3.36, -2.63] | -2.64 [-3.10, -2.55] | -5.17 [-5.60, -4.23] | 2.55 | 7.71 |
| sp-zipf-k6-r6 | sleep2 | nodus-lru2 | 66/66 | -1.07 [-1.55, -0.61] | -1.86 [-2.33, 0.99] | -2.90 [-4.27, -2.09] | 2.55 | 5.45 |