Gate G: bootstrap statistical audit (table in paper), manifest generator

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

Simon-Pierre Boucher committed 2 days ago (Nov 1, 2026) parent 068f50f Browse files

7 changed files +1,646 −2

added artifacts/manifest.json +965 −0

++{
++ "generated": "2026-11-01T01:24:31",
++ "git": "068f50f77c00921163f84652a4dccc6f84a4f554",
++ "models": {
++  "olmoe-1b-7b-0125-4bit": "mlx-community/OLMoE-1B-7B-0125-Instruct-4bit@e23844197887b031e7ddddbb0b8959c5a6853a7b",
++  "qwen3-30b-a3b-2507-4bit": "mlx-community/Qwen3-30B-A3B-Instruct-2507-4bit@e9675aa3ca5f900ccef55267914466d55ab325fa"
++ },
++ "files": [
++  {
++   "path": "artifacts/environment.json",
++   "bytes": 17537,
++   "sha256": "e05e02401355ab3a3500c181264a4a4ac038c0bef38fe8718bcc6099db0b9498"
++  },
++  {
++   "path": "benchmarks/processed/audit.json",
++   "bytes": 12184,
++   "sha256": "1bfa8337208bfa80c4dba5f9ced382e7ec4177311106da60f00754358720d874"
++  },
++  {
++   "path": "benchmarks/processed/gateC/sim_olmoe_laptop_ssd_v1.json",
++   "bytes": 123933,
++   "sha256": "d7188c80daf3db38455e330c7e414c9ce0056289e9858a2403f94f715f5b7464"
++  },
++  {
++   "path": "benchmarks/processed/gateC/sim_olmoe_v2_ssd043_m8.json",
++   "bytes": 60788,
++   "sha256": "d6e35cf4100ec34e484f1267770530d5f4c4e1a8764171f7fde6a508dde4527d"
++  },
++  {
++   "path": "benchmarks/processed/gateC/sim_olmoe_v2_ssd043_m8_probe_fixed.json",
++   "bytes": 9131,
++   "sha256": "c069f0df73cc261b82e58529b46b3c0aed6ae2b8665a6c78c6df61f364afea0c"
++  },
++  {
++   "path": "benchmarks/raw/gateA/smoke/mlx_lm_generate_laptop_20261101T011129.txt",
++   "bytes": 772,
++   "sha256": "86d3b5a63f5d8a0c20a78bcaa337a4d329c89063f26304e4d777bf057a626045"
++  },
++  {
++   "path": "benchmarks/raw/gateA/ssd/laptop-M5Max.json",
++   "bytes": 995,
++   "sha256": "93116e6ce7c0c2039525b8ee731edfddcb2ef936c40547298c1b3e3131ec864e"
++  },
++  {
++   "path": "benchmarks/raw/gateB/capacity_sweep_laptop/sweep.jsonl",
++   "bytes": 183957,
++   "sha256": "3fdc16b869407201d8544e4ee4cfe775652ead132f9539f4465f23b80948f956"
++  },
++  {
++   "path": "benchmarks/raw/gateB/design/fastpath_profile_laptop.txt",
++   "bytes": 331,
++   "sha256": "c76ecd5b8d6ec170999d37590124597141bb33e5a591fc552373dacc5f0a5519"
++  },
++  {
++   "path": "benchmarks/raw/gateB/design/slot_update_microbench_laptop.txt",
++   "bytes": 365,
++   "sha256": "bfd1f99ab8b777b2b6a3bac59fe5f0e55fa6abcc4741397d17480a6464321d4f"
++  },
++  {
++   "path": "benchmarks/raw/gateC/cold_nvme_laptop/attribution.jsonl",
++   "bytes": 120437,
++   "sha256": "9c2b80b0c6042ad3124142d332bbae73d17148fdc46b4384195726afb7862f3c"
++  },
++  {
++   "path": "benchmarks/raw/gateC/cold_nvme_laptop/quick_v2.jsonl",
++   "bytes": 3924,
++   "sha256": "e8d892ce579de09e6828d8821894c40a2f882bda2ff433ac24a224c90ddc933e"
++  },
++  {
++   "path": "benchmarks/raw/gateC/cold_nvme_laptop/quick_v3.jsonl",
++   "bytes": 3964,
++   "sha256": "52b022ff9ae01490c987bd62ac8382c05795949a13eddf1246267762682cd294"
++  },
++  {
++   "path": "benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl",
++   "bytes": 450009,
++   "sha256": "4c0a4268388dc88929b05df49ff67a9e5fa7d4e5df908b0b8e4d76cc583d259c"
++  },
++  {
++   "path": "benchmarks/raw/gateC/cold_nvme_laptop/sweep_v1_9preads_serial.jsonl",
++   "bytes": 29919,
++   "sha256": "2fff605ec66aae9672c4fe37b26fc019765392765d180b49b2935999403dbc31"
++  },
++  {
++   "path": "benchmarks/raw/gateC/design/miss_cost_breakdown_laptop.txt",
++   "bytes": 547,
++   "sha256": "94efb69aefe9ae86e9c59bcd96836f818d9d658c16ff85b4c46ab0f081572f00"
++  },
++  {
++   "path": "benchmarks/raw/gateC/prefetch_online_laptop/smoke.jsonl",
++   "bytes": 72158,
++   "sha256": "e01beefef6a42567a11cd158080cdf4f7e56e67eafa91cd00300450cb37f679f"
++  },
++  {
++   "path": "benchmarks/raw/gateC/prefetch_online_laptop/smoke_lazyinstall.jsonl",
++   "bytes": 48055,
++   "sha256": "b33f0d34e8c6172b408b1d2f864e50e6a61199ee4a8efd412f26402756bdc5ac"
++  },
++  {
++   "path": "benchmarks/raw/gateC/qwen3_m3u96a/baseline.jsonl",
++   "bytes": 80611,
++   "sha256": "4f509b6faa8722d275384181cf40583caa004edc890cecd9582566303068767b"
++  },
++  {
++   "path": "benchmarks/raw/gateC/qwen3_m3u96a/baseline_mlxlm.log",
++   "bytes": 1945,
++   "sha256": "4df7656a9a81d354b46464fda6bd80ddb93986273305402ce52a0a4aee70402c"
++  },
++  {
++   "path": "benchmarks/raw/gateC/qwen3_m3u96a/baseline_titan128.log",
++   "bytes": 2425,
++   "sha256": "50788b364c260b944aa93482c0a706177d707300f50ddc10c9ef6732baa4d413"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_laptop/code_fit_0.npz",
++   "bytes": 525273,
++   "sha256": "90267d30b9c9c6854dc6430df229e6aa13ef505a167a1a9674cd8d37f7de0314"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_laptop/code_fit_1.npz",
++   "bytes": 546094,
++   "sha256": "dfd9f17d01541206e60def9371d708a11772c437d7233c17a7760260c4a68e06"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_laptop/code_fit_2.npz",
++   "bytes": 529494,
++   "sha256": "31bf7227a1936b292359c7ec44a6375a279672b57360711134ee956ea38078db"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_laptop/code_heldout_0.npz",
++   "bytes": 533705,
++   "sha256": "5783a1755a51899de6a99b8b831a480d0730e15c9f1788a751b1d62583856ad0"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_laptop/code_heldout_1.npz",
++   "bytes": 523765,
++   "sha256": "a46f9f759d1fb5368f9735c95064b85b8fcb2ad8dd0283e7ec44312f07d0ab7a"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_laptop/general_fit_0.npz",
++   "bytes": 214468,
++   "sha256": "7066dd1e4c56e6ba7c0352dab45f80724d79fb9d553515e62f3c1b204ca183b5"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_laptop/general_fit_1.npz",
++   "bytes": 529130,
++   "sha256": "710053d323754cc5aee1ee4672e3995cf5f14d8425ee0b0edbff7a2bef62065d"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_laptop/general_fit_2.npz",
++   "bytes": 465070,
++   "sha256": "ac5a47c87592c3a9def03657dd350491d8bae8c20aa12d46e4f0d8117d55cd52"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_laptop/general_heldout_0.npz",
++   "bytes": 520565,
++   "sha256": "0ff645b4745490dbb46b00d678f4fbf5e5824be341f37520ff2d6981ef3f6ce5"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_laptop/general_heldout_1.npz",
++   "bytes": 513497,
++   "sha256": "00c51955b25b4164e204d6405f3f9e6ae2d4661811781d08875318616d66ffe3"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_laptop/index.json",
++   "bytes": 8051,
++   "sha256": "50367d2c17613fc3670db61e4d80f502514d1793ce85fc58265ee1f80cf07180"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_laptop/multilingual_fit_0.npz",
++   "bytes": 135710,
++   "sha256": "9d921d7c22b4c595d67ee68b8480e6beca041906387948b523c0018afe9b7a56"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_laptop/multilingual_fit_1.npz",
++   "bytes": 565691,
++   "sha256": "1f418dcb08c137ed8d5fdab2975afd440d21aab2edbd77552fd385105412961a"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_laptop/multilingual_heldout_0.npz",
++   "bytes": 76269,
++   "sha256": "2e136aeb8cee3daab7dc8cdde0ef9e134f067e7b15bd2da76d17342f1be5d19c"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_laptop/reasoning_fit_0.npz",
++   "bytes": 129575,
++   "sha256": "11073c5039bfab061cdedf64b0d078eef228e47b1f4e43eb25d784e949a1905a"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_laptop/reasoning_fit_1.npz",
++   "bytes": 249978,
++   "sha256": "81bb7bc96f9950d33fc598d445a92b0fecddc09e72c58c6a4c84941561971968"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_laptop/reasoning_fit_2.npz",
++   "bytes": 196539,
++   "sha256": "6d559291eb0be3054a5d4b9c261a75f4fbb591ea1b637a0a9eb8bc2425860f49"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_laptop/reasoning_heldout_0.npz",
++   "bytes": 346389,
++   "sha256": "4d17db9bb112f9e236739551f91009f19e19f4c6c76c74fcb546c1d6b539e294"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_laptop/reasoning_heldout_1.npz",
++   "bytes": 251462,
++   "sha256": "ead5a58228140725dea6259dff1c75e8852913dd453157d2fb03c471f46c479c"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_v2_laptop/code_fit_0.npz",
++   "bytes": 656354,
++   "sha256": "5081b895aec17144fa5e2658c85e057715cc15214633fefeb3f23cd01982b38e"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_v2_laptop/code_fit_1.npz",
++   "bytes": 788804,
++   "sha256": "aaba7275c8cda68efdeae893e8483a33cbe58fa1a98652e2c26b13da2e9f0abb"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_v2_laptop/code_fit_2.npz",
++   "bytes": 771988,
++   "sha256": "038cb3423715ca6bd992208bf5261c83c4180dc153e7cfb22c50570e80eab65c"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_v2_laptop/code_fit_3.npz",
++   "bytes": 770796,
++   "sha256": "b9923b51b23860def2d8a32217b3eea67cba27e8edb364d44490ad4084466bd2"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_v2_laptop/code_fit_4.npz",
++   "bytes": 718360,
++   "sha256": "9840400ce2e85170e258cabd0d2bc4d53eedc09628d08e0c9ffaf367135e34a0"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_v2_laptop/code_fit_5.npz",
++   "bytes": 789657,
++   "sha256": "33089502109cc23d0f4aaf3c3d3d42957517ab49ebd4f96c109af7c763fee9aa"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_v2_laptop/code_fit_6.npz",
++   "bytes": 764331,
++   "sha256": "e5c2ed9a15a211f3fda302164c0d4fb07220ce4867f8e0d455583522ce1a4c2e"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_v2_laptop/code_fit_7.npz",
++   "bytes": 704696,
++   "sha256": "084c843363fec70f0585caef9bfe1e3ef9419aba5e5dff2e3f070b8a3cb050e1"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_v2_laptop/code_fit_8.npz",
++   "bytes": 772710,
++   "sha256": "677ae00bb2ce15578dc2d34909162f76fed9ae2d56b1edea68579c3400773595"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_v2_laptop/code_fit_9.npz",
++   "bytes": 564005,
++   "sha256": "508d0e337e86d08f36fa77669e7c826572426f54125162b240bcf88b3e9eced2"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_v2_laptop/code_heldout_0.npz",
++   "bytes": 775944,
++   "sha256": "ea96518772ab286ee91469181f94cdfbb409c4b09d2eab1e7fd2b9ac2fb6e4b4"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_v2_laptop/code_heldout_1.npz",
++   "bytes": 731455,
++   "sha256": "b9127ab21af3fb65cd6b4f026de155e6e1dd9334108999b0f643f6e681c6c942"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_v2_laptop/code_heldout_2.npz",
++   "bytes": 766485,
++   "sha256": "1a45deb2e2d83825a3318f23bbdcce0e6fda6bda0e9527593524c4232e347735"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_v2_laptop/code_heldout_3.npz",
++   "bytes": 767985,
++   "sha256": "56c7f815093ea80ff9272bfff80dcef07c1475da32d0601f6087c4291373a39c"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_v2_laptop/drift_fit_0.npz",
++   "bytes": 759367,
++   "sha256": "80b608063a4e7a0e759345fe1ecfbda9f1cb316674965597c0fb35a48fef9a34"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_v2_laptop/drift_fit_1.npz",
++   "bytes": 786411,
++   "sha256": "91f136e58dfb01b9ee12ea519e2339bf59225f5b94e78425ac648be02305a292"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_v2_laptop/drift_fit_2.npz",
++   "bytes": 822551,
++   "sha256": "f565d674205e7235ed8b9043391d8d7a0c249c90d1dd9f39999ee288191f76f3"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_v2_laptop/drift_heldout_0.npz",
++   "bytes": 777557,
++   "sha256": "60adb0d387dd74f018437839a9b2a7e32f4b1b2b74d00edb710f61a68041e33c"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_v2_laptop/drift_heldout_1.npz",
++   "bytes": 189224,
++   "sha256": "3df66e8b80c9179284a6cb8fc608c5a7d5db7b4b24f48045714a57021fc227dc"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_v2_laptop/general_fit_0.npz",
++   "bytes": 214468,
++   "sha256": "7066dd1e4c56e6ba7c0352dab45f80724d79fb9d553515e62f3c1b204ca183b5"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_v2_laptop/general_fit_1.npz",
++   "bytes": 559492,
++   "sha256": "60a139c0aa5aa69df1e978925286ae744c2ce758aed4ab2fee0d3168b2bf3ee2"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_v2_laptop/general_fit_2.npz",
++   "bytes": 465070,
++   "sha256": "ac5a47c87592c3a9def03657dd350491d8bae8c20aa12d46e4f0d8117d55cd52"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_v2_laptop/general_fit_3.npz",
++   "bytes": 744436,
++   "sha256": "379532fa0b8e567a4d92d1eaab588e3de0e1ed948bd92ec991d6cb07761d69b9"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_v2_laptop/general_fit_4.npz",
++   "bytes": 586886,
++   "sha256": "d3c85a2dd425f88468f8a848327369901bf981cf64305473f78a5d68d08e8551"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_v2_laptop/general_fit_5.npz",
++   "bytes": 755527,
++   "sha256": "1f192d34b94945a544fb248b9017161750517e5ffc5bcad33e472cad46590efa"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_v2_laptop/general_fit_6.npz",
++   "bytes": 763897,
++   "sha256": "8e027395b485605921f5642b23c39de6e3d617dc5449797a7f2b09046c083255"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_v2_laptop/general_fit_7.npz",
++   "bytes": 486948,
++   "sha256": "b7f96e5cd07cdb0d22d8a89130442fceefbbf8bb462dcb5656e80c55d07ee770"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_v2_laptop/general_fit_8.npz",
++   "bytes": 418245,
++   "sha256": "e0224f3574bab1a8a539e7d4e5c921bdef260e9d33911d590d7f0d8dae2d5635"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_v2_laptop/general_fit_9.npz",
++   "bytes": 377742,
++   "sha256": "25ad255572ac404c83240d4c92ff4aa3c3f9fd195d1f18485afd0a32e31cf09a"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_v2_laptop/general_heldout_0.npz",
++   "bytes": 624508,
++   "sha256": "44a96860fefb1365820ef8901c45b517dc259836c7a1a077eda925dc17606eb6"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_v2_laptop/general_heldout_1.npz",
++   "bytes": 753336,
++   "sha256": "0c9e0b87b0c83032e13fdcf440b0bf7d4a1b16cdeef73fa30f7287c77fe62375"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_v2_laptop/general_heldout_2.npz",
++   "bytes": 525021,
++   "sha256": "82962233c031e7a6b986c0518e222e4784da0290c532183e28c3a67f7bc23f53"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_v2_laptop/general_heldout_3.npz",
++   "bytes": 763713,
++   "sha256": "6f5e244ca12cb77076596ff880e35c6fe895d35abeff12fb43c9932377186104"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_v2_laptop/index.json",
++   "bytes": 26533,
++   "sha256": "cd8ad1d7dc06c8c330f6da219e5ce8f88a351332ed81f4db632607a112a71ad9"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_v2_laptop/longctx_fit_0.npz",
++   "bytes": 972584,
++   "sha256": "8ca45b55c34c9484e87a6dc89c5cf361ded13adf1078887ef41890dd8a0ea18c"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_v2_laptop/longctx_fit_1.npz",
++   "bytes": 920455,
++   "sha256": "7e3afe7e1ebb873b6ad34b9788864f6c89a284e78d135080b5108b24f73f838f"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_v2_laptop/longctx_fit_2.npz",
++   "bytes": 904796,
++   "sha256": "04d5be500de14ece655cf0c2051fae05249acb9e86762beb63e98838c9b4f255"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_v2_laptop/longctx_heldout_0.npz",
++   "bytes": 925810,
++   "sha256": "08d3cdc98c05f74d110240f7086b61fd386ca0e16a2e7760eec1881dd901452a"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_v2_laptop/multilingual_fit_0.npz",
++   "bytes": 135710,
++   "sha256": "9d921d7c22b4c595d67ee68b8480e6beca041906387948b523c0018afe9b7a56"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_v2_laptop/multilingual_fit_1.npz",
++   "bytes": 701384,
++   "sha256": "99f54a73e81cbe61f32ac85cb509bd6b098f0761f454f756076a63d7b52d4ee4"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_v2_laptop/multilingual_fit_2.npz",
++   "bytes": 777501,
++   "sha256": "9c10fc1b5a81517668896d06c9dbe3478ba0f2ef7de57384cfe11ccf5ebbd84f"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_v2_laptop/multilingual_fit_3.npz",
++   "bytes": 506541,
++   "sha256": "15c4fbae207491ecd5d9c50d7beff30d49e48c26fac56103e6b4587ead4d6fcb"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_v2_laptop/multilingual_fit_4.npz",
++   "bytes": 792758,
++   "sha256": "a42656c5c583aaeed7aea9ab14537cbd844ce4d5441353c9671d1dd0f4575cab"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_v2_laptop/multilingual_fit_5.npz",
++   "bytes": 523552,
++   "sha256": "ffaaf51002d108a30b611311925f04c94c327e16dc99e03b10aceeddbc24af7b"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_v2_laptop/multilingual_fit_6.npz",
++   "bytes": 789213,
++   "sha256": "e07fcedb92273e3f591c63d764517cd8c34ef764dd8ef3fd5c34a551a515ecf0"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_v2_laptop/multilingual_fit_7.npz",
++   "bytes": 506011,
++   "sha256": "563411725c51eac9b49d4f5f9251a362027448f858b717de28d5c7172806f00b"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_v2_laptop/multilingual_heldout_0.npz",
++   "bytes": 76269,
++   "sha256": "2e136aeb8cee3daab7dc8cdde0ef9e134f067e7b15bd2da76d17342f1be5d19c"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_v2_laptop/multilingual_heldout_1.npz",
++   "bytes": 161420,
++   "sha256": "63f8cab9c6ac952e2bc5d2c6ef9b1d5b621e07dfcbb14d1f71f360115431c40d"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_v2_laptop/multilingual_heldout_2.npz",
++   "bytes": 778872,
++   "sha256": "ecb4bd5f11a311905466be5a21bc9679be2122abe1918a23d089440e895652bc"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_v2_laptop/reasoning_fit_0.npz",
++   "bytes": 129575,
++   "sha256": "11073c5039bfab061cdedf64b0d078eef228e47b1f4e43eb25d784e949a1905a"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_v2_laptop/reasoning_fit_1.npz",
++   "bytes": 249978,
++   "sha256": "81bb7bc96f9950d33fc598d445a92b0fecddc09e72c58c6a4c84941561971968"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_v2_laptop/reasoning_fit_2.npz",
++   "bytes": 196539,
++   "sha256": "6d559291eb0be3054a5d4b9c261a75f4fbb591ea1b637a0a9eb8bc2425860f49"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_v2_laptop/reasoning_fit_3.npz",
++   "bytes": 154398,
++   "sha256": "e49f157944767abedbbf5e11e8a44658dc1af232ff5573adffe7d3633bcc5e18"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_v2_laptop/reasoning_fit_4.npz",
++   "bytes": 205524,
++   "sha256": "63fbb98584d09812b649202e230e83533ae7eb13aa10b9c846e690b089b2abf8"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_v2_laptop/reasoning_fit_5.npz",
++   "bytes": 166122,
++   "sha256": "425340ee9a334cda84f6a3f56605a6b6b555d2ead304a6b78e09632e2ea0d1f6"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_v2_laptop/reasoning_fit_6.npz",
++   "bytes": 223204,
++   "sha256": "5c16696d7a7b1bc4bfe4ab513561ad139121a3e10bb5b666a59a0eac07382bf9"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_v2_laptop/reasoning_fit_7.npz",
++   "bytes": 171583,
++   "sha256": "55bfbbe7b44255fcdfcb342726a864485520f30bffa7590bc60fcec9089caedc"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_v2_laptop/reasoning_fit_8.npz",
++   "bytes": 160712,
++   "sha256": "6605a3e42c1aa56e4fa8c1574b802f36039bf3b8c919d2967bc566400bb4b40b"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_v2_laptop/reasoning_fit_9.npz",
++   "bytes": 181392,
++   "sha256": "faf93bed507e97be69e4bc621763dda0a161597de77afe3bbe602253b66095e7"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_v2_laptop/reasoning_heldout_0.npz",
++   "bytes": 346389,
++   "sha256": "4d17db9bb112f9e236739551f91009f19e19f4c6c76c74fcb546c1d6b539e294"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_v2_laptop/reasoning_heldout_1.npz",
++   "bytes": 251462,
++   "sha256": "ead5a58228140725dea6259dff1c75e8852913dd453157d2fb03c471f46c479c"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_v2_laptop/reasoning_heldout_2.npz",
++   "bytes": 185075,
++   "sha256": "de5d8a3214f3abd1847282b6a88c274729eb23e2919cbf64a8f7c45c82f90da1"
++  },
++  {
++   "path": "benchmarks/raw/gateC/traces_olmoe_v2_laptop/reasoning_heldout_3.npz",
++   "bytes": 213567,
++   "sha256": "5fa14d863660b17366ed9c28e50d7dd197bab9244cd8e36df6f55af070418ee0"
++  },
++  {
++   "path": "benchmarks/raw/gateD/fused_upgate_laptop.jsonl",
++   "bytes": 86913,
++   "sha256": "0061d8c870b522f0c4070ba562cfe437ba655c05db105aecc994d67bc387a8bb"
++  },
++  {
++   "path": "benchmarks/raw/gateE/remote_cost_breakdown_m3u96a_m3u96b.txt",
++   "bytes": 655,
++   "sha256": "dc4f40e8b782c4cd31bf9fd44a58aa59608d86cd24edf90ecf663e4bc6aecd9e"
++  },
++  {
++   "path": "benchmarks/raw/gateE/run1.log",
++   "bytes": 8356,
++   "sha256": "88326472853330864c268710bd804491dc2412508c5621356e4fabbbc123c31f"
++  },
++  {
++   "path": "benchmarks/raw/gateE/two_node_m3u96a_m3u96b.jsonl",
++   "bytes": 153008,
++   "sha256": "641405e60392332ed5661b6bbfbd974de723ae35d99d2cb3af6afcfef6ee29d0"
++  },
++  {
++   "path": "benchmarks/raw/tests/distributed_localhost_olmoe_20261101T010211.json",
++   "bytes": 1505,
++   "sha256": "eb9ac1542ab445099df37fcc485c95af8106c46535b114426b8bd0d106c2f459"
++  },
++  {
++   "path": "benchmarks/raw/tests/distributed_localhost_olmoe_20261101T010636.json",
++   "bytes": 1504,
++   "sha256": "b022f63d30bf9b837c7d40cfd256d8c5ddba5ea89bdbdb94084eeebec8c354ba"
++  },
++  {
++   "path": "benchmarks/raw/tests/distributed_localhost_olmoe_20261101T010747.json",
++   "bytes": 1504,
++   "sha256": "913e2e610f21a4aa38713942fb2ebe726e6b93381ff95a0dbb427b14f804bea8"
++  },
++  {
++   "path": "benchmarks/raw/tests/fused_upgate_olmoe_2026-11-01T010747.json",
++   "bytes": 667,
++   "sha256": "3435b57f045559d2cd8e2725934e9f2a5872dea0d4ca7b30aa9004bb7f7e5ae7"
++  },
++  {
++   "path": "benchmarks/raw/tests/greedy_cap16_20261101T011813.json",
++   "bytes": 702,
++   "sha256": "c2007a57d10668d208cadd945685a93aba009472db28de8f65670200a02b51e4"
++  },
++  {
++   "path": "benchmarks/raw/tests/greedy_cap16_20261101T012306.json",
++   "bytes": 702,
++   "sha256": "c2007a57d10668d208cadd945685a93aba009472db28de8f65670200a02b51e4"
++  },
++  {
++   "path": "benchmarks/raw/tests/greedy_cap16_20261101T013103.json",
++   "bytes": 702,
++   "sha256": "c2007a57d10668d208cadd945685a93aba009472db28de8f65670200a02b51e4"
++  },
++  {
++   "path": "benchmarks/raw/tests/greedy_cap16_20261101T013347.json",
++   "bytes": 702,
++   "sha256": "c2007a57d10668d208cadd945685a93aba009472db28de8f65670200a02b51e4"
++  },
++  {
++   "path": "benchmarks/raw/tests/greedy_olmoe_cap16_20261101T010217.json",
++   "bytes": 702,
++   "sha256": "c2007a57d10668d208cadd945685a93aba009472db28de8f65670200a02b51e4"
++  },
++  {
++   "path": "benchmarks/raw/tests/greedy_olmoe_cap16_20261101T010757.json",
++   "bytes": 702,
++   "sha256": "c2007a57d10668d208cadd945685a93aba009472db28de8f65670200a02b51e4"
++  },
++  {
++   "path": "benchmarks/raw/tests/greedy_olmoe_cap16_20261101T014415.json",
++   "bytes": 702,
++   "sha256": "c2007a57d10668d208cadd945685a93aba009472db28de8f65670200a02b51e4"
++  },
++  {
++   "path": "benchmarks/raw/tests/greedy_olmoe_cap16_20261101T014903.json",
++   "bytes": 702,
++   "sha256": "c2007a57d10668d208cadd945685a93aba009472db28de8f65670200a02b51e4"
++  },
++  {
++   "path": "benchmarks/raw/tests/greedy_olmoe_cap16_20261101T015010.json",
++   "bytes": 702,
++   "sha256": "c2007a57d10668d208cadd945685a93aba009472db28de8f65670200a02b51e4"
++  },
++  {
++   "path": "benchmarks/raw/tests/greedy_olmoe_cap16_20261101T015029.json",
++   "bytes": 702,
++   "sha256": "c2007a57d10668d208cadd945685a93aba009472db28de8f65670200a02b51e4"
++  },
++  {
++   "path": "benchmarks/raw/tests/greedy_olmoe_cap16_20261101T015233.json",
++   "bytes": 702,
++   "sha256": "c2007a57d10668d208cadd945685a93aba009472db28de8f65670200a02b51e4"
++  },
++  {
++   "path": "benchmarks/raw/tests/m3u96a/greedy_qwen3_moe_cap32_20261101T014433.json",
++   "bytes": 770,
++   "sha256": "332b59d76c129d714e334fed2be68a7497287f5ae52dfaa03545d1b5270d7e81"
++  },
++  {
++   "path": "benchmarks/raw/tests/m3u96a/numerical_qwen3_moe_cap128_2026-11-01T014417.json",
++   "bytes": 1399,
++   "sha256": "b4148933a89063f8353d1e8eaf761b6ef4db6c6fbd9626ea4d05ed127a86a65a"
++  },
++  {
++   "path": "benchmarks/raw/tests/m3u96a/numerical_qwen3_moe_cap16_2026-11-01T014425.json",
++   "bytes": 1418,
++   "sha256": "9b105addcdd512206f89406b32d186152ee7dd55c661b625a9142534da8025b7"
++  },
++  {
++   "path": "benchmarks/raw/tests/m3u96a/numerical_qwen3_moe_cap32_2026-11-01T014422.json",
++   "bytes": 1415,
++   "sha256": "74ff28f6217685d3f351c309f9552a058febd914bdb72287889bf75f5a994455"
++  },
++  {
++   "path": "benchmarks/raw/tests/m3u96a/numerical_qwen3_moe_cap64_2026-11-01T014420.json",
++   "bytes": 1396,
++   "sha256": "4407c8b71bbace6ee0f109c3406f9dd7aad2fd79ffed796cfb907430d76ab221"
++  },
++  {
++   "path": "benchmarks/raw/tests/m3u96a/pytest_qwen3_m3u96a_20261101T014415.txt",
++   "bytes": 110,
++   "sha256": "7ca257be544f21c548f743d77183601f2f935f974b9e21ea51e4eba3b0f9e620"
++  },
++  {
++   "path": "benchmarks/raw/tests/numerical_cap16_2026-11-01T011806.json",
++   "bytes": 1217,
++   "sha256": "630a98566f1f73413267579580d1b884298763c63d26938332065b59ee7430af"
++  },
++  {
++   "path": "benchmarks/raw/tests/numerical_cap16_2026-11-01T012259.json",
++   "bytes": 1218,
++   "sha256": "616058446f9965335a8fd5c7108151102ad508fc4c1806ca7277f8e480814aa0"
++  },
++  {
++   "path": "benchmarks/raw/tests/numerical_cap16_2026-11-01T013056.json",
++   "bytes": 1217,
++   "sha256": "c86c870c8eac0a14c7a442cd5f7c8bf9c6f388c229328c7ae0ad7951a9e10cb7"
++  },
++  {
++   "path": "benchmarks/raw/tests/numerical_cap16_2026-11-01T013342.json",
++   "bytes": 1219,
++   "sha256": "7afe990af8de6d4ad75c0eafa3d7963bd355994b44dcaf4eb86653813df7d3c2"
++  },
++  {
++   "path": "benchmarks/raw/tests/numerical_cap32_2026-11-01T011805.json",
++   "bytes": 1196,
++   "sha256": "eab6ae12ac67790c0894b8521c0680b71297b2bd536022f0e4e96ee4254fc67f"
++  },
++  {
++   "path": "benchmarks/raw/tests/numerical_cap32_2026-11-01T012258.json",
++   "bytes": 1195,
++   "sha256": "b39c33f5f498fb6b213cafabd1cc899de3d732ff2c79212193bb6cee85797dd3"
++  },
++  {
++   "path": "benchmarks/raw/tests/numerical_cap32_2026-11-01T013055.json",
++   "bytes": 1194,
++   "sha256": "968ae034d883bc0d51f37d44063be299148d8173b8d18183b00b1c213ac41b0c"
++  },
++  {
++   "path": "benchmarks/raw/tests/numerical_cap32_2026-11-01T013341.json",
++   "bytes": 1195,
++   "sha256": "31818d37532afec50f4bafca77f64410e64c7120050524d20d5a83d422dcc447"
++  },
++  {
++   "path": "benchmarks/raw/tests/numerical_cap64_2026-11-01T011726.json",
++   "bytes": 1195,
++   "sha256": "de7cce9a373c73b7d17d939d06e2d06274170be775231f477d78c3b834070dfe"
++  },
++  {
++   "path": "benchmarks/raw/tests/numerical_cap64_2026-11-01T011753.json",
++   "bytes": 1194,
++   "sha256": "f78ec65ca4578c2a41fcc383693c15c143ca9ed82fe2788db3b19a9b67c21c7a"
++  },
++  {
++   "path": "benchmarks/raw/tests/numerical_cap64_2026-11-01T011805.json",
++   "bytes": 1194,
++   "sha256": "60a927377b1449f957e2a06a825b35f526330e44037501a2d1816dbd9edc7ef4"
++  },
++  {
++   "path": "benchmarks/raw/tests/numerical_cap64_2026-11-01T012257.json",
++   "bytes": 1195,
++   "sha256": "742fa9be90ebb23fd7a0cc1676a088cc261d9318b6a24ca73f7a2f16e00746a8"
++  },
++  {
++   "path": "benchmarks/raw/tests/numerical_cap64_2026-11-01T013054.json",
++   "bytes": 1193,
++   "sha256": "a76c618af6e31d66422179a70d40b24d6bf1b379aa515157389080e67235de2f"
++  },
++  {
++   "path": "benchmarks/raw/tests/numerical_cap64_2026-11-01T013341.json",
++   "bytes": 1195,
++   "sha256": "5ec2516c7948eeee31d12070e2f312db13c8cfe27a0be5b415a3ab4e0343aa27"
++  },
++  {
++   "path": "benchmarks/raw/tests/numerical_cap8_2026-11-01T011808.json",
++   "bytes": 1186,
++   "sha256": "a0b2a138c2ceeefd2fa7a68edbd82068c11a25ce3d777f63a95b90d4e11c26f8"
++  },
++  {
++   "path": "benchmarks/raw/tests/numerical_cap8_2026-11-01T012300.json",
++   "bytes": 1185,
++   "sha256": "0e043911673755aa445a63d414cc7d74a8785396209627764e329c81d2d44a44"
++  },
++  {
++   "path": "benchmarks/raw/tests/numerical_cap8_2026-11-01T013058.json",
++   "bytes": 1188,
++   "sha256": "68c05df7d36d0fc557de2e5f56217e345bd1e2f54a3c997c966f7a2d992833d4"
++  },
++  {
++   "path": "benchmarks/raw/tests/numerical_cap8_2026-11-01T013343.json",
++   "bytes": 1187,
++   "sha256": "8aa27d29b405e33a7ecb0f9fe5c149d354e7acbf53f852cd1752974981d9c2b5"
++  },
++  {
++   "path": "benchmarks/raw/tests/numerical_olmoe_cap16_2026-11-01T010212.json",
++   "bytes": 1415,
++   "sha256": "8e5310a4d6309011d34c258bfd6d1c4453caf87c01969317c08c013d86859a3b"
++  },
++  {
++   "path": "benchmarks/raw/tests/numerical_olmoe_cap16_2026-11-01T010752.json",
++   "bytes": 1412,
++   "sha256": "3f66a5d445adb3e47b1b0457a9047abfbb95b495da00526ca0b645b4fbef80cf"
++  },
++  {
++   "path": "benchmarks/raw/tests/numerical_olmoe_cap16_2026-11-01T014410.json",
++   "bytes": 1412,
++   "sha256": "361306d7218ed3469589751e29ebe5a84f4d25d84706680446e62436f6668458"
++  },
++  {
++   "path": "benchmarks/raw/tests/numerical_olmoe_cap16_2026-11-01T014859.json",
++   "bytes": 1414,
++   "sha256": "744d23f25de6e21f4d8351b2caf4c6fd8147ba4426910797661d726de7d11b71"
++  },
++  {
++   "path": "benchmarks/raw/tests/numerical_olmoe_cap16_2026-11-01T015006.json",
++   "bytes": 1413,
++   "sha256": "f8f8d3b5c461ba32f4ce00a47d208a99b56bc86ff6ef6cd0b3183cc8527fd51a"
++  },
++  {
++   "path": "benchmarks/raw/tests/numerical_olmoe_cap16_2026-11-01T015025.json",
++   "bytes": 1413,
++   "sha256": "f2116b815d37865061fc187ded0540ce8ce9973e3086c1a3052a68ec2afba7f8"
++  },
++  {
++   "path": "benchmarks/raw/tests/numerical_olmoe_cap16_2026-11-01T015227.json",
++   "bytes": 1413,
++   "sha256": "632b1e522c5374ab845c6d2b8e084e9644d4c09e990701b3a68290e7d2e9f6b4"
++  },
++  {
++   "path": "benchmarks/raw/tests/numerical_olmoe_cap32_2026-11-01T010212.json",
++   "bytes": 1390,
++   "sha256": "7291c489aa22237a8fdee019bb6f889ccc12b2984c4b8fc8b8afa325a9b8f6f7"
++  },
++  {
++   "path": "benchmarks/raw/tests/numerical_olmoe_cap32_2026-11-01T010751.json",
++   "bytes": 1390,
++   "sha256": "a421c4b685555f54aa1d022705153ae750688913768ce0c06ddf3e3b1a87eb57"
++  },
++  {
++   "path": "benchmarks/raw/tests/numerical_olmoe_cap32_2026-11-01T014409.json",
++   "bytes": 1388,
++   "sha256": "37d32664308b6c27662cf1eae4e772bcd1514f8687afb15638610c384fbd97f6"
++  },
++  {
++   "path": "benchmarks/raw/tests/numerical_olmoe_cap32_2026-11-01T014858.json",
++   "bytes": 1390,
++   "sha256": "c3da2303a5094c4f0444b571b789a544fcb36d6500b65ce9b2d3d49c5ecd8763"
++  },
++  {
++   "path": "benchmarks/raw/tests/numerical_olmoe_cap32_2026-11-01T015005.json",
++   "bytes": 1388,
++   "sha256": "472f2bd9c46da1bfea04f6c5618f7dfac6a2a9d07c5f716be29e4033fba6edc9"
++  },
++  {
++   "path": "benchmarks/raw/tests/numerical_olmoe_cap32_2026-11-01T015024.json",
++   "bytes": 1391,
++   "sha256": "10b96bc0ddba360d82d33b009fd29a44378d6f36276ea8ca83779798b8fa339d"
++  },
++  {
++   "path": "benchmarks/raw/tests/numerical_olmoe_cap32_2026-11-01T015226.json",
++   "bytes": 1390,
++   "sha256": "7bfcbe825814d1fb4f1f0cfacd84f2a4ed46559b4b730a8e25b57108b135b99b"
++  },
++  {
++   "path": "benchmarks/raw/tests/numerical_olmoe_cap64_2026-11-01T010211.json",
++   "bytes": 1390,
++   "sha256": "4cf7dd958e15b49c6607fc479229e560c88cf7ae90836fff82b3b950c7cfcbb3"
++  },
++  {
++   "path": "benchmarks/raw/tests/numerical_olmoe_cap64_2026-11-01T010750.json",
++   "bytes": 1388,
++   "sha256": "2ae89b499f5b3c1dbe0fbe9425d952ca39c4d04c89462fadc56f45f38b562f6f"
++  },
++  {
++   "path": "benchmarks/raw/tests/numerical_olmoe_cap64_2026-11-01T014408.json",
++   "bytes": 1389,
++   "sha256": "999202a02163fba2a070464be8173b9d0bab7bdf389fe345989cd2a38601a20b"
++  },
++  {
++   "path": "benchmarks/raw/tests/numerical_olmoe_cap64_2026-11-01T014858.json",
++   "bytes": 1389,
++   "sha256": "b05092f278608b59801f491d9891fd035ab8058f8614257556e225de0ea49127"
++  },
++  {
++   "path": "benchmarks/raw/tests/numerical_olmoe_cap64_2026-11-01T015005.json",
++   "bytes": 1391,
++   "sha256": "2c0bbafb4bf1bc43a9f293537b842bd5c431f9455c530c1e2ac1d49f2d3af44c"
++  },
++  {
++   "path": "benchmarks/raw/tests/numerical_olmoe_cap64_2026-11-01T015024.json",
++   "bytes": 1390,
++   "sha256": "d6e00c1c48ef2427f8d1d18ff121638bd41822b8b2270051c30e128b9aa97c29"
++  },
++  {
++   "path": "benchmarks/raw/tests/numerical_olmoe_cap64_2026-11-01T015225.json",
++   "bytes": 1388,
++   "sha256": "970cdd319b2db64aecbaa5524a4aac65a7cc9b7db1640b13fd9244fa5d820868"
++  },
++  {
++   "path": "benchmarks/raw/tests/numerical_olmoe_cap8_2026-11-01T010214.json",
++   "bytes": 1385,
++   "sha256": "ee15bf0f363648a116b54bcdb3d754202339cec2f836884ffe7224680bc9c923"
++  },
++  {
++   "path": "benchmarks/raw/tests/numerical_olmoe_cap8_2026-11-01T010753.json",
++   "bytes": 1385,
++   "sha256": "289e9fab112e387e12686270bc08d64e5b77368909be3b713fffcbe19929c648"
++  },
++  {
++   "path": "benchmarks/raw/tests/numerical_olmoe_cap8_2026-11-01T014411.json",
++   "bytes": 1381,
++   "sha256": "7046e5cfe912f949b4a2a33623d5536370a2140031292ccf525b69f6ab0c2608"
++  },
++  {
++   "path": "benchmarks/raw/tests/numerical_olmoe_cap8_2026-11-01T014900.json",
++   "bytes": 1383,
++   "sha256": "10f70d6e705a5fc25776f347d3bcd6b19ea0dcc20108cbbd35c0acac1811fcea"
++  },
++  {
++   "path": "benchmarks/raw/tests/numerical_olmoe_cap8_2026-11-01T015007.json",
++   "bytes": 1384,
++   "sha256": "c91aa79562cf933f1c0ec2559f91c2a10af5bb02a8039ed7b8ec4c130be2a714"
++  },
++  {
++   "path": "benchmarks/raw/tests/numerical_olmoe_cap8_2026-11-01T015026.json",
++   "bytes": 1382,
++   "sha256": "384c9a87d1e846bc416f6b14d338d61f1db67dd66f610e2a694d19116be2d869"
++  },
++  {
++   "path": "benchmarks/raw/tests/numerical_olmoe_cap8_2026-11-01T015229.json",
++   "bytes": 1381,
++   "sha256": "2bbfa8d745776ca8b9563b12f347a71dbce0ce92231ba58ca43f72fb169e7e0f"
++  },
++  {
++   "path": "benchmarks/raw/tests/pytest_gateB_20261101T011707.txt",
++   "bytes": 2371,
++   "sha256": "0373672c9480ceb3f31bb5fdebe91f860cfd543a4f99138d0aef409d9c15a8f4"
++  },
++  {
++   "path": "benchmarks/raw/tests/pytest_gateB_20261101T011725.txt",
++   "bytes": 3243,
++   "sha256": "13e841d15b27b8091df023087e165431832e5d8aa9045fca796c04804d6a5e8f"
++  },
++  {
++   "path": "benchmarks/raw/tests/pytest_gateB_20261101T011752.txt",
++   "bytes": 3216,
++   "sha256": "7d4a6868a2b6416778aeb2fb6be9ddc95f40edcea005b54c5157580edda47409"
++  },
++  {
++   "path": "benchmarks/raw/tests/pytest_gateB_20261101T011804.txt",
++   "bytes": 98,
++   "sha256": "7e968e396e06b644a263714a78329aac6c83adb5bd8af1fa18d686856e4fee92"
++  },
++  {
++   "path": "benchmarks/raw/tests/pytest_gateB_fastpath_20261101T012256.txt",
++   "bytes": 98,
++   "sha256": "dbf9c884dbb9826afa61139b56a476705007d463d2d5a57c4e981b0c3e6dd2dd"
++  },
++  {
++   "path": "figures/attribution.pdf",
++   "bytes": 20049,
++   "sha256": "3f04ea3873d433a0a4cbfa1a25355d1ed08171653d68a5e402cee6dd3f8f3a3c"
++  },
++  {
++   "path": "figures/capacity_sweep.pdf",
++   "bytes": 19212,
++   "sha256": "30137bbdeaa5c6d45be4fa75494da3d794ce0c5fc51ac0b2594a008caf4de577"
++  },
++  {
++   "path": "figures/sim_hit_vs_capacity.pdf",
++   "bytes": 24413,
++   "sha256": "92365b86a48d35b423d3da4b96595bc275262f59ab18894ece71d5a03ae03a82"
++  },
++  {
++   "path": "figures/two_node.pdf",
++   "bytes": 14785,
++   "sha256": "6480148f9056e0b03b64687c04d19d7775b17cdc53fbbe823fee6d9e3f773bf9"
++  },
++  {
++   "path": "paper/main.pdf",
++   "bytes": 454584,
++   "sha256": "fcb20dc814dc44b33359e879174156c3ddea716518e4fba8005016879fd3ee7c"
++  },
++  {
++   "path": "paper/tables/attr.tex",
++   "bytes": 1217,
++   "sha256": "8389ed69e2e03b7d7263e88a314c2b0fdb7bc6a934d0b8b62c3d76c6daccc8e6"
++  },
++  {
++   "path": "paper/tables/audit.tex",
++   "bytes": 3423,
++   "sha256": "9463ccfccccc83052b8671a6edb1af5fcd5d94ca7a439a6c39de12b594b57be2"
++  },
++  {
++   "path": "paper/tables/cold.tex",
++   "bytes": 1439,
++   "sha256": "cc9949641149298d7fa00ba67f22e615b0398ccfca768cd88a956df9431cb130"
++  },
++  {
++   "path": "paper/tables/fused.tex",
++   "bytes": 762,
++   "sha256": "87c0f13b50dab545c9b2e7960fbe46e152bb6a46e0479a6b7a0c630fe18ecb21"
++  },
++  {
++   "path": "paper/tables/qwen3.tex",
++   "bytes": 607,
++   "sha256": "0aee76729b077d2b84afb80d3955876d2fd6e15c4bf16226287d14bd7d034b62"
++  },
++  {
++   "path": "paper/tables/sim.tex",
++   "bytes": 1821,
++   "sha256": "a15f701f749f9a258305deed8bd368b7bf2b289a0ad4094c339c7c60b14450ae"
++  },
++  {
++   "path": "research/prior_art.csv",
++   "bytes": 30241,
++   "sha256": "7e3366a429d0189a09e3edf7033976f2ba859ead39a5ebe79e2f568df59b6ba3"
++  }
++ ]
++}
+ \ No newline at end of file

added benchmarks/processed/audit.json +535 −0

++{
++ "B": 10000,
++ "seed": 0,
++ "comparisons": [
++  {
++   "comparison": "laptop: TITAN all-resident (64 slots, preloaded) vs mlx-lm",
++   "n_a": 12,
++   "n_b": 12,
++   "median_a": 489.12865342395037,
++   "ci_a": [
++    488.2082407000963,
++    489.5368828457765
++   ],
++   "median_b": 495.04061243824856,
++   "ci_b": [
++    494.1194679570178,
++    495.8904708999346
++   ],
++   "diff": -5.911959014298191,
++   "ci_diff": [
++    -7.065204357666232,
++    -4.834550062056792
++   ],
++   "rel_diff_pct": -1.1942371728209773,
++   "unit": "tok/s",
++   "source": "gateB/capacity_sweep_laptop/sweep.jsonl"
++  },
++  {
++   "comparison": "laptop: TITAN 64 slots not preloaded (sync path) vs mlx-lm",
++   "n_a": 12,
++   "n_b": 12,
++   "median_a": 201.4454551722675,
++   "ci_a": [
++    200.95837279071753,
++    202.66232324191964
++   ],
++   "median_b": 495.04061243824856,
++   "ci_b": [
++    494.1194679570178,
++    495.8904708999346
++   ],
++   "diff": -293.5951572659811,
++   "ci_diff": [
++    -294.4450157276671,
++    -291.97897074353386
++   ],
++   "rel_diff_pct": -59.307287097097344,
++   "unit": "tok/s",
++   "source": "gateB/capacity_sweep_laptop/sweep.jsonl"
++  },
++  {
++   "comparison": "M3U96a Qwen3-30B: TITAN all-resident (128 slots) vs mlx-lm",
++   "n_a": 18,
++   "n_b": 18,
++   "median_a": 121.58726022750393,
++   "ci_a": [
++    121.4222850284662,
++    121.7539110305407
++   ],
++   "median_b": 121.65810731258236,
++   "ci_b": [
++    121.37584628881791,
++    121.8112669918118
++   ],
++   "diff": -0.07084708507842663,
++   "ci_diff": [
++    -0.27269819921461647,
++    0.2832330343496068
++   ],
++   "rel_diff_pct": -0.058234577738740925,
++   "unit": "tok/s",
++   "source": "gateC/qwen3_m3u96a/baseline.jsonl"
++  },
++  {
++   "comparison": "laptop cold NVMe, 16 slots: transition m=8 vs no prefetch",
++   "n_a": 12,
++   "n_b": 12,
++   "median_a": 27.904324206563125,
++   "ci_a": [
++    26.510834259463596,
++    30.51331379185264
++   ],
++   "median_b": 36.36478535014914,
++   "ci_b": [
++    34.13753044405526,
++    42.04500706287061
++   ],
++   "diff": -8.460461143586016,
++   "ci_diff": [
++    -14.521560050815605,
++    -5.0831251036633525
++   ],
++   "rel_diff_pct": -23.265533020811077,
++   "unit": "tok/s",
++   "source": "gateC/cold_nvme_laptop/sweep.jsonl"
++  },
++  {
++   "comparison": "laptop cold NVMe, 16 slots: transition2 m=8 vs no prefetch",
++   "n_a": 12,
++   "n_b": 12,
++   "median_a": 34.96599880167474,
++   "ci_a": [
++    33.67628976796817,
++    41.49003741051476
++   ],
++   "median_b": 36.36478535014914,
++   "ci_b": [
++    34.13753044405526,
++    42.04500706287061
++   ],
++   "diff": -1.3987865484743978,
++   "ci_diff": [
++    -7.324854162049796,
++    5.998514358174624
++   ],
++   "rel_diff_pct": -3.8465414686372155,
++   "unit": "tok/s",
++   "source": "gateC/cold_nvme_laptop/sweep.jsonl"
++  },
++  {
++   "comparison": "laptop cold NVMe, 16 slots: transition3 m=8 vs no prefetch",
++   "n_a": 12,
++   "n_b": 12,
++   "median_a": 29.083938320259847,
++   "ci_a": [
++    27.671125828589567,
++    34.70508437115666
++   ],
++   "median_b": 36.36478535014914,
++   "ci_b": [
++    34.13753044405526,
++    42.04500706287061
++   ],
++   "diff": -7.280847029889294,
++   "ci_diff": [
++    -13.09924827054045,
++    -1.0543321191199624
++   ],
++   "rel_diff_pct": -20.02169670405997,
++   "unit": "tok/s",
++   "source": "gateC/cold_nvme_laptop/sweep.jsonl"
++  },
++  {
++   "comparison": "laptop cold NVMe, 16 slots: transition2 m=16 vs no prefetch",
++   "n_a": 12,
++   "n_b": 12,
++   "median_a": 25.940903169028186,
++   "ci_a": [
++    25.001461847324514,
++    29.583536873911555
++   ],
++   "median_b": 36.36478535014914,
++   "ci_b": [
++    34.13753044405526,
++    42.04500706287061
++   ],
++   "diff": -10.423882181120955,
++   "ci_diff": [
++    -16.145654899217902,
++    -5.865959852413532
++   ],
++   "rel_diff_pct": -28.66477027363563,
++   "unit": "tok/s",
++   "source": "gateC/cold_nvme_laptop/sweep.jsonl"
++  },
++  {
++   "comparison": "laptop cold NVMe, 16 slots: prev_rank m=16 vs no prefetch",
++   "n_a": 12,
++   "n_b": 12,
++   "median_a": 19.13073106755353,
++   "ci_a": [
++    18.745326302296334,
++    20.005360570414403
++   ],
++   "median_b": 36.36478535014914,
++   "ci_b": [
++    34.13753044405526,
++    42.04500706287061
++   ],
++   "diff": -17.23405428259561,
++   "ci_diff": [
++    -22.91427599531708,
++    -14.833423635496317
++   ],
++   "rel_diff_pct": -47.392151821198446,
++   "unit": "tok/s",
++   "source": "gateC/cold_nvme_laptop/sweep.jsonl"
++  },
++  {
++   "comparison": "laptop cold NVMe, 32 slots: transition m=8 vs no prefetch",
++   "n_a": 12,
++   "n_b": 12,
++   "median_a": 41.32499296378025,
++   "ci_a": [
++    35.45132953777109,
++    46.87438633199691
++   ],
++   "median_b": 50.55650193124784,
++   "ci_b": [
++    46.6179394611299,
++    56.852686944161775
++   ],
++   "diff": -9.23150896746759,
++   "ci_diff": [
++    -17.486052289083403,
++    -2.598222895951629
++   ],
++   "rel_diff_pct": -18.259785813546966,
++   "unit": "tok/s",
++   "source": "gateC/cold_nvme_laptop/sweep.jsonl"
++  },
++  {
++   "comparison": "laptop cold NVMe, 32 slots: transition2 m=8 vs no prefetch",
++   "n_a": 12,
++   "n_b": 12,
++   "median_a": 42.02425447124503,
++   "ci_a": [
++    38.459575220686936,
++    46.75470157960534
++   ],
++   "median_b": 50.55650193124784,
++   "ci_b": [
++    46.6179394611299,
++    56.852686944161775
++   ],
++   "diff": -8.532247460002807,
++   "ci_diff": [
++    -15.240124741890014,
++    -2.7179076483432
++   ],
++   "rel_diff_pct": -16.876657074901807,
++   "unit": "tok/s",
++   "source": "gateC/cold_nvme_laptop/sweep.jsonl"
++  },
++  {
++   "comparison": "laptop cold NVMe, 32 slots: transition3 m=8 vs no prefetch",
++   "n_a": 12,
++   "n_b": 12,
++   "median_a": 41.02227504595441,
++   "ci_a": [
++    37.59374509091457,
++    46.645720976690825
++   ],
++   "median_b": 50.55650193124784,
++   "ci_b": [
++    46.6179394611299,
++    56.852686944161775
++   ],
++   "diff": -9.534226885293428,
++   "ci_diff": [
++    -16.24725536552569,
++    -2.9552358624899995
++   ],
++   "rel_diff_pct": -18.858557299433205,
++   "unit": "tok/s",
++   "source": "gateC/cold_nvme_laptop/sweep.jsonl"
++  },
++  {
++   "comparison": "laptop cold NVMe, 32 slots: transition2 m=16 vs no prefetch",
++   "n_a": 12,
++   "n_b": 12,
++   "median_a": 40.9858914503088,
++   "ci_a": [
++    35.4116167818508,
++    46.03429262237789
++   ],
++   "median_b": 50.55650193124784,
++   "ci_b": [
++    46.6179394611299,
++    56.852686944161775
++   ],
++   "diff": -9.570610480939038,
++   "ci_diff": [
++    -17.710608210914906,
++    -3.556491759775213
++   ],
++   "rel_diff_pct": -18.930523504086935,
++   "unit": "tok/s",
++   "source": "gateC/cold_nvme_laptop/sweep.jsonl"
++  },
++  {
++   "comparison": "laptop cold NVMe, 32 slots: prev_rank m=16 vs no prefetch",
++   "n_a": 12,
++   "n_b": 12,
++   "median_a": 28.37905550744074,
++   "ci_a": [
++    26.997499853533533,
++    32.45499196158143
++   ],
++   "median_b": 50.55650193124784,
++   "ci_b": [
++    46.6179394611299,
++    56.852686944161775
++   ],
++   "diff": -22.177446423807098,
++   "ci_diff": [
++    -28.57031143448528,
++    -16.946638006695025
++   ],
++   "rel_diff_pct": -43.86665528000013,
++   "unit": "tok/s",
++   "source": "gateC/cold_nvme_laptop/sweep.jsonl"
++  },
++  {
++   "comparison": "laptop cold NVMe, 16 slots (general): probe m=8 vs no prefetch",
++   "n_a": 4,
++   "n_b": 4,
++   "median_a": 31.841276936782968,
++   "ci_a": [
++    31.596870661689568,
++    32.84528995900421
++   ],
++   "median_b": 32.99759250772945,
++   "ci_b": [
++    32.636451601945915,
++    33.273626374779
++   ],
++   "diff": -1.1563155709464823,
++   "ci_diff": [
++    -1.5500378457589257,
++    -0.1523025487252383
++   ],
++   "rel_diff_pct": -3.5042422282038412,
++   "unit": "tok/s",
++   "source": "gateC/cold_nvme_laptop/attribution.jsonl"
++  },
++  {
++   "comparison": "laptop cold NVMe, 16 slots (general): probe m=16 vs no prefetch",
++   "n_a": 4,
++   "n_b": 4,
++   "median_a": 25.317370635052725,
++   "ci_a": [
++    24.76517223527836,
++    26.203663395431743
++   ],
++   "median_b": 32.99759250772945,
++   "ci_b": [
++    32.636451601945915,
++    33.273626374779
++   ],
++   "diff": -7.680221872676725,
++   "ci_diff": [
++    -8.398426644951499,
++    -6.752416400732631
++   ],
++   "rel_diff_pct": -23.275097632888485,
++   "unit": "tok/s",
++   "source": "gateC/cold_nvme_laptop/attribution.jsonl"
++  },
++  {
++   "comparison": "laptop cold NVMe, 16 slots (general): probe2 m=8 vs no prefetch",
++   "n_a": 4,
++   "n_b": 4,
++   "median_a": 31.34039585473687,
++   "ci_a": [
++    30.97991967772089,
++    32.02471239861806
++   ],
++   "median_b": 32.99759250772945,
++   "ci_b": [
++    32.636451601945915,
++    33.273626374779
++   ],
++   "diff": -1.6571966529925817,
++   "ci_diff": [
++    -2.1243054121081393,
++    -0.8894574043585379
++   ],
++   "rel_diff_pct": -5.022174428647772,
++   "unit": "tok/s",
++   "source": "gateC/cold_nvme_laptop/attribution.jsonl"
++  },
++  {
++   "comparison": "laptop cold NVMe, 16 slots (general): probe3 m=8 vs no prefetch",
++   "n_a": 4,
++   "n_b": 4,
++   "median_a": 30.747024815645197,
++   "ci_a": [
++    29.277490438874107,
++    32.56493375331503
++   ],
++   "median_b": 32.99759250772945,
++   "ci_b": [
++    32.636451601945915,
++    33.273626374779
++   ],
++   "diff": -2.2505676920842532,
++   "ci_diff": [
++    -3.7201020688553434,
++    -0.4326587544144189
++   ],
++   "rel_diff_pct": -6.820399674785588,
++   "unit": "tok/s",
++   "source": "gateC/cold_nvme_laptop/attribution.jsonl"
++  },
++  {
++   "comparison": "laptop cold NVMe, 16 slots (general): transition2 m=8 vs no prefetch",
++   "n_a": 4,
++   "n_b": 4,
++   "median_a": 27.12807201818323,
++   "ci_a": [
++    26.024585035633084,
++    28.245109293687104
++   ],
++   "median_b": 32.99759250772945,
++   "ci_b": [
++    32.636451601945915,
++    33.273626374779
++   ],
++   "diff": -5.869520489546222,
++   "ci_diff": [
++    -7.147820761442922,
++    -4.541891551805659
++   ],
++   "rel_diff_pct": -17.787723417007857,
++   "unit": "tok/s",
++   "source": "gateC/cold_nvme_laptop/attribution.jsonl"
++  },
++  {
++   "comparison": "laptop cold NVMe, 16 slots (general): transition2t50 m=8 vs no prefetch",
++   "n_a": 4,
++   "n_b": 4,
++   "median_a": 27.924338730957444,
++   "ci_a": [
++    26.832091171562322,
++    29.41238423146309
++   ],
++   "median_b": 32.99759250772945,
++   "ci_b": [
++    32.636451601945915,
++    33.273626374779
++   ],
++   "diff": -5.073253776772006,
++   "ci_diff": [
++    -6.165501336167129,
++    -3.5852082762663606
++   ],
++   "rel_diff_pct": -15.37461793791057,
++   "unit": "tok/s",
++   "source": "gateC/cold_nvme_laptop/attribution.jsonl"
++  },
++  {
++   "comparison": "laptop cold NVMe, 16 slots (general): transition2t70 m=8 vs no prefetch",
++   "n_a": 4,
++   "n_b": 4,
++   "median_a": 27.822276415206307,
++   "ci_a": [
++    26.652845588205146,
++    29.388223861291042
++   ],
++   "median_b": 32.99759250772945,
++   "ci_b": [
++    32.636451601945915,
++    33.273626374779
++   ],
++   "diff": -5.175316092523143,
++   "ci_diff": [
++    -6.3447469195243045,
++    -3.6093686464384085
++   ],
++   "rel_diff_pct": -15.683920247547945,
++   "unit": "tok/s",
++   "source": "gateC/cold_nvme_laptop/attribution.jsonl"
++  },
++  {
++   "comparison": "laptop H4: fused up+gate vs two kernels, 64 slots resident",
++   "n_a": 12,
++   "n_b": 12,
++   "median_a": 475.590753313536,
++   "ci_a": [
++    473.54619144727985,
++    476.23632189642444
++   ],
++   "median_b": 476.71854951152625,
++   "ci_b": [
++    475.7441354752446,
++    478.1937096893877
++   ],
++   "diff": -1.1277961979902216,
++   "ci_diff": [
++    -3.608229933155485,
++    0.07714985221900861
++   ],
++   "rel_diff_pct": -0.23657485095678102,
++   "unit": "tok/s",
++   "source": "gateD/fused_upgate_laptop.jsonl"
++  },
++  {
++   "comparison": "laptop H4: fused up+gate vs two kernels, 16 slots cold",
++   "n_a": 4,
++   "n_b": 4,
++   "median_a": 35.94742494598419,
++   "ci_a": [
++    35.731574910456565,
++    36.7122725819878
++   ],
++   "median_b": 34.98584079217508,
++   "ci_b": [
++    34.12741716795578,
++    35.49684771325741
++   ],
++   "diff": 0.9615841538091061,
++   "ci_diff": [
++    0.42788351347727627,
++    1.964214448416044
++   ],
++   "rel_diff_pct": 2.7484951970174563,
++   "unit": "tok/s",
++   "source": "gateD/fused_upgate_laptop.jsonl"
++  },
++  {
++   "comparison": "M3U96a+M3U96b OLMoE 32 slots: odd experts remote vs local",
++   "n_a": 16,
++   "n_b": 16,
++   "median_a": 15.099750302534616,
++   "ci_a": [
++    14.832823372534836,
++    15.478581724182211
++   ],
++   "median_b": 81.29106238077739,
++   "ci_b": [
++    70.45778798608657,
++    86.91246514746022
++   ],
++   "diff": -66.19131207824277,
++   "ci_diff": [
++    -71.76838200410603,
++    -55.195792166972154
++   ],
++   "rel_diff_pct": -81.42507938719571,
++   "unit": "tok/s",
++   "source": "gateE/two_node_m3u96a_m3u96b.jsonl"
++  }
++ ]
++}
+ \ No newline at end of file

modified paper/main.pdf +0 −0

Binary file not shown.

modified paper/sections/robustness.tex +8 −2

				`@@ -4,5 +4,11 @@ Held-out prompt families give hit rates within 0.01 of the fit families at every`
4	predictor's recall drops from 0.53 (fit) to 0.50 (held-out), so the locality statistics are not an artefact of the fitting	4	predictor's recall drops from 0.53 (fit) to 0.50 (held-out), so the locality statistics are not an artefact of the fitting
5	prompts. Throughput measurements on the laptop vary by less than 1\,\% across repeats for the all-resident configurations and by	5	prompts. Throughput measurements on the laptop vary by less than 1\,\% across repeats for the all-resident configurations and by
6	up to 40\,\% across prompt families for demand loading (code prompts route more repetitively than general prose, hence higher	6	up to 40\,\% across prompt families for demand loading (code prompts route more repetitively than general prose, hence higher
7	−hit rates). Thermal state and competing processes are not yet controlled in the smoke tests reported here; the final sweeps	7	+hit rates). Thermal state and competing processes are not controlled in the smoke tests; the cold-NVMe sweeps were run alone on the
8	−sample \code{powermetrics} thermal pressure every second and discard runs under throttling.	8	+laptop. Table~\ref{tab:audit} gives bootstrap confidence intervals for every comparison used in the text
		9	+(\code{scripts/audit\_stats.py}, 10\,000 resamples, seed 0, raw logs only): the all-resident parity with mlx-lm holds within
		10	+$-5.9$ [$-7.1$, $-4.8$]~\tps{} on the laptop and $-0.1$ [$-0.3$, $+0.3$] on M3U96a; every prefetch configuration's interval
		11	+excludes a gain except the depth-2 transition predictor at 16 slots, whose interval straddles zero ($-1.4$ [$-7.3$, $+6.0$]); the
		12	+fused-kernel differences are within $\pm 1$~\tps{}; and the two-node loss is $-66$ [$-72$, $-55$]~\tps{}.
		13	+
		14	+\input{tables/audit}

added paper/tables/audit.tex +34 −0

++\begin{table}[t]
++\centering\scriptsize
++\caption{Statistical audit: medians with bootstrap 95\,\% CIs (10\,000 resamples) of decode throughput (tok/s) and of the difference A $-$ B.}
++\label{tab:audit}
++\begin{tabular}{L{7.2cm}rrrr}
++\toprule
++Comparison (A vs B) & $n_A/n_B$ & A [CI] & B [CI] & A$-$B [CI] \\
++\midrule
++laptop: TITAN all-resident (64 slots, preloaded) vs mlx-lm & 12/12 & 489.1 [488.2, 489.5] & 495.0 [494.1, 495.9] & -5.9 [-7.1, -4.8] \\
++laptop: TITAN 64 slots not preloaded (sync path) vs mlx-lm & 12/12 & 201.4 [201.0, 202.7] & 495.0 [494.1, 495.9] & -293.6 [-294.4, -292.0] \\
++M3U96a Qwen3-30B: TITAN all-resident (128 slots) vs mlx-lm & 18/18 & 121.6 [121.4, 121.8] & 121.7 [121.4, 121.8] & -0.1 [-0.3, +0.3] \\
++laptop cold NVMe, 16 slots: transition m=8 vs no prefetch & 12/12 & 27.9 [26.5, 30.5] & 36.4 [34.1, 42.0] & -8.5 [-14.5, -5.1] \\
++laptop cold NVMe, 16 slots: transition2 m=8 vs no prefetch & 12/12 & 35.0 [33.7, 41.5] & 36.4 [34.1, 42.0] & -1.4 [-7.3, +6.0] \\
++laptop cold NVMe, 16 slots: transition3 m=8 vs no prefetch & 12/12 & 29.1 [27.7, 34.7] & 36.4 [34.1, 42.0] & -7.3 [-13.1, -1.1] \\
++laptop cold NVMe, 16 slots: transition2 m=16 vs no prefetch & 12/12 & 25.9 [25.0, 29.6] & 36.4 [34.1, 42.0] & -10.4 [-16.1, -5.9] \\
++laptop cold NVMe, 16 slots: prev\_rank m=16 vs no prefetch & 12/12 & 19.1 [18.7, 20.0] & 36.4 [34.1, 42.0] & -17.2 [-22.9, -14.8] \\
++laptop cold NVMe, 32 slots: transition m=8 vs no prefetch & 12/12 & 41.3 [35.5, 46.9] & 50.6 [46.6, 56.9] & -9.2 [-17.5, -2.6] \\
++laptop cold NVMe, 32 slots: transition2 m=8 vs no prefetch & 12/12 & 42.0 [38.5, 46.8] & 50.6 [46.6, 56.9] & -8.5 [-15.2, -2.7] \\
++laptop cold NVMe, 32 slots: transition3 m=8 vs no prefetch & 12/12 & 41.0 [37.6, 46.6] & 50.6 [46.6, 56.9] & -9.5 [-16.2, -3.0] \\
++laptop cold NVMe, 32 slots: transition2 m=16 vs no prefetch & 12/12 & 41.0 [35.4, 46.0] & 50.6 [46.6, 56.9] & -9.6 [-17.7, -3.6] \\
++laptop cold NVMe, 32 slots: prev\_rank m=16 vs no prefetch & 12/12 & 28.4 [27.0, 32.5] & 50.6 [46.6, 56.9] & -22.2 [-28.6, -16.9] \\
++laptop cold NVMe, 16 slots (general): probe m=8 vs no prefetch & 4/4 & 31.8 [31.6, 32.8] & 33.0 [32.6, 33.3] & -1.2 [-1.6, -0.2] \\
++laptop cold NVMe, 16 slots (general): probe m=16 vs no prefetch & 4/4 & 25.3 [24.8, 26.2] & 33.0 [32.6, 33.3] & -7.7 [-8.4, -6.8] \\
++laptop cold NVMe, 16 slots (general): probe2 m=8 vs no prefetch & 4/4 & 31.3 [31.0, 32.0] & 33.0 [32.6, 33.3] & -1.7 [-2.1, -0.9] \\
++laptop cold NVMe, 16 slots (general): probe3 m=8 vs no prefetch & 4/4 & 30.7 [29.3, 32.6] & 33.0 [32.6, 33.3] & -2.3 [-3.7, -0.4] \\
++laptop cold NVMe, 16 slots (general): transition2 m=8 vs no prefetch & 4/4 & 27.1 [26.0, 28.2] & 33.0 [32.6, 33.3] & -5.9 [-7.1, -4.5] \\
++laptop cold NVMe, 16 slots (general): transition2t50 m=8 vs no prefetch & 4/4 & 27.9 [26.8, 29.4] & 33.0 [32.6, 33.3] & -5.1 [-6.2, -3.6] \\
++laptop cold NVMe, 16 slots (general): transition2t70 m=8 vs no prefetch & 4/4 & 27.8 [26.7, 29.4] & 33.0 [32.6, 33.3] & -5.2 [-6.3, -3.6] \\
++laptop H4: fused up+gate vs two kernels, 64 slots resident & 12/12 & 475.6 [473.5, 476.2] & 476.7 [475.7, 478.2] & -1.1 [-3.6, +0.1] \\
++laptop H4: fused up+gate vs two kernels, 16 slots cold & 4/4 & 35.9 [35.7, 36.7] & 35.0 [34.1, 35.5] & +1.0 [+0.4, +2.0] \\
++M3U96a+M3U96b OLMoE 32 slots: odd experts remote vs local & 16/16 & 15.1 [14.8, 15.5] & 81.3 [70.5, 86.9] & -66.2 [-71.8, -55.2] \\
++\bottomrule
++\end{tabular}
++\end{table}

added scripts/audit_stats.py +81 −0

++#!/usr/bin/env python3
++"""Gate G statistical audit: bootstrap 95% confidence intervals (10 000 resamples, seed 0) of median decode throughput per
++configuration and of paired/unpaired differences for the pre-registered comparisons. Reads only raw JSONL logs. Writes
++benchmarks/processed/audit.json and paper/tables/audit.tex."""
++import json, os, glob, random, statistics
++ROOT = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
++rng = random.Random(0); B = 10000
++
++def load(path):
++    return [json.loads(l) for l in open(path) if l.strip()]
++
++def boot_median(xs):
++    meds = []
++    for _ in range(B):
++        s = [xs[rng.randrange(len(xs))] for _ in xs]; meds.append(statistics.median(s))
++    meds.sort(); return statistics.median(xs), meds[int(0.025 * B)], meds[int(0.975 * B)]
++
++def boot_diff(a, b):
++    ds = []
++    for _ in range(B):
++        sa = [a[rng.randrange(len(a))] for _ in a]; sb = [b[rng.randrange(len(b))] for _ in b]
++        ds.append(statistics.median(sa) - statistics.median(sb))
++    ds.sort(); return statistics.median(a) - statistics.median(b), ds[int(0.025 * B)], ds[int(0.975 * B)]
++
++def sel(recs, **kw):
++    out = []
++    for r in recs:
++        if r.get("warmup"): continue
++        if all((r.get(k) == v) if not callable(v) else v(r.get(k)) for k, v in kw.items()): out.append(r)
++    return out
++
++comparisons = []
++def add(name, a, b, field="gen_tps", unit="tok/s", src=""):
++    xa = [r[field] for r in a if r.get(field) is not None]; xb = [r[field] for r in b if r.get(field) is not None]
++    if not xa or not xb: return
++    ma, la, ha = boot_median(xa); mb, lb, hb = boot_median(xb); d, ld, hd = boot_diff(xa, xb)
++    comparisons.append({"comparison": name, "n_a": len(xa), "n_b": len(xb), "median_a": ma, "ci_a": [la, ha], "median_b": mb, "ci_b": [lb, hb],
++                        "diff": d, "ci_diff": [ld, hd], "rel_diff_pct": 100 * d / mb if mb else None, "unit": unit, "source": src})
++
++sw = load(f"{ROOT}/benchmarks/raw/gateB/capacity_sweep_laptop/sweep.jsonl")
++add("laptop: TITAN all-resident (64 slots, preloaded) vs mlx-lm", sel(sw, backend="titan", capacity=64, tag="warmall"), sel(sw, backend="mlx_lm"), src="gateB/capacity_sweep_laptop/sweep.jsonl")
++add("laptop: TITAN 64 slots not preloaded (sync path) vs mlx-lm", sel(sw, backend="titan", capacity=64, tag=""), sel(sw, backend="mlx_lm"), src="gateB/capacity_sweep_laptop/sweep.jsonl")
++q = f"{ROOT}/benchmarks/raw/gateC/qwen3_m3u96a/baseline.jsonl"
++if os.path.exists(q):
++    qq = load(q); add("M3U96a Qwen3-30B: TITAN all-resident (128 slots) vs mlx-lm", sel(qq, backend="titan"), sel(qq, backend="mlx_lm"), src="gateC/qwen3_m3u96a/baseline.jsonl")
++cold = load(f"{ROOT}/benchmarks/raw/gateC/cold_nvme_laptop/sweep.jsonl")
++for cap in (16, 32):
++    base = sel(cold, capacity=cap, predictor="none")
++    for pr, m in (("transition", 8), ("transition2", 8), ("transition3", 8), ("transition2", 16), ("prev_rank", 16)):
++        add(f"laptop cold NVMe, {cap} slots: {pr} m={m} vs no prefetch", sel(cold, capacity=cap, predictor=pr, m=m), base, src="gateC/cold_nvme_laptop/sweep.jsonl")
++attr = load(f"{ROOT}/benchmarks/raw/gateC/cold_nvme_laptop/attribution.jsonl")
++base = sel(attr, predictor="none")
++for pr, m in (("probe", 8), ("probe", 16), ("probe2", 8), ("probe3", 8), ("transition2", 8), ("transition2t50", 8), ("transition2t70", 8)):
++    add(f"laptop cold NVMe, 16 slots (general): {pr} m={m} vs no prefetch", sel(attr, predictor=pr, m=m), base, src="gateC/cold_nvme_laptop/attribution.jsonl")
++fz = f"{ROOT}/benchmarks/raw/gateD/fused_upgate_laptop.jsonl"
++if os.path.exists(fz):
++    f = load(fz)
++    add("laptop H4: fused up+gate vs two kernels, 64 slots resident", sel(f, capacity=64, fused_upgate=True), sel(f, capacity=64, fused_upgate=False), src="gateD/fused_upgate_laptop.jsonl")
++    add("laptop H4: fused up+gate vs two kernels, 16 slots cold", sel(f, capacity=16, fused_upgate=True), sel(f, capacity=16, fused_upgate=False), src="gateD/fused_upgate_laptop.jsonl")
++tn = f"{ROOT}/benchmarks/raw/gateE/two_node_m3u96a_m3u96b.jsonl"
++if os.path.exists(tn):
++    t = load(tn)
++    add("M3U96a+M3U96b OLMoE 32 slots: odd experts remote vs local", sel(t, capacity=32, placement="odd"), sel(t, capacity=32, placement="none"), src="gateE/two_node_m3u96a_m3u96b.jsonl")
++gf = f"{ROOT}/benchmarks/raw/gateF/two_node_qwen3_m3u96a_m3u96b.jsonl"
++if os.path.exists(gf):
++    g = load(gf)
++    for cap in (32, 64):
++        base = sel(g, capacity=cap, placement="none")
++        for pl in ("odd", "popular:benchmarks/raw/gateF/traces_qwen3_m3u96a"):
++            add(f"M3U96a+M3U96b Qwen3-30B {cap} slots cold: {pl.split(':')[0]} placement vs local cold NVMe", sel(g, capacity=cap, placement=pl), base, src="gateF/two_node_qwen3_m3u96a_m3u96b.jsonl")
++os.makedirs(f"{ROOT}/benchmarks/processed", exist_ok=True)
++json.dump({"B": B, "seed": 0, "comparisons": comparisons}, open(f"{ROOT}/benchmarks/processed/audit.json", "w"), indent=1)
++os.makedirs(f"{ROOT}/paper/tables", exist_ok=True)
++L = ["\\begin{table}[t]", "\\centering\\scriptsize", "\\caption{Statistical audit: medians with bootstrap 95\\,\\% CIs (10\\,000 resamples) of decode throughput (tok/s) and of the difference A $-$ B.}", "\\label{tab:audit}",
++     "\\begin{tabular}{L{7.2cm}rrrr}", "\\toprule", "Comparison (A vs B) & $n_A/n_B$ & A [CI] & B [CI] & A$-$B [CI] \\\\", "\\midrule"]
++for c in comparisons:
++    L.append(f"{c['comparison'].replace('_', '\\_').replace('%', '\\%')} & {c['n_a']}/{c['n_b']} & {c['median_a']:.1f} [{c['ci_a'][0]:.1f}, {c['ci_a'][1]:.1f}] & {c['median_b']:.1f} [{c['ci_b'][0]:.1f}, {c['ci_b'][1]:.1f}] & {c['diff']:+.1f} [{c['ci_diff'][0]:+.1f}, {c['ci_diff'][1]:+.1f}] \\\\")
++L += ["\\bottomrule", "\\end{tabular}", "\\end{table}"]
++open(f"{ROOT}/paper/tables/audit.tex", "w").write("\n".join(L) + "\n")
++for c in comparisons:
++    print(f"{c['comparison'][:70]:70s} A={c['median_a']:6.1f} [{c['ci_a'][0]:6.1f},{c['ci_a'][1]:6.1f}]  B={c['median_b']:6.1f}  diff={c['diff']:+6.1f} [{c['ci_diff'][0]:+6.1f},{c['ci_diff'][1]:+6.1f}]")

added scripts/make_manifest.py +23 −0

++#!/usr/bin/env python3
++"""artifacts/manifest.json: sha256 and size of every raw log, processed result, figure and table, plus the git commit and the
++model revisions — the reproducibility package index (Gate G)."""
++import hashlib, json, os, subprocess, time
++ROOT = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
++def sha(p):
++    h = hashlib.sha256()
++    with open(p, "rb") as f:
++        for b in iter(lambda: f.read(1 << 20), b""): h.update(b)
++    return h.hexdigest()
++files = []
++for sub in ("benchmarks/raw", "benchmarks/processed", "figures", "paper/tables", "paper/main.pdf", "artifacts/environment.json", "research/prior_art.csv"):
++    p = os.path.join(ROOT, sub)
++    if os.path.isfile(p): files.append(p)
++    else:
++        for d, _, fs in os.walk(p):
++            files += [os.path.join(d, f) for f in fs if not f.startswith(".")]
++man = {"generated": time.strftime("%Y-%m-%dT%H:%M:%S"), "git": subprocess.run(["git", "rev-parse", "HEAD"], capture_output=True, text=True, cwd=ROOT).stdout.strip(),
++       "models": {"olmoe-1b-7b-0125-4bit": "mlx-community/OLMoE-1B-7B-0125-Instruct-4bit@e23844197887b031e7ddddbb0b8959c5a6853a7b",
++                  "qwen3-30b-a3b-2507-4bit": "mlx-community/Qwen3-30B-A3B-Instruct-2507-4bit@e9675aa3ca5f900ccef55267914466d55ab325fa"},
++       "files": [{"path": os.path.relpath(p, ROOT), "bytes": os.path.getsize(p), "sha256": sha(p)} for p in sorted(files)]}
++json.dump(man, open(os.path.join(ROOT, "artifacts", "manifest.json"), "w"), indent=1)
++print(len(man["files"]), "files,", sum(f["bytes"] for f in man["files"]) / 1e6, "MB")

spb/wp33_uqo

Gate G: bootstrap statistical audit (table in paper), manifest generator

Theme