Dataset Preview
Duplicate
The full dataset viewer is not available (click to read why). Only showing a preview of the rows.
The dataset generation failed
Error code:   DatasetGenerationError
Exception:    TypeError
Message:      int() argument must be a string, a bytes-like object or a real number, not 'NoneType'
Traceback:    Traceback (most recent call last):
                File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1539, in _prepare_split_single
                  for key, record in generator:
                                     ^^^^^^^^^
                File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 621, in wrapped
                  for item in generator(*args, **kwargs):
                              ~~~~~~~~~^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/webdataset/webdataset.py", line 131, in _generate_examples
                  for example_idx, example in enumerate(self._get_pipeline_from_tar(tar_path, tar_iterator)):
                                              ~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/webdataset/webdataset.py", line 32, in _get_pipeline_from_tar
                  for filename, f in tar_iterator:
                                     ^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/utils/track.py", line 49, in __iter__
                  for x in self.generator(*self.args):
                           ~~~~~~~~~~~~~~^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/utils/file_utils.py", line 1419, in _iter_from_urlpath
                  with xopen(urlpath, "rb", download_config=download_config, block_size=0) as f:
                       ~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/utils/file_utils.py", line 976, in xopen
                  file_obj = fs.open(paths[0], mode)
                File "<string>", line 3, in open
                File "/usr/local/lib/python3.14/unittest/mock.py", line 1176, in __call__
                  return self._mock_call(*args, **kwargs)
                         ~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/unittest/mock.py", line 1180, in _mock_call
                  return self._execute_mock_call(*args, **kwargs)
                         ~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/unittest/mock.py", line 1247, in _execute_mock_call
                  result = effect(*args, **kwargs)
                File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 794, in wrapped
                  tracker.files[urlpath] = {"read": 0, "size": int(f.size)}
                                                               ~~~^^^^^^^^
              TypeError: int() argument must be a string, a bytes-like object or a real number, not 'NoneType'
              
              The above exception was the direct cause of the following exception:
              
              Traceback (most recent call last):
                File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 1379, in compute_config_parquet_and_info_response
                  parquet_operations, partial, estimated_dataset_info = stream_convert_to_parquet(
                                                                        ~~~~~~~~~~~~~~~~~~~~~~~~~^
                      builder, max_dataset_size_bytes=max_dataset_size_bytes
                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                  )
                  ^
                File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 956, in stream_convert_to_parquet
                  builder._prepare_split(split_generator=splits_generators[split], file_format="parquet")
                  ~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1401, in _prepare_split
                  for job_id, done, content in self._prepare_split_single(
                                               ~~~~~~~~~~~~~~~~~~~~~~~~~~^
                      gen_kwargs=gen_kwargs, job_id=job_id, **_prepare_split_args
                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                  ):
                  ^
                File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1579, in _prepare_split_single
                  raise DatasetGenerationError("An error occurred while generating the dataset") from e
              datasets.exceptions.DatasetGenerationError: An error occurred while generating the dataset

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

json.gz
list
__key__
string
__url__
string
[ 9, 24, 13, 24, 10, 714, 714, 714, 714, 714, 714, 427, 427, 427, 427, 187, 9, 23, 13, 23, 10, 714, 714, 714, 714, 714, 427, 427, 427, 427, 427, 187, 9, 22, 13, 22, 10, 427, 427, 427, 427, 427, 427, 427, 427, 427, 427, 187, ...
00000000_00000000_057d8a28-cfaf-4600-9a4c-39c8640e5b76
hf://datasets/jerryyan/crawl-pretrain@b243b44f0d68440d974e47c150b6fec6bcaf5697/2026-10-26/common_crawl_baseline_processed_14_4b/shard_00000000.tar
[ 187, 187, 34, 27, 11793, 14980, 25978, 2506, 457, 85, 3523, 387, 253, 1566, 15, 733, 1364, 3986, 253, 2718, 835, 941, 4852, 285, 43279, 15, 443, 1189, 920, 253, 5447, 3828, 20096, 1046, 14980, 14, 17477, 1818, 310, 10711, 494, 13, 2...
00000000_00000001_f72ccb33-5dbd-453b-bd01-535de64cf49d
hf://datasets/jerryyan/crawl-pretrain@b243b44f0d68440d974e47c150b6fec6bcaf5697/2026-10-26/common_crawl_baseline_processed_14_4b/shard_00000000.tar
[ 14, 8547, 16, 42532, 25379, 64, 5397, 15, 46415, 1873, 4428, 16012, 323, 1016, 18329, 15722, 13, 1690, 253, 1511, 273, 15722, 313, 74, 15, 70, 904, 496, 3219, 390, 657, 2088, 582, 253, 9632, 313, 74, 15, 70, 904, 322, 20, 390, 391...
00000000_00000002_c1d69798-a16a-4ef1-b569-f9cdb03e1b9f
hf://datasets/jerryyan/crawl-pretrain@b243b44f0d68440d974e47c150b6fec6bcaf5697/2026-10-26/common_crawl_baseline_processed_14_4b/shard_00000000.tar
[ 285, 44559, 2300, 1194, 1864, 13, 285, 8138, 4849, 7281, 84, 13, 403, 7470, 323, 897, 1309, 253, 5909, 3038, 273, 22777, 9189, 15, 1583, 403, 4158, 281, 10196, 275, 12620, 835, 952, 403, 1246, 13, 347, 597, 403, 4158, 281, 320, 7107...
00000000_00000003_36554a10-9a3e-4150-a576-02cef11707be
hf://datasets/jerryyan/crawl-pretrain@b243b44f0d68440d974e47c150b6fec6bcaf5697/2026-10-26/common_crawl_baseline_processed_14_4b/shard_00000000.tar
[ 407, 14604, 64, 3281, 1955, 281, 253, 1039, 941, 310, 7141, 275, 44864, 6391, 15, 187, 187, 47416, 4373, 22041, 187, 187, 1394, 476, 36219, 253, 11454, 2990, 10336, 407, 19427, 253, 1180, 273, 8763, 8090, 313, 79, 64, 19057, 582, 253,...
00000000_00000004_d9f7f979-05aa-4e88-93dc-134e4580617d
hf://datasets/jerryyan/crawl-pretrain@b243b44f0d68440d974e47c150b6fec6bcaf5697/2026-10-26/common_crawl_baseline_processed_14_4b/shard_00000000.tar
[ 85, 2810, 18996, 1128, 15617, 273, 534, 2847, 253, 2798, 281, 789, 12150, 685, 7312, 281, 755, 2614, 562, 281, 253, 2133, 15, 187, 187, 9586, 247, 6283, 4447, 22678, 11059, 476, 1818, 347, 359, 2363, 13, 390, 320, 13572, 984, 273, 1...
00000000_00000005_1504bf75-d5c6-4d4d-850b-afde484fcbfb
hf://datasets/jerryyan/crawl-pretrain@b243b44f0d68440d974e47c150b6fec6bcaf5697/2026-10-26/common_crawl_baseline_processed_14_4b/shard_00000000.tar
[ 5559, 32664, 14980, 14, 33527, 17215, 32, 187, 187, 4374, 15, 5559, 556, 13644, 4767, 326, 14980, 14, 20419, 2600, 310, 4136, 13, 2530, 326, 253, 2600, 310, 9371, 285, 3542, 8558, 323, 952, 15, 187, 187, 17664, 457, 84, 3186, 9600, ...
00000000_00000006_3eac3112-0db0-4829-8f13-e333e5d62680
hf://datasets/jerryyan/crawl-pretrain@b243b44f0d68440d974e47c150b6fec6bcaf5697/2026-10-26/common_crawl_baseline_processed_14_4b/shard_00000000.tar
[ 444, 4576, 13, 11936, 56, 74, 7864, 1108, 11936, 14, 1374, 272, 36181, 7327, 14, 1171, 14, 1374, 322, 11313, 285, 39823, 1830, 1108, 4559, 5034, 36347, 33034, 17143, 6045, 11282, 11955, 432, 18228, 13, 19048, 6453, 1108, 9773, 917, 496,...
00000000_00000007_1dcc4b10-af3a-4dac-b691-d163586f88b6
hf://datasets/jerryyan/crawl-pretrain@b243b44f0d68440d974e47c150b6fec6bcaf5697/2026-10-26/common_crawl_baseline_processed_14_4b/shard_00000000.tar
[ 31733, 273, 253, 4484, 2329, 26575, 689, 619, 787, 25462, 2454, 15, 309, 3047, 6114, 875, 619, 4669, 15, 401, 11822, 1481, 7053, 13, 309, 1869, 42708, 15, 1310, 436, 369, 13889, 352, 369, 2717, 281, 320, 9202, 273, 13, 760, 247, 171...
00000000_00000008_f8c64df1-bdad-44b3-8f26-84eb70fad446
hf://datasets/jerryyan/crawl-pretrain@b243b44f0d68440d974e47c150b6fec6bcaf5697/2026-10-26/common_crawl_baseline_processed_14_4b/shard_00000000.tar
[ 6322, 6510, 326, 2792, 281, 247, 1852, 27423, 2245, 273, 247, 41288, 13, 368, 3365, 897, 432, 43118, 586, 10983, 10244, 13, 323, 31238, 247, 6322, 12095, 326, 2792, 281, 271, 1852, 21294, 2245, 273, 247, 41288, 13, 368, 476, 897, 281,...
00000000_00000009_80fa098f-0c33-4cf3-91f5-a4793a1c56b4
hf://datasets/jerryyan/crawl-pretrain@b243b44f0d68440d974e47c150b6fec6bcaf5697/2026-10-26/common_crawl_baseline_processed_14_4b/shard_00000000.tar
[ 9338, 15, 187, 187, 1628, 20696, 11172, 2289, 13, 627, 310, 642, 11286, 13, 285, 642, 4839, 387, 1495, 1806, 2296, 8784, 5210, 73, 596, 518, 263, 13, 10399, 13, 23011, 13, 8188, 10364, 342, 253, 44039, 5383, 26743, 7754, 10537, 2285, ...
00000000_00000010_d2f3005f-96bb-4b69-b1df-09e2b98a2fc7
hf://datasets/jerryyan/crawl-pretrain@b243b44f0d68440d974e47c150b6fec6bcaf5697/2026-10-26/common_crawl_baseline_processed_14_4b/shard_00000000.tar
[ 352, 943, 1379, 15, 586, 10983, 27120, 281, 1555, 2515, 973, 3345, 253, 1948, 48603, 985, 15, 586, 10983, 914, 88, 3867, 1255, 281, 3730, 368, 11358, 762, 667, 5989, 15, 187, 187, 2347, 1048, 1078, 309, 871, 604, 352, 310, 2444, 32,...
00000000_00000011_0468c50b-29a4-4a88-8618-02aef8172194
hf://datasets/jerryyan/crawl-pretrain@b243b44f0d68440d974e47c150b6fec6bcaf5697/2026-10-26/common_crawl_baseline_processed_14_4b/shard_00000000.tar
[ 3630, 13, 1014, 2167, 368, 1053, 457, 85, 1663, 871, 670, 352, 390, 5046, 368, 457, 250, 417, 4680, 670, 352, 13, 3707, 1024, 368, 513, 871, 670, 352, 984, 368, 8028, 281, 436, 9037, 15, 187, 187, 60, 361, 27, 1797, 27, 746, 62,...
00000000_00000012_ba240ab3-8636-4780-9540-64f3c464f3c9
hf://datasets/jerryyan/crawl-pretrain@b243b44f0d68440d974e47c150b6fec6bcaf5697/2026-10-26/common_crawl_baseline_processed_14_4b/shard_00000000.tar
[ 2159, 14, 3945, 11947, 9091, 326, 7407, 29262, 5724, 8244, 15, 659, 4121, 2879, 281, 253, 3605, 6427, 11280, 432, 16864, 444, 13837, 15071, 665, 3540, 253, 22791, 15, 659, 4121, 5176, 2454, 10156, 3473, 323, 253, 1072, 1921, 15, 1496, ...
00000000_00000013_5881ea1c-b985-4b46-894a-79cdf83cb51a
hf://datasets/jerryyan/crawl-pretrain@b243b44f0d68440d974e47c150b6fec6bcaf5697/2026-10-26/common_crawl_baseline_processed_14_4b/shard_00000000.tar
[ 15, 16404, 253, 1673, 327, 1165, 20, 2289, 1078, 253, 1735, 1265, 5149, 15, 187, 187, 1276, 359, 1119, 187, 187, 1542, 253, 31140, 1421, 586, 337, 15, 19263, 9058, 15, 11075, 84, 6497, 432, 337, 13, 14028, 281, 337, 13, 8196, 247, ...
00000000_00000014_c9dc58db-52e5-4292-a091-a251720f6892
hf://datasets/jerryyan/crawl-pretrain@b243b44f0d68440d974e47c150b6fec6bcaf5697/2026-10-26/common_crawl_baseline_processed_14_4b/shard_00000000.tar
[ 897, 2219, 273, 512, 273, 1110, 1146, 8379, 908, 987, 1024, 534, 310, 1027, 432, 436, 2934, 326, 627, 457, 84, 1469, 281, 320, 247, 10721, 2181, 3304, 634, 4382, 326, 816, 1057, 634, 2628, 15, 187, 187, 55, 47, 27, 1737, 9118, 84,...
00000000_00000015_255392c2-5e26-49f3-bd03-841cda02ca06
hf://datasets/jerryyan/crawl-pretrain@b243b44f0d68440d974e47c150b6fec6bcaf5697/2026-10-26/common_crawl_baseline_processed_14_4b/shard_00000000.tar
[ 285, 671, 634, 1442, 17988, 19358, 323, 253, 4067, 3394, 824, 347, 253, 444, 893, 365, 3443, 11997, 15, 25330, 187, 52, 7694, 378, 675, 248, 285, 48678, 5429, 187, 17737, 1384, 1099, 407, 378, 675, 248, 35, 2804, 15, 681, 44067, 45,...
00000000_00000016_951de5ef-b7cd-4dfe-aa48-4dbd3ea24135
hf://datasets/jerryyan/crawl-pretrain@b243b44f0d68440d974e47c150b6fec6bcaf5697/2026-10-26/common_crawl_baseline_processed_14_4b/shard_00000000.tar
[ 1694, 1110, 5044, 12783, 10141, 13, 627, 434, 581, 625, 2181, 4409, 6240, 27, 247, 1652, 1029, 14, 38172, 40992, 15, 49042, 111, 187, 187, 4125, 13, 1078, 368, 2810, 436, 4579, 285, 1616, 479, 745, 347, 773, 1439, 14, 1542, 14, 5658...
00000000_00000017_7551a834-3a25-4480-b25c-9fd659cb3a26
hf://datasets/jerryyan/crawl-pretrain@b243b44f0d68440d974e47c150b6fec6bcaf5697/2026-10-26/common_crawl_baseline_processed_14_4b/shard_00000000.tar
[ 380, 3946, 5644, 281, 1534, 11784, 11655, 323, 13001, 285, 10953, 3013, 253, 12510, 273, 1345, 3238, 15, 586, 10983, 2006, 42527, 281, 3270, 14, 29706, 49270, 5323, 310, 9560, 323, 11850, 5213, 5054, 7688, 285, 598, 16514, 247, 2586, 45...
00000000_00000018_aa2d35ff-cadc-4ae2-9a79-7f27008cfe98
hf://datasets/jerryyan/crawl-pretrain@b243b44f0d68440d974e47c150b6fec6bcaf5697/2026-10-26/common_crawl_baseline_processed_14_4b/shard_00000000.tar
[ 5893, 2561, 15, 187, 187, 5703, 625, 275, 776, 18196, 5074, 37351, 12234, 187, 187, 20865, 253, 48762, 459, 187, 39, 10719, 12234, 187, 187, 8665, 5300, 187, 187, 44, 54, 2070, 86, 1261, 187, 28743, 7668, 9965, 5074, 187, 27748, 5399,...
00000000_00000019_3bbabf2f-b764-4324-8867-f73d0bce623a
hf://datasets/jerryyan/crawl-pretrain@b243b44f0d68440d974e47c150b6fec6bcaf5697/2026-10-26/common_crawl_baseline_processed_14_4b/shard_00000000.tar
[ 43363, 285, 23790, 91, 1342, 2335, 247, 1007, 1475, 387, 253, 15573, 24674, 13016, 15, 19760, 369, 642, 3356, 247, 9058, 1178, 350, 15, 380, 4491, 285, 8446, 574, 4395, 327, 15, 1292, 752, 574, 253, 8446, 4395, 327, 281, 32, 187, 18...
00000000_00000020_5aacbf7f-e35f-4593-b1e8-afa30c676c04
hf://datasets/jerryyan/crawl-pretrain@b243b44f0d68440d974e47c150b6fec6bcaf5697/2026-10-26/common_crawl_baseline_processed_14_4b/shard_00000000.tar
End of preview.

crawl-pretrain

Tokenized pretraining data from crawl simulations on the production crawl graph (job 148132, graph v3), Crawl4LLM project, CMU. Each simulation changes only the crawl priority (which URL is fetched first); everything else is fixed: at most 250 pages per registered domain per cycle, same graph, seeds (seed 42) and max 100 outlinks per page.

Every folder <experiment date>/<experiment>/ holds one simulation's selected pages after full processing: English filter β†’ DCLM RefinedWeb rules β†’ BFF near-dup removal β†’ DCLM fastText filter (threshold 0.018112), tokenized with tokshuf-rust (EleutherAI/gpt-neox-20b via tiktoken, seqlen 2049) as webdataset shard_*.tar + manifest.jsonl, ready for DCLM / open_lm 1B-1x training.

Status: uploaded = complete in this repo; pending = still simulating or processing, folder added when tokenized.

Results (DCLM 1B-1x, 28.8B training tokens, Core = centered accuracy over the low-variance tasks without commonsense_qa)

folder Commonsense Reasoning Language Understanding Reading Comprehension Symbolic Problem Solving World Knowledge Core
2026-10-27/sim_233150_source_quality_cash 0.33256 0.37992 0.26287 0.20513 0.30070 0.29977
2026-10-29/sim_233150_opic_content_trust_decay 0.38559 0.37873 0.27001 0.17054 0.30646 0.30110
2026-10-29/sim_233150_opic_content_template_split 0.36602 0.37097 0.22619 0.19049 0.29642 0.29259
2026-10-29/sim_233150_opic_content_self_damp 0.34941 0.36451 0.20528 0.18064 0.28735 0.28141
2026-11-02/sim_2m_host_explore (2M setting, fastText threshold version) 0.33684 0.32389 0.24529 0.20221 0.28221 0.27781
2026-11-02/sim_2m_head_tail (2M setting, fastText threshold version) 0.33221 0.33506 0.21993 0.19018 0.28504 0.27468
2026-10-29/sim_233150_inlink_content_domain_once 0.34817 0.34998 0.27306 0.22121 0.29514 0.29751
2026-10-29/sim_233150_inlink_content_template_once 0.35944 0.36181 0.25579 0.20663 0.29576 0.29675
2026-10-27/sim_233150_quality_density 0.33988 0.36631 0.25677 0.20270 0.29853 0.29518
2026-10-29/sim_233150_domain_prior_full 0.33594 0.36640 0.27076 0.19242 0.29418 0.29325
2026-10-29/sim_233150_inlink_content 0.33765 0.35172 0.26463 0.19592 0.30954 0.29293
2026-10-29/sim_233150_opic_content_only 0.35840 0.36012 0.26737 0.17879 0.29841 0.29200
2026-10-29/sim_233150_opic_content_domain_once 0.32515 0.38188 0.26402 0.16249 0.30417 0.29055
2026-10-29/sim_233150_inlink_content_damp 0.35416 0.34827 0.23856 0.18891 0.29422 0.28561
2026-10-27/sim_233150_host_prior 0.35326 0.34470 0.24528 0.16677 0.29569 0.28073
2026-10-29/sim_233150_inlink_full 0.33820 0.34643 0.24828 0.17889 0.28700 0.28034
2026-10-27/sim_233150_quality_density_fasttext_top12b 0.34922 0.38724 0.25565 0.19070 0.29342 0.29812
2026-10-29/sim_233150_domain_prior_full_fasttext_top12b 0.38005 0.35502 0.26730 0.19586 0.29689 0.29709
2026-10-27/common_crawl_baseline_processed_28_8b 0.35383 0.38069 0.26679 0.18341 0.27754 0.29322
2026-10-27/sim_233150_source_quality_cash_fasttext_top12b 0.34491 0.37732 0.27291 0.17031 0.28670 0.29102
2026-10-29/common_crawl_baseline_processed_c56_110 0.33462 0.36239 0.25603 0.20500 0.27693 0.28890
2026-10-27/sim_233150_source_quality_cash_dedup_no_fasttext 0.32496 0.37223 0.25932 0.19102 0.27988 0.28822
2026-10-29/sim_233150_inlink_content_fasttext_top12b 0.31763 0.34796 0.25714 0.19480 0.30892 0.28776
2026-11-02/sim_2m_edu_site_fasttext_top50pct 0.36540 0.33901 0.24008 0.18149 0.31359 0.28754
2026-11-03/sim_2m_c5_mix_source_quality_inlink_fasttext_top50pct 0.34652 0.33578 0.25521 0.19723 0.28569 0.28339
2026-10-29/sim_233150_inlink_content_domain_once_dedup_no_fasttext 0.32694 0.34777 0.25126 0.18659 0.28218 0.28023
2026-11-02/sim_2m_pass_target_fasttext_top50pct 0.34650 0.34655 0.24466 0.17078 0.29105 0.28009
2026-11-02/sim_2m_ftavg_v5_full 0.35642 0.33193 0.24604 0.18405 0.28630 0.27958
2026-11-03/sim_2m_c2_nc_vote_once_both_fasttext_top50pct 0.33655 0.33415 0.21710 0.20978 0.28583 0.27927
2026-10-29/sim_233150_inlink_content_domain_once_fasttext_top12b 0.32522 0.34991 0.24540 0.17442 0.29010 0.27881
2026-11-02/sim_2m_nc_yield_prior_fasttext_top50pct 0.34058 0.33362 0.26026 0.17309 0.29186 0.27859
2026-11-02/sim_2m_pass_target 0.34847 0.33284 0.20598 0.19507 0.29789 0.27842
2026-11-03/sim_2m_c4_nc_domain_once_quality_votes_fasttext_top50pct 0.34556 0.33141 0.18253 0.20412 0.29397 0.27560
2026-10-26/common_crawl_baseline_processed_14_4b (our m2 run; an HF-trained run gave 0.27500, being checked) 0.34272 0.36613 0.26205 0.19208 0.28957 0.29179
2026-11-02/sim_2m_dedup_aware_fasttext_top50pct 0.34169 0.32266 0.24009 0.18682 0.28170 0.27381
2026-11-03/sim_2m_c3_nc_domain_once_domain_prior_fasttext_top50pct 0.32986 0.32584 0.21307 0.20194 0.28470 0.27350
2026-11-02/sim_2m_nc_lang_url_fasttext_top50pct 0.33571 0.32615 0.24121 0.18067 0.28422 0.27328
2026-11-02/sim_2m_ftavg_v5_full_fasttext_top50pct 0.33352 0.32335 0.21942 0.19064 0.28527 0.27175
2026-11-02/sim_2m_head_tail_fasttext_top50pct 0.34451 0.31292 0.23058 0.19411 0.27875 0.27123
2026-11-03/sim_2m_nc_discovery_fasttext_top50pct 0.30835 0.33719 0.21305 0.18118 0.28720 0.26951
2026-11-02/sim_2m_site_bandit_fasttext_top50pct 0.32075 0.32632 0.21698 0.18704 0.28355 0.26927
2026-11-02/sim_2m_nc_base_fasttext_top50pct 0.34099 0.31851 0.20652 0.18666 0.28024 0.26764
2026-11-02/sim_2m_inlink_max_fasttext_top50pct 0.32208 0.32802 0.19559 0.18042 0.28956 0.26686
2026-11-02/sim_2m_learned_url_fasttext_top50pct 0.30812 0.29668 0.19537 0.17877 0.28468 0.25490
2026-11-02/sim_2m_host_explore_fasttext_top50pct 0.31600 0.31798 0.14376 0.17889 0.27042 0.25153

Prod-like setting: same as prod cycles 1-9 (9 cycles, 34.1M pages, 250 per domain, page score Β½ fastText + Β½ FineWeb-Edu); folders named sim_233150_*

folder priority (what the simulation changes) tokens status
2026-10-29/sim_233150_inlink_full inlink average of page scores (prod's priority; the control) 11.49B uploaded
2026-10-29/sim_233150_domain_prior_full 0.5 Γ— registered-domain mean of fetched pages + 0.5 Γ— inlink average 11.36B uploaded
2026-10-27/sim_233150_host_prior 0.5 Γ— host mean of fetched pages (10-page history) + 0.5 Γ— inlink average 11.29B uploaded
2026-10-29/sim_233150_inlink_content inlink average; links to non-content URLs (assets, forms, parked, bot-check) get no vote 12.70B uploaded
2026-10-29/sim_233150_inlink_content_damp as inlink_content, same-domain votes Γ— 0.2 12.70B uploaded
2026-10-29/sim_233150_inlink_content_template_once as inlink_content, each source URL template votes once per URL 12.47B uploaded
2026-10-29/sim_233150_inlink_content_domain_once as inlink_content, each source domain votes once per URL 12.41B uploaded
2026-10-29/sim_233150_opic_content_only OPIC cash; non-content URLs get no cash 8.43B uploaded
2026-10-29/sim_233150_opic_content_domain_once as opic_content_only, a source domain's pages count once per target domain per cycle 8.51B uploaded
2026-10-27/sim_233150_source_quality_cash OPIC cash weighted by the source page's quality score 7.23B uploaded
2026-10-27/sim_233150_quality_density OPIC cash split by the targets' predicted quality density (domain/host/section history) 6.97B uploaded
2026-10-29/sim_233150_opic_content_self_damp as opic_content_only, at most 0.2 of a page's cash to its own domain 8.35B uploaded
2026-10-29/sim_233150_opic_content_template_split as opic_content_only, cash split equally per URL template first 7.84B uploaded
2026-10-29/sim_233150_opic_content_trust_decay as opic_content_only, cash enters only at the seeds and Γ— 0.85 per hop (TrustRank-style) 8.46B uploaded

Simulations aligned with Zichun's run ftavg_2m_250dom_v5 (cycle 1 = 814,021 seeds, then 2,000,000 pages per cycle, 10 cycles, page score = DCLM fastText of English pages)

folder priority (what the simulation changes) tokens status
2026-11-02/sim_2m_ftavg_v5_full Zichun's v5 itself: fastText inlink average (production's priority; the control) 8.58B uploaded
2026-11-02/sim_2m_host_explore 20% of each cycle probes hosts not fetched yet (2 URLs each), the rest by host mean fastText (Host-Aware Web Crawling) 6.21B uploaded
2026-11-02/sim_2m_inlink_max best fastText among English inlinking pages instead of the average (Document Quality Scoring for Web Crawling) 8.65B uploaded
2026-11-02/sim_2m_head_tail half of each cycle from the 10,000 most-linked domains, half from the rest (WebGraphMix) 7.38B uploaded
2026-11-02/sim_2m_learned_url online model trained each cycle on fetched pages (URL tokens, host, inlink stats) predicts fastText 10.75B uploaded
2026-11-02/sim_2m_site_bandit Thompson sampling over hosts on the share of pages passing the fastText filter 5.80B uploaded
2026-11-02/sim_2m_pass_target share of English inlinking pages that pass the fastText filter (0.018112) instead of their mean score 8.01B uploaded
2026-11-02/sim_2m_dedup_aware inlink average Γ— (1 βˆ’ host's rate of duplicate pages among fetched pages) 8.69B uploaded
2026-11-02/sim_2m_edu_site 0.5 Γ— inlink average + 0.5 Γ— host mean FineWeb-Edu score of fetched pages 6.48B uploaded
2026-11-02/sim_2m_nc_base Zichun's v5 + no votes for non-content URLs (assets, forms, parked, bot-check); the new base - pending (simulating)
2026-11-02/sim_2m_nc_lang_url nc_base + no votes for URLs that name another language (/de/, fr.host, ?hl=xx) - pending (simulating)
2026-11-02/sim_2m_nc_yield_prior nc_lang_url, priority Γ— the URL template's / host's share of fetched pages that survive processing 11.72B uploaded
2026-11-03/sim_2m_nc_discovery nc_lang_url, priority Γ— (1 + host's rate of discovering new pages that pass fastText / global rate) 10.20B uploaded

Real crawls (not simulations): the hero Common Crawl-style OPIC crawl and the quality-weighted OPIC crawls, same processing

folder what it is tokens status
2026-10-27/common_crawl_baseline_processed_28_8b Common Crawl-style OPIC crawl (crawl 239994 and its earlier jobs), cycles 1-122, rules + dedup + fastText, about 28.8B tokens (1B-1x sees each document once) 26.60B uploaded
2026-10-26/common_crawl_baseline_processed_14_4b Common Crawl-style OPIC crawl, cycles 1-55, rules + dedup + fastText, 13.8B tokens (Core 0.29179 on the team sheet) 13.82B uploaded
2026-10-29/common_crawl_baseline_processed_c56_110 Common Crawl-style OPIC crawl, cycles 56-110 only, rules + dedup + fastText (later cycles; compare with cycles 1-55) 12.36B uploaded
2026-11-06/common_crawl_baseline_processed_c111_165 Common Crawl-style OPIC crawl, cycles 111-165 only, rules + dedup + fastText 11.81B uploaded
2026-11-11/opic_quality_real_cycles_1_49_fasttext_threshold Quality-weighted OPIC crawl (crawls 259887 / 261543), cycles 1-49 (as many pages as Common Crawl-style cycles 1-55), rules + dedup + fastText 21.63B uploaded
2026-11-15/opic_quality_250_real_cycles_1_46_fasttext_threshold Quality-weighted OPIC crawl, flat 250 pages per site (crawl 265639), cycles 1-46 (as many pages as Common Crawl-style cycles 1-55), rules + dedup + fastText; opted-out sites removed 23.58B uploaded
2026-11-18/opic_quality_real_cycles_1_49_fasttext_top_13_8b the cycles 1-49 threshold data above, opted-out sites removed, cut to the tokens of Common Crawl-style cycles 1-55 (13.82B) by the original fastText score (keeps score >= 0.04359) 13.84B uploaded
2026-11-18/opic_quality_250_real_cycles_1_46_fasttext_top_13_8b the flat-250 cycles 1-46 threshold data above, opted-out sites removed, cut to the tokens of Common Crawl-style cycles 1-55 (13.82B) by the original fastText score (keeps score >= 0.04879) 13.84B uploaded

Combinations and stability runs (added 11-03)

folder what the simulation does tokens status
2026-11-03/sim_2m_c1_opic_content_source_quality Simulation aligned with Zichun's 2M x 10: OPIC source-quality cash + no cash to non-content URLs; 250/domain, full processing 4.63B uploaded
2026-11-03/sim_2m_opic_plain_250 Simulation aligned with Zichun's 2M x 10: plain OPIC priority, 250/domain, full processing 3.33B uploaded
2026-11-03/sim_2m_opic_plain_harmonic Simulation aligned with Zichun's 2M x 10: plain OPIC priority, the hero's Common Crawl harmonic per-domain limits (8M file) cut at 2M per cycle, short cycles kept; full processing 2.97B uploaded
2026-11-03/sim_2m_c2_nc_vote_once_both Simulation aligned with Zichun's 2M x 10: inlink average, non-content rule, one vote per source domain and per source template; 250/domain, full processing - pending (simulating)
2026-11-03/sim_2m_c3_nc_domain_once_domain_prior Simulation aligned with Zichun's 2M x 10: 0.5 domain prior + 0.5 inlink average, non-content rule, domain once; 250/domain, full processing - pending (simulating)
2026-11-03/sim_2m_c4_nc_domain_once_quality_votes Simulation aligned with Zichun's 2M x 10: quality-weighted inlink average, non-content rule, domain once; 250/domain, full processing - pending (simulating)
2026-11-03/sim_2m_c5_mix_source_quality_inlink Simulation aligned with Zichun's 2M x 10: half by source-quality OPIC cash, half by inlink average with vote-once rules, non-content rule in both; 250/domain, full processing - simulated; data in the top-50% table below
2026-11-03/sim_2m_c1_opic_content_source_quality_30cycles Simulation aligned with Zichun's 2M setting, 30 cycles: OPIC source-quality cash + no cash to non-content URLs; 250/domain, full processing - pending (simulating)
2026-11-03/sim_2m_c2_nc_vote_once_both_30cycles Simulation aligned with Zichun's 2M setting, 30 cycles: inlink average, non-content rule, one vote per source domain and per source template; 250/domain, full processing - pending (simulating)

Post-dedup, no fastText filter (English β†’ rules β†’ BFF dedup only)

folder source run tokens status
2026-10-29/sim_233150_inlink_content_domain_once_dedup_no_fasttext sim_233150_inlink_content_domain_once 25.91B uploaded
2026-10-27/sim_233150_source_quality_cash_dedup_no_fasttext sim_233150_source_quality_cash 21.35B uploaded

Top 12B tokens by 原ζ₯ηš„ fastText θ΄¨ι‡εˆ† after dedup (equal size for every run; prod-like setting)

folder source run tokens status
2026-10-29/sim_233150_domain_prior_full_fasttext_top12b sim_233150_domain_prior_full 12.00B uploaded
2026-10-29/sim_233150_inlink_content_fasttext_top12b sim_233150_inlink_content 12.00B uploaded
2026-10-29/sim_233150_inlink_content_domain_once_fasttext_top12b sim_233150_inlink_content_domain_once 12.00B uploaded
2026-10-29/sim_233150_inlink_content_template_once_fasttext_top12b sim_233150_inlink_content_template_once 12.00B uploaded
2026-10-29/sim_233150_inlink_full_fasttext_top12b sim_233150_inlink_full 12.00B uploaded
2026-10-29/sim_233150_opic_content_only_fasttext_top12b sim_233150_opic_content_only 12.00B uploaded
2026-10-27/sim_233150_quality_density_fasttext_top12b sim_233150_quality_density 12.00B uploaded
2026-10-27/sim_233150_source_quality_cash_fasttext_top12b sim_233150_source_quality_cash 12.00B uploaded
2026-10-29/sim_233150_opic_content_trust_decay_fasttext_top12b sim_233150_opic_content_trust_decay 12.00B uploaded

Top 50% of post-dedup tokens by 原ζ₯ηš„ fastText θ΄¨ι‡εˆ† (2M setting; replaces the fixed 0.018112 threshold)

folder source run tokens status
2026-11-03/sim_2m_c3_nc_domain_once_domain_prior_fasttext_top50pct sim_2m_c3_nc_domain_once_domain_prior 7.68B uploaded
2026-11-02/sim_2m_dedup_aware_fasttext_top50pct sim_2m_dedup_aware 7.32B uploaded
2026-11-02/sim_2m_edu_site_fasttext_top50pct sim_2m_edu_site 6.51B uploaded
2026-11-02/sim_2m_ftavg_v5_full_fasttext_top50pct sim_2m_ftavg_v5_full 7.21B uploaded
2026-11-02/sim_2m_head_tail_fasttext_top50pct sim_2m_head_tail 6.59B uploaded
2026-11-02/sim_2m_host_explore_fasttext_top50pct sim_2m_host_explore 5.65B uploaded
2026-11-02/sim_2m_nc_yield_prior_fasttext_top50pct sim_2m_nc_yield_prior 9.87B uploaded
2026-11-02/sim_2m_pass_target_fasttext_top50pct sim_2m_pass_target 7.02B uploaded
2026-11-03/sim_2m_c1_opic_content_source_quality_fasttext_top50pct sim_2m_c1_opic_content_source_quality 7.22B uploaded
2026-11-03/sim_2m_c2_nc_vote_once_both_fasttext_top50pct sim_2m_c2_nc_vote_once_both 7.87B uploaded
2026-11-03/sim_2m_c4_nc_domain_once_quality_votes_fasttext_top50pct sim_2m_c4_nc_domain_once_quality_votes 7.87B uploaded
2026-11-03/sim_2m_c5_mix_source_quality_inlink_fasttext_top50pct sim_2m_c5_mix_source_quality_inlink 7.22B uploaded
2026-11-02/sim_2m_inlink_max_fasttext_top50pct sim_2m_inlink_max 7.37B uploaded
2026-11-02/sim_2m_learned_url_fasttext_top50pct sim_2m_learned_url 8.35B uploaded
2026-11-02/sim_2m_nc_base_fasttext_top50pct sim_2m_nc_base 8.12B uploaded
2026-11-03/sim_2m_nc_discovery_fasttext_top50pct sim_2m_nc_discovery 8.69B uploaded
2026-11-02/sim_2m_nc_lang_url_fasttext_top50pct sim_2m_nc_lang_url 8.62B uploaded
2026-11-02/sim_2m_site_bandit_fasttext_top50pct sim_2m_site_bandit 5.93B uploaded
2026-11-04/sim_2m_opic_plain_250_seed_quality Zichun's setting: plain OPIC, each seed's starting score by its quality (0.2 + 0.8Β·fastText), otherwise as OPIC + 250 - pending (simulating)

Page-score grid and reruns (added 11-05)

Zichun's 2M x 10 setting, 250 per domain, processing English -> rules -> BFF -> top 50% of the post-dedup tokens by fastText. Page score: prod's fastText and/or FineWeb-Edu (graph-text CPU score: the kenhktsui FineWeb-Edu fastText model run correctly on the simulation text), see each row.

folder what the simulation does tokens status
2026-11-05/sim_2m_c1_opic_content_source_quality_30cycles_fasttext_top50pct Simulation aligned with Zichun's 2M setting, 30 cycles: OPIC source-quality cash + no cash to non-content URLs; 250/domain, full processing 19.71B uploaded
2026-11-05/sim_2m_c1_opic_content_source_quality_edu_fasttext_top50pct Simulation aligned with Zichun's 2M x 10: OPIC source-quality cash + non-content rule, page score FineWeb-Edu only (graph-text CPU score); 250/domain, top 50% processing 7.38B uploaded
2026-11-05/sim_2m_c1_opic_content_source_quality_ft50edu50_fasttext_top50pct Simulation aligned with Zichun's 2M x 10: OPIC source-quality cash + non-content rule, page score 1/2 prod fastText + 1/2 FineWeb-Edu (graph-text CPU score); 250/domain, top 50% processing 7.31B uploaded
2026-11-05/sim_2m_c1_opic_content_source_quality_ft50edu50_gpuedu_fasttext_top50pct Simulation aligned with Zichun's 2M x 10: OPIC source-quality cash + non-content rule, page score 1/2 prod fastText + 1/2 FineWeb-Edu (official GPU classifier, comparison); 250/domain, top 50% processing 7.28B uploaded
2026-11-05/sim_2m_c1_opic_seed_quality_cash_fasttext_top50pct Simulation aligned with Zichun's 2M x 10: OPIC source-quality cash + non-content rule, seeds' starting cash weighted by their fastText score too; 250/domain, full processing 7.22B uploaded
2026-11-05/sim_2m_c2_nc_vote_once_both_30cycles_fasttext_top50pct Simulation aligned with Zichun's 2M setting, 30 cycles: inlink average, non-content rule, one vote per source domain and per source template; 250/domain, full processing 20.72B uploaded
2026-11-05/sim_2m_c2_nc_vote_once_both_edu_fasttext_top50pct Simulation aligned with Zichun's 2M x 10: inlink average, non-content rule, one vote per linking site, page score FineWeb-Edu only (graph-text CPU score); 250/domain, top 50% processing 7.08B uploaded
2026-11-05/sim_2m_c2_nc_vote_once_both_ft50edu50_fasttext_top50pct Simulation aligned with Zichun's 2M x 10: inlink average, non-content rule, one vote per linking site, page score 1/2 prod fastText + 1/2 FineWeb-Edu (graph-text CPU score); 250/domain, top 50% processing 7.75B uploaded
2026-11-05/sim_2m_opic_plain_250_seed_quality_fasttext_top50pct Simulation aligned with Zichun's 2M x 10: plain OPIC priority with the seeds' starting cash weighted by their fastText score, 250/domain 5.51B uploaded
2026-11-05/sim_2m_v5_edu_fasttext_top50pct Simulation aligned with Zichun's 2M x 10: v5 inlink average, page score FineWeb-Edu only (graph-text CPU score); 250/domain, top 50% processing 6.32B uploaded
2026-11-05/sim_2m_v5_ft50edu50_fasttext_top50pct Simulation aligned with Zichun's 2M x 10: v5 inlink average, page score 1/2 prod fastText + 1/2 FineWeb-Edu (graph-text CPU score); 250/domain, top 50% processing 7.00B uploaded
2026-11-05/sim_2m_v5_ft50edu50_gpuedu_fasttext_top50pct Simulation aligned with Zichun's 2M x 10: v5 inlink average, page score 1/2 prod fastText + 1/2 FineWeb-Edu (official GPU classifier, comparison); 250/domain, top 50% processing 6.39B uploaded
2026-11-04/sim_2m_opic_content_trust_decay_fasttext_top50pct Simulation aligned with Zichun's 2M x 10: OPIC, cash only from the seeds x0.85 per link + no cash to non-content URLs, page score fastText only; 250/domain, top 50% processing 7.23B uploaded
2026-11-03/sim_2m_opic_plain_250_fasttext_top50pct Simulation aligned with Zichun's 2M x 10: plain OPIC priority, 250/domain, full processing 5.51B uploaded
2026-11-03/sim_2m_opic_plain_harmonic_fasttext_top50pct Simulation aligned with Zichun's 2M x 10: plain OPIC priority, the hero's Common Crawl harmonic per-domain limits (8M file) cut at 2M per cycle, short cycles kept; top 50% processing 4.57B uploaded
2026-11-05/sim_2m_c1_siteprior_ft_fasttext_top50pct Simulation aligned with Zichun's 2M x 10: C1 (OPIC source-quality cash + non-content rule) + site quality prior, page score prod fastText; 250/domain, top 50% processing 7.07B uploaded
2026-11-05/sim_2m_c1_siteprior_ft50edu50_fasttext_top50pct Simulation aligned with Zichun's 2M x 10: C1 (OPIC source-quality cash + non-content rule) + site quality prior, page score 1/2 prod fastText + 1/2 FineWeb-Edu (graph-text CPU score); 250/domain, top 50% processing 7.39B uploaded
2026-11-06/sim_2m_quality_density_fasttext_top50pct Simulation aligned with Zichun's 2M x 10: quality-density OPIC (cash split by the target site/section's pass rate so far, fastText >= 0.018112), no non-content rule; 250/domain, top 50% processing 5.51B uploaded
2026-11-06/sim_2m_opic_source_quality_fasttext_top50pct Simulation aligned with Zichun's 2M x 10: quality-weighted OPIC (a page passes cash x (0.2 + 0.8 its own fastText score)), no non-content rule; 250/domain, top 50% processing 5.61B uploaded
2026-11-06/sim_2m_opic_source_quality_x_density_fasttext_top50pct Simulation aligned with Zichun's 2M x 10: quality-weighted OPIC x quality density (outflow by the page's own fastText score, split by the target site/section's pass rate so far), no non-content rule; 250/domain, top 50% processing 5.61B uploaded
2026-11-06/sim_2m_opic_source_quality_lang_url_fasttext_top50pct Simulation aligned with Zichun's 2M x 10: quality-weighted OPIC + no cash to URLs naming another language, no non-content rule; 250/domain, top 50% processing 5.77B uploaded
2026-11-06/sim_2m_opic_source_quality_domain_once_fasttext_top50pct Simulation aligned with Zichun's 2M x 10: quality-weighted OPIC + each linking site counts once per target site, no non-content rule; 250/domain, top 50% processing 5.67B uploaded
2026-11-06/sim_2m_head_tail_10pct_fasttext_top50pct Simulation aligned with Zichun's 2M x 10: 10% of each cycle from the 10,000 most-linked domains, 90% from the rest (inlink average); 250/domain, top 50% processing 7.22B uploaded
2026-11-06/sim_2m_opic_plain_hybrid_caps_fasttext_top50pct Simulation aligned with Zichun's 2M x 10: plain OPIC priority, hybrid per-domain caps (harmonic x quality, mean 250); top 50% processing 5.52B uploaded
2026-11-06/sim_2m_opic_source_quality_hybrid_caps_fasttext_top50pct Simulation aligned with Zichun's 2M x 10: quality-weighted OPIC, no non-content rule, hybrid per-domain caps (harmonic x quality, mean 250); top 50% processing 5.62B uploaded
2026-11-06/sim_2m_opic_content_trust_decay_hybrid_caps_fasttext_top50pct Simulation aligned with Zichun's 2M x 10: OPIC, cash only from the seeds x0.85 per link + no cash to non-content URLs, hybrid per-domain caps (harmonic x quality, mean 250); top 50% processing 7.26B uploaded
2026-11-06/sim_2m_quality_density_hybrid_caps_fasttext_top50pct Simulation aligned with Zichun's 2M x 10: quality-density OPIC (fastText >= 0.018112), no non-content rule, hybrid per-domain caps (harmonic x quality, mean 250); top 50% processing 5.53B uploaded
2026-11-08/sim_2m_c1_lang_url_ft50edu50_harmonic_fasttext_top50pct Simulation aligned with Zichun's 2M x 10, twin of the real crawl: OPIC source-quality cash (1/2 fastText + 1/2 FineWeb-Edu) + no cash to non-content or other-language URLs, the hero's Common Crawl per-domain limits (8M file) cut at 2M per cycle in random order, short cycles kept; top 50% processing 5.33B uploaded
2026-11-08/sim_2m_c1_lang_url_ft50edu50_harmonic_score_cut_fasttext_top50pct Simulation aligned with Zichun's 2M x 10: the real-crawl twin (source-quality cash 1/2 fastText + 1/2 FineWeb-Edu, non-content and language rules, Common Crawl per-domain limits) with the cut to 2M per cycle in global cash order instead of random; top 50% processing 6.12B uploaded
2026-11-08/sim_2m_c1_opic_content_source_quality_harmonic_fasttext_top50pct Simulation aligned with Zichun's 2M x 10: OPIC source-quality cash + no cash to non-content URLs, the hero's Common Crawl harmonic per-domain limits (8M file) cut at 2M per cycle, short cycles kept; top 50% processing 5.25B uploaded
2026-11-08/sim_2m_c1_lang_url_trust_decay_ft50edu50_fasttext_top50pct Simulation aligned with Zichun's 2M x 10: TrustRank x quality-weighted OPIC (cash only from the seeds, x0.85 per link, each page passes cash x (0.2 + 0.8 q)), non-content and language rules, page score 1/2 prod fastText + 1/2 FineWeb-Edu; 250/domain, top 50% processing 7.57B uploaded
2026-11-08/sim_2m_c1_lang_url_siteprior_ft50edu50_fasttext_top50pct Simulation aligned with Zichun's 2M x 10: quality-weighted OPIC cash + non-content and language rules + site quality prior (priority 1/2 F(cash) + 1/2 F(site mean page score)), page score 1/2 prod fastText + 1/2 FineWeb-Edu; 250/domain, top 50% processing 7.62B uploaded
2026-11-08/sim_2m_c1_lang_url_ft50edu50_harmonic_score_cut_max250_fasttext_top50pct Simulation aligned with Zichun's 2M x 10: the real-crawl twin with the score cut (source-quality cash 1/2 fastText + 1/2 FineWeb-Edu, non-content and language rules, Common Crawl per-domain limits cut in cash order), each domain's limit min(Common Crawl limit, 250); top 50% processing 6.12B uploaded
2026-11-08/sim_2m_c1_lang_url_ft50edu50_hybrid_caps_fasttext_top50pct Simulation aligned with Zichun's 2M x 10: the real-crawl twin's priority (source-quality cash 1/2 fastText + 1/2 FineWeb-Edu, non-content and language rules) with Sai's hybrid per-domain caps instead of Common Crawl's limits, cut in cash order; top 50% processing 7.52B uploaded
2026-11-09/sim_2m_ccgraph_opic_plain_harmonic_fasttext_top50pct Simulation on the graph of our Common Crawl-style crawl (simgraph-239994), 898,157 seeds + 2M x 9: plain OPIC priority, Common Crawl harmonic per-domain limits (8M file) cut at 2M per cycle, short cycles kept; top 50% processing 3.28B uploaded
2026-11-09/sim_2m_ccgraph_opic_plain_250_fasttext_top50pct Simulation on the graph of our Common Crawl-style crawl (simgraph-239994), 898,157 seeds + 2M x 9: plain OPIC priority, 250/domain; top 50% processing 3.92B uploaded
2026-11-09/sim_2m_ccgraph_c1_opic_content_source_quality_fasttext_top50pct Simulation on the graph of our Common Crawl-style crawl (simgraph-239994), 898,157 seeds + 2M x 9: OPIC source-quality cash (page score fastText) + no cash to non-content URLs; 250/domain, top 50% processing 3.99B uploaded
2026-11-09/sim_2m_ccgraph_opic_content_trust_decay_fasttext_top50pct Simulation on the graph of our Common Crawl-style crawl (simgraph-239994), 898,157 seeds + 2M x 9: OPIC, cash only from the seeds x0.85 per link + no cash to non-content URLs; 250/domain, top 50% processing 3.96B uploaded
2026-11-12/sim_5m_ccgraph_opic_plain_harmonic_fasttext_top50pct Simulation on the graph of our Common Crawl-style crawl (simgraph-239994), 898,157 seeds + 5M x 9: plain OPIC priority, Common Crawl harmonic per-domain limits (8M file) cut at 5M per cycle, short cycles kept; top 50% processing 6.98B uploaded
2026-11-12/sim_5m_ccgraph_opic_plain_250_fasttext_top50pct Simulation on the graph of our Common Crawl-style crawl (simgraph-239994), 898,157 seeds + 5M x 9: plain OPIC priority, 250/domain; top 50% processing 9.35B uploaded
2026-11-12/sim_5m_ccgraph_c1_opic_content_source_quality_fasttext_top50pct Simulation on the graph of our Common Crawl-style crawl (simgraph-239994), 898,157 seeds + 5M x 9: OPIC source-quality cash (page score fastText) + no cash to non-content URLs; 250/domain, top 50% processing 9.47B uploaded
2026-11-12/sim_5m_ccgraph_opic_content_trust_decay_fasttext_top50pct Simulation on the graph of our Common Crawl-style crawl (simgraph-239994), 898,157 seeds + 5M x 9: OPIC, cash only from the seeds x0.85 per link + no cash to non-content URLs; 250/domain, top 50% processing 9.49B uploaded
2026-11-13/sim_5m_ccgraph_c1_opic_content_source_quality_ftanchor_cut_fasttext_top50pct Simulation on the graph of our Common Crawl-style crawl (simgraph-239994), 898,157 seeds + 5M x 9: OPIC source-quality cash + no cash to non-content URLs, page score = pass / fail at the processing cut (0 below fastText 0.00108, 1 from 0.003); 250/domain, top 50% processing - pending (simulating or processing)
2026-11-13/sim_5m_ccgraph_c1_opic_content_source_quality_ftanchor_dclm_fasttext_top50pct Simulation on the graph of our Common Crawl-style crawl (simgraph-239994), 898,157 seeds + 5M x 9: OPIC source-quality cash + no cash to non-content URLs, page score = fastText mapped through anchors from the measured DCLM-cut pass share of each score band's outlinks (0 below 3e-4, concave up to 1); 250/domain, top 50% processing - pending (simulating or processing)
2026-11-13/sim_5m_ccgraph_c1_opic_content_source_quality_ftanchor_top50_fasttext_top50pct Simulation on the graph of our Common Crawl-style crawl (simgraph-239994), 898,157 seeds + 5M x 9: OPIC source-quality cash + no cash to non-content URLs, page score = fastText mapped through anchors from the measured top-50% usefulness of each score band's outlinks (0 at the floor, 1 from 0.03); 250/domain, top 50% processing 10.06B uploaded
2026-11-14/sim_5m_ccgraph_c1_opic_content_source_quality_edu_anchor_fasttext_top50pct Simulation on the graph of our Common Crawl-style crawl (simgraph-239994), 898,157 seeds + 5M x 9: OPIC source-quality cash + no cash to non-content URLs, page score = CPU FineWeb-Edu mapped through anchors from its measured top-50% usefulness of outlinks (0 up to 0.75, 1 from 2.875); 250/domain, top 50% processing 10.13B uploaded
2026-11-14/sim_5m_ccgraph_c1_opic_content_source_quality_ft50edu50_anchor_fasttext_top50pct Simulation on the graph of our Common Crawl-style crawl (simgraph-239994), 898,157 seeds + 5M x 9: OPIC source-quality cash + no cash to non-content URLs, page score = 1/2 anchored fastText + 1/2 anchored FineWeb-Edu (both maps from measured outlink usefulness); 250/domain, top 50% processing 10.13B uploaded
2026-11-15/sim_5m_ccgraph_opic_plain_250_quality_feedback_fasttext_top50pct Simulation on the graph of our Common Crawl-style crawl (simgraph-239994), 898,157 seeds + 5M x 9: plain OPIC priority, 250/domain with quality feedback every 2 cycles (cap x clip(domain mean fastText score / overall mean, 0.3, 4)); top 50% processing 9.39B uploaded

One fastText threshold for the 5M second-graph simulations (added 11-16)

Same simulations and processing as their top 50% folders (English -> rules -> BFF dedup), but every run keeps the post-dedup docs with 原ζ₯ηš„ fastText θ΄¨ι‡εˆ† >= 0.0017535686 (exact 0.0017535686492919922), the score where the top 50% of the plain OPIC 250/domain run (2026-11-12/sim_5m_ccgraph_opic_plain_250) ends, so all runs meet one quality bar. For that baseline run the data is the same docs as its top 50% folder. From 11-16 (~22:00) this is the processing of every new 5M second-graph simulation: those runs have only this folder, no top 50% one.

folder what the simulation does tokens status
2026-11-12/sim_5m_ccgraph_opic_plain_250_fasttext_threshold_0.00175357 Simulation on the graph of our Common Crawl-style crawl (simgraph-239994), 898,157 seeds + 5M x 9: plain OPIC priority, 250/domain; fastText >= 0.00175357 processing 9.35B uploaded
2026-11-12/sim_5m_ccgraph_c1_opic_content_source_quality_fasttext_threshold_0.00175357 Simulation on the graph of our Common Crawl-style crawl (simgraph-239994), 898,157 seeds + 5M x 9: OPIC source-quality cash (page score fastText) + no cash to non-content URLs; 250/domain, fastText >= 0.00175357 processing 9.51B uploaded
2026-11-13/sim_5m_ccgraph_c1_opic_content_source_quality_ftanchor_top50_fasttext_threshold_0.00175357 Simulation on the graph of our Common Crawl-style crawl (simgraph-239994), 898,157 seeds + 5M x 9: OPIC source-quality cash + no cash to non-content URLs, page score = fastText mapped through anchors from the measured top-50% usefulness of each score band's outlinks (0 at the floor, 1 from 0.03); 250/domain, fastText >= 0.00175357 processing 10.67B uploaded
2026-11-12/sim_5m_ccgraph_opic_content_trust_decay_fasttext_threshold_0.00175357 Simulation on the graph of our Common Crawl-style crawl (simgraph-239994), 898,157 seeds + 5M x 9: OPIC, cash only from the seeds x0.85 per link + no cash to non-content URLs; 250/domain, fastText >= 0.00175357 processing 9.49B uploaded
2026-11-14/sim_5m_ccgraph_c1_opic_content_source_quality_ft50edu50_anchor_fasttext_threshold_0.00175357 Simulation on the graph of our Common Crawl-style crawl (simgraph-239994), 898,157 seeds + 5M x 9: OPIC source-quality cash + no cash to non-content URLs, page score = 1/2 anchored fastText + 1/2 anchored FineWeb-Edu (both maps from measured outlink usefulness); 250/domain, fastText >= 0.00175357 processing 10.66B uploaded
2026-11-15/sim_5m_ccgraph_c1_opic_content_source_quality_site_type_caps_fasttext_threshold_0.00175357 Simulation on the graph of our Common Crawl-style crawl (simgraph-239994), 898,157 seeds + 5M x 9: OPIC source-quality cash + no cash to non-content URLs; per-domain cap by site type: 125 shopping / tech / blog / country-code sites, 500 mid-size sites (251-5,000 pages), 250 otherwise; fastText >= 0.00175357 processing 9.51B uploaded
2026-11-15/sim_5m_ccgraph_c1_opic_content_source_quality_sample_deepen_fasttext_threshold_0.00175357 Simulation on the graph of our Common Crawl-style crawl (simgraph-239994), 898,157 seeds + 5M x 9: OPIC source-quality cash + no cash to non-content URLs; a new domain gets 5 pages first, then every cycle cap = 250 x clip(domain mean fastText score / overall mean, 0.3, 4); fastText >= 0.00175357 processing 9.60B uploaded
2026-11-15/sim_5m_ccgraph_c1_opic_content_source_quality_central_peripheral_fasttext_threshold_0.00175357 Simulation on the graph of our Common Crawl-style crawl (simgraph-239994), 898,157 seeds + 5M x 9: OPIC source-quality cash + no cash to non-content URLs; from cycle 2 half of each cycle from central domains (CC harmonic-centrality rank <= 4,991,256, half the graph's pages), half from the rest; 250/domain, fastText >= 0.00175357 processing 8.76B uploaded
2026-11-16/sim_5m_ccgraph_c1_opic_content_source_quality_dataset_domain_prior_fasttext_threshold_0.00175357 Simulation on the graph of our Common Crawl-style crawl (simgraph-239994), 898,157 seeds + 5M x 9: OPIC source-quality cash + no cash to non-content URLs, 250/domain; per-domain cap = 250 x a bounded multiplier from how often the domain's URLs appear in public curated datasets (FineWeb-Edu, FineMath, MegaMath) relative to its size in the graph; mean 250; fastText >= 0.00175357 processing - pending (simulating or processing)
2026-11-16/sim_5m_ccgraph_c1_opic_content_source_quality_fact_reserve_fasttext_threshold_0.00175357 Simulation on the graph of our Common Crawl-style crawl (simgraph-239994), 898,157 seeds + 5M x 9: OPIC source-quality cash + no cash to non-content URLs, 250/domain; 15% of each cycle first from fact-rich sites (reference / news / government by URL list; or sites whose fetched pages are mostly name-and-number dense); fastText >= 0.00175357 processing - pending (simulating or processing)
2026-11-16/sim_5m_ccgraph_c1_opic_content_source_quality_topic_min_share_fasttext_threshold_0.00175357 Simulation on the graph of our Common Crawl-style crawl (simgraph-239994), 898,157 seeds + 5M x 9: OPIC source-quality cash + no cash to non-content URLs, 250/domain; at least 15% of each cycle from math / science / code sites (math-dataset URL list, academic URLs, or fetched pages mostly math / science / code by keywords); fastText >= 0.00175357 processing - pending (simulating or processing)
2026-11-16/sim_5m_ccgraph_c1_opic_content_source_quality_repetition_cut_fasttext_threshold_0.00175357 Simulation on the graph of our Common Crawl-style crawl (simgraph-239994), 898,157 seeds + 5M x 9: OPIC source-quality cash + no cash to non-content URLs, 250/domain; a site whose fetched pages mostly repeat each other (same text or same line template) gets 25 per cycle instead of 250 from the next cycle; fastText >= 0.00175357 processing - pending (simulating or processing)
2026-11-16/sim_5m_ccgraph_c1_opic_content_source_quality_link_similarity_fasttext_threshold_0.00175357 Simulation on the graph of our Common Crawl-style crawl (simgraph-239994), 898,157 seeds + 5M x 9: OPIC source-quality cash + no cash to non-content URLs, 250/domain; priority = OPIC cash x (0.2 + 0.8 x how similar the best linking page is to health / mid-size-site pages, small CPU embedding model); fastText >= 0.00175357 processing - pending (simulating or processing)
2026-11-16/sim_5m_ccgraph_edu_site_fasttext_threshold_0.00175357 Simulation on the graph of our Common Crawl-style crawl (simgraph-239994), 898,157 seeds + 5M x 9: 1/2 rank of the fastText inlink average + 1/2 rank of the host's FineWeb-Edu history; 250/domain, fastText >= 0.00175357 processing - pending (simulating or processing)
2026-11-16/sim_5m_ccgraph_inlink_v5_fasttext_threshold_0.00175357 Simulation on the graph of our Common Crawl-style crawl (simgraph-239994), 898,157 seeds + 5M x 9: plain inlink average of 原ζ₯ηš„ fastText θ΄¨ι‡εˆ† (prod's method, Zichun's v5 config); 250/domain, fastText >= 0.00175357 processing 18.37B uploaded
2026-11-16/sim_5m_ccgraph_c2_nc_vote_once_both_edu_fasttext_threshold_0.00175357 Simulation on the graph of our Common Crawl-style crawl (simgraph-239994), 898,157 seeds + 5M x 9: inlink average of the FineWeb-Edu page score, non-content rule, one vote per linking site and URL pattern; 250/domain, fastText >= 0.00175357 processing 15.79B uploaded
2026-11-16/sim_5m_ccgraph_c5_mix_source_quality_inlink_fasttext_threshold_0.00175357 Simulation on the graph of our Common Crawl-style crawl (simgraph-239994), 898,157 seeds + 5M x 9: half of each cycle by quality-weighted OPIC (+ non-content rule), half by inlink average with one vote per linking site / URL pattern; 250/domain over both, fastText >= 0.00175357 processing - pending (simulating or processing)
2026-11-16/sim_5m_ccgraph_opic_source_quality_hybrid_caps_fasttext_threshold_0.00175357 Simulation on the graph of our Common Crawl-style crawl (simgraph-239994), 898,157 seeds + 5M x 9: quality-weighted OPIC with hybrid per-site caps (harmonic-centrality rank x site fastText quality, mean 250) rebuilt for this graph; fastText >= 0.00175357 processing 9.51B uploaded
2026-11-16/sim_5m_ccgraph_c1_opic_content_source_quality_topic_mix_fasttext_threshold_0.00175357 Simulation on the graph of our Common Crawl-style crawl (simgraph-239994), 898,157 seeds + 5M x 9: OPIC source-quality cash + no cash to non-content URLs, 250/domain; each cycle split over WebOrganizer topics by the paper's MMLU + HellaSwag mixture (site topic = majority of its fetched pages); fastText >= 0.00175357 processing - pending (simulating or processing)
2026-11-16/sim_5m_ccgraph_c1_opic_content_source_quality_site_value_fasttext_threshold_0.00175357 Simulation on the graph of our Common Crawl-style crawl (simgraph-239994), 898,157 seeds + 5M x 9: OPIC source-quality cash + no cash to non-content URLs; every 2 cycles a site's cap = 250 x (0.3-1.7 by rank) from how 8 small LMs' losses on 8 of its fetched pages track their benchmark ability (perplexity correlations; sites with >= 20 fetched text pages, others 250); fastText >= 0.00175357 processing - pending (simulating or processing)

Data ablations and mixes on the 5M second-graph data (added 11-16)

Each folder has the token count of the 5M quality-weighted OPIC run's top-50% data (9.47B, 2026-11-12/sim_5m_ccgraph_c1_opic_content_source_quality_fasttext_top50pct) and the same processing (post-dedup docs of 5M second-graph runs, unchanged) and tokenizer. Health = URL topic rule.

folder what it is tokens status
2026-11-16/ablation_qw_opic_5m_topic_swap_health base data (5M quality-weighted OPIC, top 50%) with every health-URL doc replaced by non-health docs of other 5M runs from the same fastText x FineWeb-Edu score bins (same tokens) 9.48B uploaded
2026-11-16/ablation_qw_opic_5m_topic_upweight_health base data with non-health docs replaced by health-URL docs of other 5M runs from the same score bins, as many as those runs have (health 4.2% -> 5.7% of tokens) 9.48B uploaded
2026-11-16/ablation_qw_opic_5m_random_swap_control control: 0.40B tokens of random base docs replaced by random docs of other 5M runs from the same score bins (the health swap's amount) 9.49B uploaded
2026-11-16/ablation_qw_opic_5m_score_swap_within_topic base data, URL-topic shares kept: in each topic the highest-fastText docs (1.65B tokens) replaced by docs of that topic from the middle third of its fastText range (same FineWeb-Edu bins) 9.49B uploaded
2026-11-16/mix_5m_ftmapped_trustrank_half half the tokens from the ft-mapped quality-weighted OPIC run's top-50% data, half from the TrustRank-style OPIC run's (random docs, each doc once; BFF rerun over the mix) 9.48B uploaded
2026-11-16/mix_5m_union_all_runs_top_fasttext reference, not a crawl policy: union of the 8 finished 5M runs' processed docs (one copy per page; BFF rerun over them, fastText rescored), top fastText down to the base data's tokens 9.49B uploaded
Downloads last month
687