case_id stringlengths 9 19 | graph_family stringclasses 2
values | n int32 2 17 | r int32 1 16 | status stringclasses 3
values | max_distance_from_canonical_start int64 0 127 ⌀ | observed_states int64 0 1,308B | evidence_prefix stringclasses 6
values | source_repo stringclasses 2
values | source_revision stringclasses 2
values | evidence_kind stringclasses 4
values | retained_unique_states int64 0 2k | layers listlengths 0 128 | states listlengths 0 2k |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
lrx:n2:r1 | lrx | 2 | 1 | COMPLETE | 1 | 2 | runs/b300-sweep-eight-20261011 | TryDotAtwo/LRX-BFS-8xB300-2026-10-11 | b23cc8c1629a395d057ef9cb009da3bfdd84889b | source_reported_two_pass_gpu | 2 | [
{
"depth": 0,
"new_unique_states": 1,
"cumulative_states": 1
},
{
"depth": 1,
"new_unique_states": 1,
"cumulative_states": 2
}
] | [
{
"state": [
0,
1
],
"seeds": [
0,
13
],
"retention_roles": [
"previous_small"
],
"evidence": [
{
"seed": 0,
"retention_role": "previous_small",
"source_file": "runs/b300-sweep-eight-20261011/cohorts/chunk-00000.tar.gz",
... |
lrx:n2:r2 | lrx | 2 | 2 | COMPLETE | 0 | 1 | runs/b300-sweep-eight-20261011 | TryDotAtwo/LRX-BFS-8xB300-2026-10-11 | b23cc8c1629a395d057ef9cb009da3bfdd84889b | source_reported_two_pass_gpu | 1 | [
{
"depth": 0,
"new_unique_states": 1,
"cumulative_states": 1
}
] | [
{
"state": [
0,
0
],
"seeds": [
0,
13
],
"retention_roles": [
"current"
],
"evidence": [
{
"seed": 0,
"retention_role": "current",
"source_file": "runs/b300-sweep-eight-20261011/cohorts/chunk-00000.tar.gz",
"archive_memb... |
lrx:n3:r1 | lrx | 3 | 1 | COMPLETE | 2 | 6 | runs/b300-sweep-eight-20261011 | TryDotAtwo/LRX-BFS-8xB300-2026-10-11 | b23cc8c1629a395d057ef9cb009da3bfdd84889b | source_reported_two_pass_gpu | 6 | [
{
"depth": 0,
"new_unique_states": 1,
"cumulative_states": 1
},
{
"depth": 1,
"new_unique_states": 3,
"cumulative_states": 4
},
{
"depth": 2,
"new_unique_states": 2,
"cumulative_states": 6
}
] | [
{
"state": [
0,
1,
2
],
"seeds": [
13
],
"retention_roles": [
"current"
],
"evidence": [
{
"seed": 13,
"retention_role": "current",
"source_file": "runs/b300-sweep-eight-20261011/cohorts/chunk-00000.tar.gz",
"archive_mem... |
lrx:n3:r2 | lrx | 3 | 2 | COMPLETE | 1 | 3 | runs/b300-sweep-eight-20261011 | TryDotAtwo/LRX-BFS-8xB300-2026-10-11 | b23cc8c1629a395d057ef9cb009da3bfdd84889b | source_reported_two_pass_gpu | 3 | [
{
"depth": 0,
"new_unique_states": 1,
"cumulative_states": 1
},
{
"depth": 1,
"new_unique_states": 2,
"cumulative_states": 3
}
] | [
{
"state": [
0,
1,
1
],
"seeds": [
0,
13
],
"retention_roles": [
"previous_small"
],
"evidence": [
{
"seed": 0,
"retention_role": "previous_small",
"source_file": "runs/b300-sweep-eight-20261011/cohorts/chunk-00000.tar.gz"... |
lrx:n3:r3 | lrx | 3 | 3 | COMPLETE | 0 | 1 | runs/b300-sweep-eight-20261011 | TryDotAtwo/LRX-BFS-8xB300-2026-10-11 | b23cc8c1629a395d057ef9cb009da3bfdd84889b | source_reported_two_pass_gpu | 1 | [
{
"depth": 0,
"new_unique_states": 1,
"cumulative_states": 1
}
] | [
{
"state": [
0,
0,
0
],
"seeds": [
0,
13
],
"retention_roles": [
"current"
],
"evidence": [
{
"seed": 0,
"retention_role": "current",
"source_file": "runs/b300-sweep-eight-20261011/cohorts/chunk-00000.tar.gz",
"arc... |
lrx:n4:r1 | lrx | 4 | 1 | COMPLETE | 6 | 24 | runs/b300-sweep-eight-20261011 | TryDotAtwo/LRX-BFS-8xB300-2026-10-11 | b23cc8c1629a395d057ef9cb009da3bfdd84889b | source_reported_two_pass_gpu | 6 | [
{
"depth": 0,
"new_unique_states": 1,
"cumulative_states": 1
},
{
"depth": 1,
"new_unique_states": 3,
"cumulative_states": 4
},
{
"depth": 2,
"new_unique_states": 5,
"cumulative_states": 9
},
{
"depth": 3,
"new_unique_states": 6,
"cumulative_states": 15
... | [
{
"state": [
0,
1,
3,
2
],
"seeds": [
0
],
"retention_roles": [
"previous_small"
],
"evidence": [
{
"seed": 0,
"retention_role": "previous_small",
"source_file": "runs/b300-sweep-eight-20261011/cohorts/chunk-00000.tar.gz",... |
lrx:n4:r2 | lrx | 4 | 2 | COMPLETE | 3 | 12 | runs/b300-sweep-eight-20261011 | TryDotAtwo/LRX-BFS-8xB300-2026-10-11 | b23cc8c1629a395d057ef9cb009da3bfdd84889b | source_reported_two_pass_gpu | 16 | [
{
"depth": 0,
"new_unique_states": 1,
"cumulative_states": 1
},
{
"depth": 1,
"new_unique_states": 3,
"cumulative_states": 4
},
{
"depth": 2,
"new_unique_states": 5,
"cumulative_states": 9
},
{
"depth": 3,
"new_unique_states": 3,
"cumulative_states": 12
... | [
{
"state": [
0,
1,
1,
2
],
"seeds": [
13
],
"retention_roles": [
"previous_small"
],
"evidence": [
{
"seed": 13,
"retention_role": "previous_small",
"source_file": "runs/b300-sweep-eight-20261011/cohorts/chunk-00000.tar.gz... |
lrx:n4:r3 | lrx | 4 | 3 | COMPLETE | 2 | 4 | runs/b300-sweep-eight-20261011 | TryDotAtwo/LRX-BFS-8xB300-2026-10-11 | b23cc8c1629a395d057ef9cb009da3bfdd84889b | source_reported_two_pass_gpu | 3 | [
{
"depth": 0,
"new_unique_states": 1,
"cumulative_states": 1
},
{
"depth": 1,
"new_unique_states": 2,
"cumulative_states": 3
},
{
"depth": 2,
"new_unique_states": 1,
"cumulative_states": 4
}
] | [
{
"state": [
1,
0,
1,
1
],
"seeds": [
0,
13
],
"retention_roles": [
"previous_small"
],
"evidence": [
{
"seed": 0,
"retention_role": "previous_small",
"source_file": "runs/b300-sweep-eight-20261011/cohorts/chunk-0000... |
lrx:n4:r4 | lrx | 4 | 4 | COMPLETE | 0 | 1 | runs/b300-sweep-eight-20261011 | TryDotAtwo/LRX-BFS-8xB300-2026-10-11 | b23cc8c1629a395d057ef9cb009da3bfdd84889b | source_reported_two_pass_gpu | 1 | [
{
"depth": 0,
"new_unique_states": 1,
"cumulative_states": 1
}
] | [
{
"state": [
0,
0,
0,
0
],
"seeds": [
0,
13
],
"retention_roles": [
"current"
],
"evidence": [
{
"seed": 0,
"retention_role": "current",
"source_file": "runs/b300-sweep-eight-20261011/cohorts/chunk-00000.tar.gz",
... |
lrx:n5:r1 | lrx | 5 | 1 | COMPLETE | 10 | 120 | runs/b300-sweep-eight-20261011 | TryDotAtwo/LRX-BFS-8xB300-2026-10-11 | b23cc8c1629a395d057ef9cb009da3bfdd84889b | source_reported_two_pass_gpu | 8 | [
{
"depth": 0,
"new_unique_states": 1,
"cumulative_states": 1
},
{
"depth": 1,
"new_unique_states": 3,
"cumulative_states": 4
},
{
"depth": 2,
"new_unique_states": 6,
"cumulative_states": 10
},
{
"depth": 3,
"new_unique_states": 10,
"cumulative_states": 20
... | [
{
"state": [
0,
1,
2,
4,
3
],
"seeds": [
13
],
"retention_roles": [
"current"
],
"evidence": [
{
"seed": 13,
"retention_role": "current",
"source_file": "runs/b300-sweep-eight-20261011/cohorts/chunk-00000.tar.gz",
... |
lrx:n5:r2 | lrx | 5 | 2 | COMPLETE | 7 | 60 | runs/b300-sweep-eight-20261011 | TryDotAtwo/LRX-BFS-8xB300-2026-10-11 | b23cc8c1629a395d057ef9cb009da3bfdd84889b | source_reported_two_pass_gpu | 30 | [
{
"depth": 0,
"new_unique_states": 1,
"cumulative_states": 1
},
{
"depth": 1,
"new_unique_states": 3,
"cumulative_states": 4
},
{
"depth": 2,
"new_unique_states": 6,
"cumulative_states": 10
},
{
"depth": 3,
"new_unique_states": 9,
"cumulative_states": 19
... | [
{
"state": [
0,
2,
1,
3,
2
],
"seeds": [
13
],
"retention_roles": [
"previous_small"
],
"evidence": [
{
"seed": 13,
"retention_role": "previous_small",
"source_file": "runs/b300-sweep-eight-20261011/cohorts/chunk-000... |
lrx:n5:r3 | lrx | 5 | 3 | COMPLETE | 5 | 20 | runs/b300-sweep-eight-20261011 | TryDotAtwo/LRX-BFS-8xB300-2026-10-11 | b23cc8c1629a395d057ef9cb009da3bfdd84889b | source_reported_two_pass_gpu | 8 | [
{
"depth": 0,
"new_unique_states": 1,
"cumulative_states": 1
},
{
"depth": 1,
"new_unique_states": 3,
"cumulative_states": 4
},
{
"depth": 2,
"new_unique_states": 6,
"cumulative_states": 10
},
{
"depth": 3,
"new_unique_states": 6,
"cumulative_states": 16
... | [
{
"state": [
0,
1,
1,
2,
1
],
"seeds": [
13
],
"retention_roles": [
"previous_small"
],
"evidence": [
{
"seed": 13,
"retention_role": "previous_small",
"source_file": "runs/b300-sweep-eight-20261011/cohorts/chunk-000... |
lrx:n5:r4 | lrx | 5 | 4 | COMPLETE | 2 | 5 | runs/b300-sweep-eight-20261011 | TryDotAtwo/LRX-BFS-8xB300-2026-10-11 | b23cc8c1629a395d057ef9cb009da3bfdd84889b | source_reported_two_pass_gpu | 4 | [
{
"depth": 0,
"new_unique_states": 1,
"cumulative_states": 1
},
{
"depth": 1,
"new_unique_states": 2,
"cumulative_states": 3
},
{
"depth": 2,
"new_unique_states": 2,
"cumulative_states": 5
}
] | [
{
"state": [
1,
0,
1,
1,
1
],
"seeds": [
0,
13
],
"retention_roles": [
"previous_small"
],
"evidence": [
{
"seed": 0,
"retention_role": "previous_small",
"source_file": "runs/b300-sweep-eight-20261011/cohorts/c... |
lrx:n5:r5 | lrx | 5 | 5 | COMPLETE | 0 | 1 | runs/b300-sweep-eight-20261011 | TryDotAtwo/LRX-BFS-8xB300-2026-10-11 | b23cc8c1629a395d057ef9cb009da3bfdd84889b | source_reported_two_pass_gpu | 1 | [
{
"depth": 0,
"new_unique_states": 1,
"cumulative_states": 1
}
] | [
{
"state": [
0,
0,
0,
0,
0
],
"seeds": [
0,
13
],
"retention_roles": [
"current"
],
"evidence": [
{
"seed": 0,
"retention_role": "current",
"source_file": "runs/b300-sweep-eight-20261011/cohorts/chunk-00000.tar... |
lrx:n6:r1 | lrx | 6 | 1 | COMPLETE | 15 | 720 | runs/b300-sweep-eight-20261011 | TryDotAtwo/LRX-BFS-8xB300-2026-10-11 | b23cc8c1629a395d057ef9cb009da3bfdd84889b | source_reported_two_pass_gpu | 8 | [
{
"depth": 0,
"new_unique_states": 1,
"cumulative_states": 1
},
{
"depth": 1,
"new_unique_states": 3,
"cumulative_states": 4
},
{
"depth": 2,
"new_unique_states": 6,
"cumulative_states": 10
},
{
"depth": 3,
"new_unique_states": 11,
"cumulative_states": 21
... | [
{
"state": [
0,
1,
5,
4,
3,
2
],
"seeds": [
0
],
"retention_roles": [
"previous_small"
],
"evidence": [
{
"seed": 0,
"retention_role": "previous_small",
"source_file": "runs/b300-sweep-eight-20261011/cohorts/ch... |
lrx:n6:r2 | lrx | 6 | 2 | COMPLETE | 12 | 360 | runs/b300-sweep-eight-20261011 | TryDotAtwo/LRX-BFS-8xB300-2026-10-11 | b23cc8c1629a395d057ef9cb009da3bfdd84889b | source_reported_two_pass_gpu | 38 | [
{
"depth": 0,
"new_unique_states": 1,
"cumulative_states": 1
},
{
"depth": 1,
"new_unique_states": 3,
"cumulative_states": 4
},
{
"depth": 2,
"new_unique_states": 6,
"cumulative_states": 10
},
{
"depth": 3,
"new_unique_states": 10,
"cumulative_states": 20
... | [
{
"state": [
0,
1,
2,
2,
3,
4
],
"seeds": [
13
],
"retention_roles": [
"previous_small"
],
"evidence": [
{
"seed": 13,
"retention_role": "previous_small",
"source_file": "runs/b300-sweep-eight-20261011/cohorts/... |
lrx:n6:r3 | lrx | 6 | 3 | COMPLETE | 9 | 120 | runs/b300-sweep-eight-20261011 | TryDotAtwo/LRX-BFS-8xB300-2026-10-11 | b23cc8c1629a395d057ef9cb009da3bfdd84889b | source_reported_two_pass_gpu | 42 | [
{
"depth": 0,
"new_unique_states": 1,
"cumulative_states": 1
},
{
"depth": 1,
"new_unique_states": 3,
"cumulative_states": 4
},
{
"depth": 2,
"new_unique_states": 6,
"cumulative_states": 10
},
{
"depth": 3,
"new_unique_states": 10,
"cumulative_states": 20
... | [
{
"state": [
0,
1,
3,
3,
2,
3
],
"seeds": [
0
],
"retention_roles": [
"previous_small"
],
"evidence": [
{
"seed": 0,
"retention_role": "previous_small",
"source_file": "runs/b300-sweep-eight-20261011/cohorts/ch... |
lrx:n6:r4 | lrx | 6 | 4 | COMPLETE | 6 | 30 | runs/b300-sweep-eight-20261011 | TryDotAtwo/LRX-BFS-8xB300-2026-10-11 | b23cc8c1629a395d057ef9cb009da3bfdd84889b | source_reported_two_pass_gpu | 12 | [
{
"depth": 0,
"new_unique_states": 1,
"cumulative_states": 1
},
{
"depth": 1,
"new_unique_states": 3,
"cumulative_states": 4
},
{
"depth": 2,
"new_unique_states": 6,
"cumulative_states": 10
},
{
"depth": 3,
"new_unique_states": 7,
"cumulative_states": 17
... | [
{
"state": [
1,
0,
1,
1,
2,
1
],
"seeds": [
13
],
"retention_roles": [
"previous_small"
],
"evidence": [
{
"seed": 13,
"retention_role": "previous_small",
"source_file": "runs/b300-sweep-eight-20261011/cohorts/... |
lrx:n6:r5 | lrx | 6 | 5 | COMPLETE | 3 | 6 | runs/b300-sweep-eight-20261011 | TryDotAtwo/LRX-BFS-8xB300-2026-10-11 | b23cc8c1629a395d057ef9cb009da3bfdd84889b | source_reported_two_pass_gpu | 3 | [
{
"depth": 0,
"new_unique_states": 1,
"cumulative_states": 1
},
{
"depth": 1,
"new_unique_states": 2,
"cumulative_states": 3
},
{
"depth": 2,
"new_unique_states": 2,
"cumulative_states": 5
},
{
"depth": 3,
"new_unique_states": 1,
"cumulative_states": 6
}... | [
{
"state": [
1,
1,
0,
1,
1,
1
],
"seeds": [
0,
13
],
"retention_roles": [
"previous_small"
],
"evidence": [
{
"seed": 0,
"retention_role": "previous_small",
"source_file": "runs/b300-sweep-eight-20261011/... |
lrx:n6:r6 | lrx | 6 | 6 | COMPLETE | 0 | 1 | runs/b300-sweep-eight-20261011 | TryDotAtwo/LRX-BFS-8xB300-2026-10-11 | b23cc8c1629a395d057ef9cb009da3bfdd84889b | source_reported_two_pass_gpu | 1 | [
{
"depth": 0,
"new_unique_states": 1,
"cumulative_states": 1
}
] | [
{
"state": [
0,
0,
0,
0,
0,
0
],
"seeds": [
0,
13
],
"retention_roles": [
"current"
],
"evidence": [
{
"seed": 0,
"retention_role": "current",
"source_file": "runs/b300-sweep-eight-20261011/cohorts/chunk-... |
lrx:n7:r1 | lrx | 7 | 1 | COMPLETE | 21 | 5,040 | runs/b300-sweep-eight-20261011 | TryDotAtwo/LRX-BFS-8xB300-2026-10-11 | b23cc8c1629a395d057ef9cb009da3bfdd84889b | source_reported_two_pass_gpu | 8 | [
{
"depth": 0,
"new_unique_states": 1,
"cumulative_states": 1
},
{
"depth": 1,
"new_unique_states": 3,
"cumulative_states": 4
},
{
"depth": 2,
"new_unique_states": 6,
"cumulative_states": 10
},
{
"depth": 3,
"new_unique_states": 12,
"cumulative_states": 22
... | [
{
"state": [
0,
1,
6,
5,
4,
3,
2
],
"seeds": [
0
],
"retention_roles": [
"previous_small"
],
"evidence": [
{
"seed": 0,
"retention_role": "previous_small",
"source_file": "runs/b300-sweep-eight-20261011/c... |
lrx:n7:r2 | lrx | 7 | 2 | COMPLETE | 19 | 2,520 | runs/b300-sweep-eight-20261011 | TryDotAtwo/LRX-BFS-8xB300-2026-10-11 | b23cc8c1629a395d057ef9cb009da3bfdd84889b | source_reported_two_pass_gpu | 12 | [
{
"depth": 0,
"new_unique_states": 1,
"cumulative_states": 1
},
{
"depth": 1,
"new_unique_states": 3,
"cumulative_states": 4
},
{
"depth": 2,
"new_unique_states": 6,
"cumulative_states": 10
},
{
"depth": 3,
"new_unique_states": 11,
"cumulative_states": 21
... | [
{
"state": [
0,
1,
5,
5,
4,
3,
2
],
"seeds": [
0
],
"retention_roles": [
"previous_small"
],
"evidence": [
{
"seed": 0,
"retention_role": "previous_small",
"source_file": "runs/b300-sweep-eight-20261011/c... |
lrx:n7:r3 | lrx | 7 | 3 | COMPLETE | 15 | 840 | runs/b300-sweep-eight-20261011 | TryDotAtwo/LRX-BFS-8xB300-2026-10-11 | b23cc8c1629a395d057ef9cb009da3bfdd84889b | source_reported_two_pass_gpu | 40 | [
{
"depth": 0,
"new_unique_states": 1,
"cumulative_states": 1
},
{
"depth": 1,
"new_unique_states": 3,
"cumulative_states": 4
},
{
"depth": 2,
"new_unique_states": 6,
"cumulative_states": 10
},
{
"depth": 3,
"new_unique_states": 11,
"cumulative_states": 21
... | [
{
"state": [
0,
1,
2,
2,
3,
2,
4
],
"seeds": [
13
],
"retention_roles": [
"previous_small"
],
"evidence": [
{
"seed": 13,
"retention_role": "previous_small",
"source_file": "runs/b300-sweep-eight-20261011... |
lrx:n7:r4 | lrx | 7 | 4 | COMPLETE | 11 | 210 | runs/b300-sweep-eight-20261011 | TryDotAtwo/LRX-BFS-8xB300-2026-10-11 | b23cc8c1629a395d057ef9cb009da3bfdd84889b | source_reported_two_pass_gpu | 36 | [
{
"depth": 0,
"new_unique_states": 1,
"cumulative_states": 1
},
{
"depth": 1,
"new_unique_states": 3,
"cumulative_states": 4
},
{
"depth": 2,
"new_unique_states": 6,
"cumulative_states": 10
},
{
"depth": 3,
"new_unique_states": 11,
"cumulative_states": 21
... | [
{
"state": [
0,
2,
2,
1,
2,
3,
2
],
"seeds": [
13
],
"retention_roles": [
"previous_small"
],
"evidence": [
{
"seed": 13,
"retention_role": "previous_small",
"source_file": "runs/b300-sweep-eight-20261011... |
lrx:n7:r5 | lrx | 7 | 5 | COMPLETE | 7 | 42 | runs/b300-sweep-eight-20261011 | TryDotAtwo/LRX-BFS-8xB300-2026-10-11 | b23cc8c1629a395d057ef9cb009da3bfdd84889b | source_reported_two_pass_gpu | 16 | [
{
"depth": 0,
"new_unique_states": 1,
"cumulative_states": 1
},
{
"depth": 1,
"new_unique_states": 3,
"cumulative_states": 4
},
{
"depth": 2,
"new_unique_states": 6,
"cumulative_states": 10
},
{
"depth": 3,
"new_unique_states": 8,
"cumulative_states": 18
... | [
{
"state": [
0,
1,
1,
1,
2,
1,
1
],
"seeds": [
13
],
"retention_roles": [
"previous_small"
],
"evidence": [
{
"seed": 13,
"retention_role": "previous_small",
"source_file": "runs/b300-sweep-eight-20261011... |
lrx:n7:r6 | lrx | 7 | 6 | COMPLETE | 3 | 7 | runs/b300-sweep-eight-20261011 | TryDotAtwo/LRX-BFS-8xB300-2026-10-11 | b23cc8c1629a395d057ef9cb009da3bfdd84889b | source_reported_two_pass_gpu | 4 | [
{
"depth": 0,
"new_unique_states": 1,
"cumulative_states": 1
},
{
"depth": 1,
"new_unique_states": 2,
"cumulative_states": 3
},
{
"depth": 2,
"new_unique_states": 2,
"cumulative_states": 5
},
{
"depth": 3,
"new_unique_states": 2,
"cumulative_states": 7
}... | [
{
"state": [
1,
1,
0,
1,
1,
1,
1
],
"seeds": [
0,
13
],
"retention_roles": [
"previous_small"
],
"evidence": [
{
"seed": 0,
"retention_role": "previous_small",
"source_file": "runs/b300-sweep-eight-... |
lrx:n7:r7 | lrx | 7 | 7 | COMPLETE | 0 | 1 | runs/b300-sweep-eight-20261011 | TryDotAtwo/LRX-BFS-8xB300-2026-10-11 | b23cc8c1629a395d057ef9cb009da3bfdd84889b | source_reported_two_pass_gpu | 1 | [
{
"depth": 0,
"new_unique_states": 1,
"cumulative_states": 1
}
] | [
{
"state": [
0,
0,
0,
0,
0,
0,
0
],
"seeds": [
0,
13
],
"retention_roles": [
"current"
],
"evidence": [
{
"seed": 0,
"retention_role": "current",
"source_file": "runs/b300-sweep-eight-20261011/cohor... |
lrx:n8:r1 | lrx | 8 | 1 | COMPLETE | 28 | 40,320 | runs/b300-sweep-eight-20261011 | TryDotAtwo/LRX-BFS-8xB300-2026-10-11 | b23cc8c1629a395d057ef9cb009da3bfdd84889b | source_reported_two_pass_gpu | 8 | [
{
"depth": 0,
"new_unique_states": 1,
"cumulative_states": 1
},
{
"depth": 1,
"new_unique_states": 3,
"cumulative_states": 4
},
{
"depth": 2,
"new_unique_states": 6,
"cumulative_states": 10
},
{
"depth": 3,
"new_unique_states": 12,
"cumulative_states": 22
... | [
{
"state": [
0,
1,
7,
6,
5,
4,
3,
2
],
"seeds": [
0
],
"retention_roles": [
"previous_small"
],
"evidence": [
{
"seed": 0,
"retention_role": "previous_small",
"source_file": "runs/b300-sweep-eight-2... |
Graph BFS Atlas
One research table for recorded breadth-first searches on finite graphs. Each row describes one graph parameter pair: its run status, recorded layer counts and distinct retained state vectors. The initial collection covers the LRX and Rapaport M2 graph families. Source artifacts and provenance are recorded; a case may combine multiple physical runs.
| Family | Graph cases | Complete | Incomplete | Heuristically skipped | Recorded layers | Distinct retained vectors |
|---|---|---|---|---|---|---|
| LRX | 140 | 129 | 5 | 6 | 4,460 | 22,753 |
| Rapaport M2 | 141 | 132 | 5 | 4 | 5,890 | 40,639 |
| Total | 281 | 261 | 10 | 10 | 10,350 | 63,392 |
The canonical data file is atlas.parquet. Human-readable layer counts are available as derived CSV views: lrx-layers.csv and rapaport-m2-layers.csv. State vectors remain in Parquet. It is organized in one Parquet row group per graph, with statistics on family and parameters, so a filtered Parquet reader can select relevant row groups without downloading the entire table. The Hub dataset viewer also displays the table.
Select a graph
Install pyarrow, huggingface_hub and fsspec, then read just the desired graph:
import pyarrow.parquet as pq
from huggingface_hub import HfFileSystem
fs = HfFileSystem()
path = "datasets/TryDotAtwo/graph-bfs-atlas/atlas.parquet"
graph = pq.read_table(
path,
filesystem=fs,
filters=[("graph_family", "=", "lrx"), ("n", "=", 10), ("r", "=", 1)],
).to_pylist()[0]
print(graph["case_id"], graph["status"])
print(graph["layers"]) # depth, new_unique_states, cumulative_states
print([entry["state"] for entry in graph["states"]])
To select several graphs, supply a list of filter lists (logical OR), for example:
graphs = pq.read_table(path, filesystem=fs, filters=[
[("graph_family", "=", "lrx"), ("n", "=", 10), ("r", "=", 1)],
[("graph_family", "=", "rapaport_m2"), ("n", "=", 10), ("r", "=", 2)],
]).to_pylist()
You can also download the one Parquet file or load it with load_dataset("TryDotAtwo/graph-bfs-atlas", split="train").
Schema
case_id,graph_family,n,r: graph identifier and parameters. Join or filter bycase_id; its format isfamily:nN:rR.status: source status, one ofCOMPLETE,INCOMPLETEandRESOURCE_PRUNED_HEURISTIC. Incomplete runs have only completed depths. A heuristic skip is not a traversal.max_distance_from_canonical_start,observed_states: source-reported summary; missing values are null, not zero.layers: ordered list of records containingdepth,new_unique_statesandcumulative_states. Empty means no layer records are available for the case.states: list of distinct retained vectors for the graph. Each entry containsstate(signed 64-bit integer vector),seeds,retention_rolesandevidence.retained_unique_statesis the list length. Empty means no state vector was supplied in the imported payloads, not that the graph has no vertices.evidenceis a list of source occurrences. Each hasseed,retention_role,source_file, nullablearchive_member,source_index,source_schemaandcurrent_sample_limit. One vector occurs once per graph instates; its original occurrences are preserved inevidence(81,602 occurrences in total). Different seeds may use relabelled conventions: consult the source run before comparing vectors across seeds.source_repo,source_revision,evidence_prefix,evidence_kind: provenance of the combined case. Each graph family in this release comes from one pinned source revision;source_fileandarchive_memberlocate the state payload within it.
current denotes the retained current frontier in the source run, potentially sampled to at most 1,000 states. In an incomplete run it is not necessarily a final layer. previous_small is a retained previous frontier. The source files do not consistently attach explicit depth to each saved vector; do not infer it from the list or the reported eccentricity.
Scope and interpretation
- The table contains all vectors extracted from 787 available
states.jsonpayload occurrences in the pinned LRX and Rapaport sources (including their cohort archives), deduplicated by(case_id, state). It does not contain every vertex visited during BFS: historical layer vectors were not retained, and frontier retention was bounded. Ten cases have no saved vector in the imported payloads. - Rapaport
(15,2)and(16,4)use externally supplied CPU results for the combined summaries. Their final states were not supplied; historical incomplete GPU run samples, if present, must not be identified with those final layers. The CPU traversals were not independently rerun or verified by a second complete relabelled pass. - GPU
COMPLETEresults were reported as matching two relabelled passes and reachingn!/r!states. This publication checks imported data consistency; it does not rerun BFS. Distance is from the canonical start and, with repeated labels, is not automatically a proven global diameter. - Full original reports, graph definitions, executable code and validation artifacts are still in the pinned source repositories, not duplicated in this Parquet file. This release is therefore self-contained for its published case summaries, layer counts and retained state vectors, but not for all reproduction artifacts. These sweeps cover only part of the requested grid.
Provenance and extending the atlas
sources.json records the pinned source revisions and hashes for imported tables. Original datasets: LRX and Rapaport M2.
For a new case, publish available vectors in this Parquet table, preserve their exact encoding and source evidence, and explicitly record sampling, run completeness, starts and generators. Keep one vector per case with all occurrences in its evidence; do not fabricate missing state lists or measurements. Validate keys, layer sequences and sums, source coverage, deduplication, Parquet round trip, and filtered reading before publication.
Rapaport M2 n=15, r=1 update (2026-11-13)
Alexander C supplied all 118 layer counts and reported diameter 117 in a private message. The first 74 counts match the earlier incomplete source run; all counts sum to 15!. The result has not been independently rerun. No final-layer vectors were supplied; retained vectors for this case come from the earlier incomplete run and must not be interpreted as the final layer. The r=1 identification is inferred from the matching prefix.
- Downloads last month
- 329