{ "seed": 0, "split": "file_split", "n_theorems": 64031, "n_test_theorems": 6422, "n_usable_steps": 303870, "n_test_steps": 30856, "tasks": { "t1": { "n": 5695, "labels": "silver labels: dotted identifiers mined from tactic [...] groups", "file": "t1.jsonl" }, "t2": { "n": 30856, "labels": "positives = all tactics used at this canonical state (multi-positive)", "file": "t2.jsonl" }, "t3": { "n": 30856, "labels": "positives = all tactics for this (before, after) transition", "file": "t3.jsonl" }, "t4": { "n": 3516, "labels": "positives = states with identical canonical hash", "file": "t4.jsonl" }, "t5": { "n": 55, "labels": "positives = statements with identical canonical hash; may be empty", "file": "t5.jsonl" }, "t6": { "n": 154280, "labels": "weak labels: 1 = tactic used at this state (true); 0 = assumed negative", "file": "t6.jsonl" }, "t7": { "n": 30358, "labels": "positive = canonical renamed variant (hyps -> v0, v1, ...); negatives = 5 token-overlap hard + 4 random", "file": "t7.jsonl" }, "t8": { "n": 6418, "labels": "area = second path component of file_path (e.g. Mathlib/Algebra -> Algebra)", "file": "t8.jsonl" } } }