Datasets:
task_id large_stringlengths 7 7 | task_type large_stringclasses 1
value | difficulty large_stringclasses 3
values | difficulty_score float64 0.22 7.6 | repo large_stringclasses 11
values | pr_number int64 37 37.1k | pr_title large_stringlengths 12 122 | pr_url large_stringlengths 43 63 | pr_labels large_stringclasses 83
values | base_commit large_stringlengths 40 40 | merge_commit large_stringlengths 40 40 | files_changed large_stringlengths 19 452 | primary_file large_stringlengths 8 87 | problem_statement large_stringlengths 159 20.6k | golden_patch large_stringlengths 294 77.4k | loc_added int64 0 1.21k | loc_deleted int64 0 972 | loc int64 1 1.45k | task_quality_score float64 4 9.5 | statement_strategy large_stringclasses 2
values | has_linked_issue bool 2
classes | pr_body_length int64 0 2.68k | files_changed_count int64 1 8 | new_theorems_count int64 0 112 | new_lemmas_count int64 0 61 | new_definitions_count int64 0 30 | proof_source large_stringclasses 1
value | human_verified bool 1
class | lean_toolchain large_stringclasses 1
value | docker_image large_stringclasses 1
value | verification_command large_stringclasses 1
value | timeout_seconds int64 600 600 | verification_method large_stringclasses 1
value | domain large_stringclasses 8
values | created_at large_stringdate 2026-04-28 12:39:47 2026-04-28 12:43:22 | created_by large_stringclasses 1
value |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
LB-0001 | pr_completion | easy | 0.44 | ImperialCollegeLondon/FLT | 910 | fix(CI): Blueprint test driver | https://github.com/ImperialCollegeLondon/FLT/pull/910 | ed4af4822818ad935393f983051e3bf27308e855 | 47c2d775b991e9f8dacf77a375b02e4c0edf613a | FLTTest.lean,FLTTest/FLTTest.lean | FLTTest.lean | # Task: fix(CI): Blueprint test driver
## Context
Checking if the band aid works in the main repo, consists of a CI step to forcibly relocate the `FLTTest.olean` for the blueprint checkdecls to see.
## Files affected
- FLTTest.lean (+1/-0)
- FLTTest/FLTTest.lean (+1/-0)
## Verification
The patched repository must co... | diff --git a/FLTTest.lean b/FLTTest.lean
--- a/FLTTest.lean
+++ b/FLTTest.lean
@@ -1 +1,2 @@
+import FLTTest.FLTTest
import FLTTest.MathlibCompatibility
diff --git a/FLTTest/FLTTest.lean b/FLTTest/FLTTest.lean
--- /dev/null
+++ b/FLTTest/FLTTest.lean
@@ -0,0 +1 @@
+import FLTTest.MathlibCompatibility | 2 | 0 | 2 | 6 | diff_derived | false | 148 | 2 | 0 | 0 | 0 | pr_merge | false | leanprover/lean4:v4.29.0 | leanbench/lean4-env:4.29.0 | lake build | 600 | lean_compiler | general | 2026-04-28T12:39:47Z | extraction_pipeline_v2 | |
LB-0002 | pr_completion | easy | 3.1 | ImperialCollegeLondon/FLT | 898 | feat(Patching/Utils/CompactHausdorffRings): add partial proof that compact Hausdorff rings are totally disconnected | https://github.com/ImperialCollegeLondon/FLT/pull/898 | d346976c67e8437079ff1fd2b7878506e9ad0f37 | e8c35753f741cf30ffc59dcec88b5ce4e36f37b7 | FLT.lean,FLT/Patching/Utils/AdicTopology.lean,FLT/Patching/Utils/CompactHausdorffRings.lean | FLT.lean | # Task: feat(Patching/Utils/CompactHausdorffRings): add partial proof that compact Hausdorff rings are totally disconnected
## Context
I don't have immediately plans to finish this off (since @YaelDillies is actively working in this area and has ideas for how to do things in the right generality for mathlib), so I'm p... | diff --git a/FLT.lean b/FLT.lean
--- a/FLT.lean
+++ b/FLT.lean
@@ -161,6 +161,7 @@ import FLT.Patching.REqualsT
import FLT.Patching.System
import FLT.Patching.Ultraproduct
import FLT.Patching.Utils.AdicTopology
+import FLT.Patching.Utils.CompactHausdorffRings
import FLT.Patching.Utils.Depth
import FLT.Patching.Uti... | 137 | 0 | 137 | 8 | diff_derived | false | 224 | 3 | 3 | 0 | 0 | pr_merge | false | leanprover/lean4:v4.29.0 | leanbench/lean4-env:4.29.0 | lake build | 600 | lean_compiler | algebra | 2026-04-28T12:39:47Z | extraction_pipeline_v2 | |
LB-0003 | pr_completion | easy | 1.6 | ImperialCollegeLondon/FLT | 887 | fix: finiteness files in `4.29` bump | https://github.com/ImperialCollegeLondon/FLT/pull/887 | 4554fdf0c9ff937384bda4692e603572c6adb7f0 | 504213abdafa05a5508d31f25b03dc9fc3c09e99 | FLT/AutomorphicForm/QuaternionAlgebra/FiniteDimensional.lean,FLT/DivisionAlgebra/Finiteness.lean,FLT/Mathlib/MeasureTheory/Haar/Extension.lean | FLT/AutomorphicForm/QuaternionAlgebra/FiniteDimensional.lean | # Task: fix: finiteness files in `4.29` bump
## Context
- A couple more `maxHeartbeats 0` required
- Remove upstreamed material in `MeasureTheory.Haar.Extension`
## Files affected
- FLT/AutomorphicForm/QuaternionAlgebra/FiniteDimensional.lean (+1/-0)
- FLT/DivisionAlgebra/Finiteness.lean (+76/-9)
- FLT/Mathlib/Measu... | diff --git a/FLT/AutomorphicForm/QuaternionAlgebra/FiniteDimensional.lean b/FLT/AutomorphicForm/QuaternionAlgebra/FiniteDimensional.lean
--- a/FLT/AutomorphicForm/QuaternionAlgebra/FiniteDimensional.lean
+++ b/FLT/AutomorphicForm/QuaternionAlgebra/FiniteDimensional.lean
@@ -28,6 +28,7 @@ open TotallyDefiniteQuaternionA... | 85 | 298 | 383 | 5 | diff_derived | false | 106 | 3 | 0 | 0 | 0 | pr_merge | false | leanprover/lean4:v4.29.0 | leanbench/lean4-env:4.29.0 | lake build | 600 | lean_compiler | algebra | 2026-04-28T12:39:47Z | extraction_pipeline_v2 | |
LB-0004 | pr_completion | easy | 1.06 | ImperialCollegeLondon/FLT | 886 | fix: `4.29` fixes for `HaarChar.AdeleRing` | https://github.com/ImperialCollegeLondon/FLT/pull/886 | 7c8ac5865d778d3c2eafd3fe7dfb83803d8dc41c | 4554fdf0c9ff937384bda4692e603572c6adb7f0 | FLT/HaarMeasure/HaarChar/AdeleRing.lean | FLT/HaarMeasure/HaarChar/AdeleRing.lean | # Task: fix: `4.29` fixes for `HaarChar.AdeleRing`
## Files affected
- FLT/HaarMeasure/HaarChar/AdeleRing.lean (+22/-21)
## Verification
The patched repository must compile with `lake build`. | diff --git a/FLT/HaarMeasure/HaarChar/AdeleRing.lean b/FLT/HaarMeasure/HaarChar/AdeleRing.lean
--- a/FLT/HaarMeasure/HaarChar/AdeleRing.lean
+++ b/FLT/HaarMeasure/HaarChar/AdeleRing.lean
@@ -40,6 +40,7 @@ open scoped TensorProduct
open NumberField MeasureTheory
+set_option backward.isDefEq.respectTransparency fals... | 22 | 21 | 43 | 6 | diff_derived | false | 0 | 1 | 0 | 0 | 0 | pr_merge | false | leanprover/lean4:v4.29.0 | leanbench/lean4-env:4.29.0 | lake build | 600 | lean_compiler | algebra | 2026-04-28T12:39:47Z | extraction_pipeline_v2 | |
LB-0005 | pr_completion | medium | 5 | ImperialCollegeLondon/FLT | 885 | fix: `4.29` fixes for finite adele files | https://github.com/ImperialCollegeLondon/FLT/pull/885 | 166dbe151303e083f2d18a2478f7d3bb218ee511 | b30cfb53c2a30eb96fb19c2b62b670075dd44fd1 | FLT/DedekindDomain/Completion/BaseChange.lean,FLT/DedekindDomain/FiniteAdeleRing/BaseChange.lean,FLT/DedekindDomain/IntegralClosure.lean,FLT/HaarMeasure/HaarChar/FiniteAdeleRing.lean,FLT/NumberField/AdeleRing.lean | FLT/DedekindDomain/Completion/BaseChange.lean | # Task: fix: `4.29` fixes for finite adele files
## Context
Some `maxHeartbeats 0` required in
- `Completion.BaseChange`
- `FiniteAdeleRing.BaseChange`
- `HaarCharacter.FiniteAdeleRing`
## Files affected
- FLT/DedekindDomain/Completion/BaseChange.lean (+134/-97)
- FLT/DedekindDomain/FiniteAdeleRing/BaseChange.lea... | diff --git a/FLT/DedekindDomain/Completion/BaseChange.lean b/FLT/DedekindDomain/Completion/BaseChange.lean
--- a/FLT/DedekindDomain/Completion/BaseChange.lean
+++ b/FLT/DedekindDomain/Completion/BaseChange.lean
@@ -40,6 +40,7 @@ map is continuous, `K_v`-linear and restricts to an isomorphism `B ⊗_A 𝓞_v
-/
open sco... | 267 | 167 | 434 | 6.5 | diff_derived | false | 129 | 5 | 1 | 3 | 2 | pr_merge | false | leanprover/lean4:v4.29.0 | leanbench/lean4-env:4.29.0 | lake build | 600 | lean_compiler | algebra | 2026-04-28T12:39:48Z | extraction_pipeline_v2 | |
LB-0006 | pr_completion | easy | 2.4 | ImperialCollegeLondon/FLT | 883 | fix: `WithZeroMulInt` issues for bump to 4.29.0 | https://github.com/ImperialCollegeLondon/FLT/pull/883 | 6e04ca51c224b2ef37db017617e56bf917858a68 | df1bd1124322a770996e7cb67e747e6802c25ea8 | FLT/Mathlib/Topology/Algebra/Valued/WithZeroMulInt.lean,FLT/NumberField/Completion/Finite.lean | FLT/Mathlib/Topology/Algebra/Valued/WithZeroMulInt.lean | # Task: fix: `WithZeroMulInt` issues for bump to 4.29.0
## Files affected
- FLT/Mathlib/Topology/Algebra/Valued/WithZeroMulInt.lean (+21/-46)
- FLT/NumberField/Completion/Finite.lean (+1/-0)
## Theorems to prove
### `finite_cover_of_uniformity_basis` in `FLT/Mathlib/Topology/Algebra/Valued/WithZeroMulInt.lean`
```le... | diff --git a/FLT/Mathlib/Topology/Algebra/Valued/WithZeroMulInt.lean b/FLT/Mathlib/Topology/Algebra/Valued/WithZeroMulInt.lean
--- a/FLT/Mathlib/Topology/Algebra/Valued/WithZeroMulInt.lean
+++ b/FLT/Mathlib/Topology/Algebra/Valued/WithZeroMulInt.lean
@@ -9,6 +9,9 @@ import Mathlib.RingTheory.DiscreteValuationRing.Basic... | 22 | 46 | 68 | 7 | diff_derived | false | 0 | 2 | 2 | 0 | 0 | pr_merge | false | leanprover/lean4:v4.29.0 | leanbench/lean4-env:4.29.0 | lake build | 600 | lean_compiler | algebra | 2026-04-28T12:39:48Z | extraction_pipeline_v2 | |
LB-0007 | pr_completion | hard | 7.4 | ImperialCollegeLondon/FLT | 882 | fix: `WithAbs` issues in 4.29.0 bump | https://github.com/ImperialCollegeLondon/FLT/pull/882 | bd1033acf6742eb55c375c2f5a7fafb9e0ef6d25 | 5b0ecd9f4770fe3425cff1af11a8ad5bf1d7e243 | "FLT.lean,FLT/Mathlib/Algebra/Algebra/Hom.lean,FLT/Mathlib/Analysis/Normed/Ring/WithAbs.lean,FLT/Num(...TRUNCATED) | FLT.lean | "# Task: fix: `WithAbs` issues in 4.29.0 bump\n\n## Context\n- `WeakApproximation.lean` is now in ma(...TRUNCATED) | "diff --git a/FLT.lean b/FLT.lean\n--- a/FLT.lean\n+++ b/FLT.lean\n@@ -152,7 +152,6 @@ import FLT.Nu(...TRUNCATED) | 210 | 972 | 1,182 | 4.5 | diff_derived | false | 450 | 7 | 9 | 0 | 2 | pr_merge | false | leanprover/lean4:v4.29.0 | leanbench/lean4-env:4.29.0 | lake build | 600 | lean_compiler | algebra | 2026-04-28T12:39:48Z | extraction_pipeline_v2 | |
LB-0008 | pr_completion | easy | 2.4 | leanprover-community/mathlib4 | 37,059 | review: #36709 | https://github.com/leanprover-community/mathlib4/pull/37059 | ready-to-merge,t-meta | 9450d934e6c23ce226e8ddfa16ebd28f9cab1df4 | 23a8036a7a7f35719b883ebf20f5fa25c4b2a65e | Mathlib/Tactic/DefEqAbuse.lean,MathlibTest/DefEqAbuse.lean | Mathlib/Tactic/DefEqAbuse.lean | "# Task: review: #36709\n\n## Context\nThis review suggests\r\n- various improvements to the code st(...TRUNCATED) | "diff --git a/Mathlib/Tactic/DefEqAbuse.lean b/Mathlib/Tactic/DefEqAbuse.lean\n--- a/Mathlib/Tactic/(...TRUNCATED) | 77 | 66 | 143 | 8 | diff_derived | false | 557 | 2 | 0 | 0 | 2 | pr_merge | false | leanprover/lean4:v4.29.0 | leanbench/lean4-env:4.29.0 | lake build | 600 | lean_compiler | general | 2026-04-28T12:39:49Z | extraction_pipeline_v2 |
LB-0009 | pr_completion | easy | 1.7 | leanprover-community/mathlib4 | 36,880 | review: MessageData.ForExprs | https://github.com/leanprover-community/mathlib4/pull/36880 | t-meta | 35d565bba7a120c88be1e421e84ff70e8778dc3c | 94acb1345e3767c673cd2cf656f86e25d5f4e6ec | Mathlib/Lean/MessageData/ForExprs.lean | Mathlib/Lean/MessageData/ForExprs.lean | "# Task: review: MessageData.ForExprs\n\n## Context\nThis review of #36796 suggests implementing `fo(...TRUNCATED) | "diff --git a/Mathlib/Lean/MessageData/ForExprs.lean b/Mathlib/Lean/MessageData/ForExprs.lean\n--- a(...TRUNCATED) | 71 | 37 | 108 | 8.5 | diff_derived | false | 729 | 1 | 0 | 0 | 1 | pr_merge | false | leanprover/lean4:v4.29.0 | leanbench/lean4-env:4.29.0 | lake build | 600 | lean_compiler | general | 2026-04-28T12:39:53Z | extraction_pipeline_v2 |
LB-0010 | pr_completion | easy | 3.4 | leanprover-community/mathlib4 | 35,843 | review: just the basics | https://github.com/leanprover-community/mathlib4/pull/35843 | delegated,t-meta | be3f919ded3ff56ae14edfbbe776bb5542a929f0 | 93584a945344b3c995cac4249c4e1ae747998b02 | Mathlib/Lean/MessageData/Trace.lean,Mathlib/Tactic/DefEqAbuse.lean | Mathlib/Lean/MessageData/Trace.lean | "# Task: review: just the basics\n\n## Context\nThis ultra-quick review of #35750 doesn't check for (...TRUNCATED) | "diff --git a/Mathlib/Lean/MessageData/Trace.lean b/Mathlib/Lean/MessageData/Trace.lean\n--- a/Mathl(...TRUNCATED) | 139 | 128 | 267 | 8.5 | diff_derived | false | 421 | 2 | 0 | 0 | 4 | pr_merge | false | leanprover/lean4:v4.29.0 | leanbench/lean4-env:4.29.0 | lake build | 600 | lean_compiler | general | 2026-04-28T12:39:53Z | extraction_pipeline_v2 |
End of preview. Expand in Data Studio
LeanBench Dataset
A benchmark dataset for evaluating AI systems on Lean 4 theorem proving tasks.
Dataset Description
This dataset contains 482 tasks extracted from real Lean 4 pull requests.
Files
leanbench_tasks.csv- Full dataset in CSV formatdata/train-00000-of-00001.parquet- Dataset in Parquet format (fordatasetslibrary)
Task Format
Each row represents a single task with the following key fields:
| Field | Description |
|---|---|
task_id |
Unique identifier (e.g., LB-0001) |
task_type |
Type of task (e.g., pr_completion) |
difficulty |
Difficulty level (easy/medium/hard) |
difficulty_score |
Numeric difficulty score |
repo |
Source GitHub repository |
pr_number |
Pull request number |
problem_statement |
Natural language description of the task |
golden_patch |
Expected solution (diff format) |
verification_command |
Command to verify the solution |
Usage
from datasets import load_dataset
dataset = load_dataset("foundry-ai/leanbench")
# Access tasks
for task in dataset["train"]:
print(task["task_id"], task["difficulty"])
Statistics
- Total tasks: 482
- Easy: 330
- Medium: 121
- Hard: 31
License
Apache 2.0
- Downloads last month
- 49