File size: 3,100 Bytes
8979fc9
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
---
pretty_name: Anonymized Iterative Code Benchmark Problems
license: mit
task_categories:
- text-generation
tags:
- code
- benchmark
- iterative-specification
- python
- js
- cpp
- rust
- java
- anonymized
size_categories:
- n<1K
configs:
- config_name: python
  data_files:
  - split: test
    path: data/python/test.parquet
- config_name: js
  data_files:
  - split: test
    path: data/js/test.parquet
- config_name: cpp
  data_files:
  - split: test
    path: data/cpp/test.parquet
- config_name: rust
  data_files:
  - split: test
    path: data/rust/test.parquet
- config_name: java
  data_files:
  - split: test
    path: data/java/test.parquet
---
# Anonymized Iterative Code Benchmark Problems

This anonymized dataset contains multi-checkpoint coding tasks for
evaluating agents under iterative specification refinement. Problem
identifiers, source repository identifiers, and author fields are
intentionally omitted.

## Dataset summary

- Language subsets: `python`, `js`, `cpp`, `rust`, `java`
- Problems: 36 problems
- Rows: 980 rows
- Columns: `problem`, `Checkpoint number`, `language`, `difficulty`, `tags`, `instruction`, `tests`

Rows per subset:

- `python`: 196 rows
- `js`: 196 rows
- `cpp`: 196 rows
- `rust`: 196 rows
- `java`: 196 rows

Category distribution:

- developer-tools: 8
- web: 7
- data-processing: 6
- cli-tools: 5
- algorithms: 2
- configuration-management: 2
- dsl: 2
- databases: 1
- file-systems: 1
- networking: 1
- simulation: 1

Difficulty distribution:

- Easy: 12
- Hard: 12
- Medium: 12

## Using the `tests` field

Each row includes a `tests` value containing a complete Bash test runner
for that checkpoint. Save it as a file, then run it against a submission
directory whose contents implement the entry file named in the
`instruction` field.

### Test script examples

Write and run one checkpoint from Python:

```python
from pathlib import Path
import subprocess

from datasets import load_dataset

dataset = load_dataset("anon/dsn26", "python", split="test")
row = dataset[0]

test_path = Path("test.sh")
test_path.write_text(row["tests"])
test_path.chmod(0o755)

subprocess.run(["bash", str(test_path), "/path/to/submission"], check=True)
```

Run a saved test script directly:

```bash
bash test.sh /path/to/submission
```

Use a local problems checkout to avoid cloning:

```bash
BENCH_PROBLEMS_REPO=/path/to/problems bash test.sh /path/to/submission
```

Keep the temporary workspace for debugging:

```bash
BENCH_KEEP_WORKDIR=1 bash test.sh /path/to/submission
```

The runner creates a temporary workspace, copies the submission into it,
materializes the relevant checkpoint tests and static assets, installs
Python test dependencies in a virtual environment, and invokes `pytest`.
Set `BENCH_PROBLEMS_REPO` to an existing problems checkout to
avoid cloning. Set `BENCH_KEEP_WORKDIR=1` to keep the temporary
workspace for debugging.

For `cpp`, `rust`, and `java` subsets, the runner compiles the declared
source file with `g++`, `rustc`, or `javac` before running the Python
test harness. The `js` subset expects `node` to be available.