README.md

---
license: cc-by-4.0
task_categories:
- text-classification
language:
- en
tags:
- code
- debugging
- behavioral-data
- software-engineering
pretty_name: DebugTraj-50K
size_categories:
- 100K<n<1M
---
# DebugTraj-50K: Developer Debugging Trajectory Dataset

50,000 debugging sessions with 665,364 step-by-step behavioral events across 8 programming languages and 71 error types.

## Load in one line

```python
import pandas as pd
sessions = pd.read_csv("hf://datasets/pyofpython/debugtraj-50k/code_debugging_sessions_fixed.csv")
events   = pd.read_csv("hf://datasets/pyofpython/debugtraj-50k/code_debugging_events_fixed.csv")
```

## What makes this unique

Most bug datasets only say "bug was fixed in 23 minutes."
This dataset records every step taken during those 23 minutes — every compile, edit, search, test, and commit.

## Files

| File | Rows | Description |
|------|------|-------------|
| code_debugging_sessions_fixed.csv | 50,000 | Session features (38 columns) |
| code_debugging_events_fixed.csv | 665,364 | Event trajectories (9 columns) |
| DATA_DICTIONARY_updated.md | — | Full column documentation |

## Use cases

- Predict debugging time from early session features
- Classify whether a session will succeed or be abandoned
- Next-action prediction for AI coding assistants
- Study junior vs senior developer behavior patterns

## Citation

```bibtex
@dataset{singh2026debugtraj,
  author    = {Abhishek Singh},
  title     = {DebugTraj-50K: Developer Debugging Trajectory Dataset},
  year      = {2026},
  publisher = {HuggingFace},
  url       = {https://huggingface.co/datasets/pyofpython/debugtraj-50k}
}
```