CODEOWNERS & Review Accountability (2024–2025)

This repository reproduces the dataset and analyses used to study how CODEOWNERS governance affects pull-request (PR) review latency and merge latency across open-source repositories.
The workflow combines BigQuery SQL queries with lightweight Python utilities for metadata extraction and data validation.

⚠️ Data Volume & Cost Notice
Running the full workflow on the GitHub Archive involves multi-terabyte scans and can incur significant BigQuery costs depending on your billing configuration.
For experimentation, restrict queries to a smaller date range (for example, a few months) without affecting reproducibility.
Always enable cost controls or query quotas before execution.

The full dataset and supporting files are permanently archived on Zenodo (DOI 10.5281/zenodo.17375757).

---

Repository Layout

code/
├── scripts/
│   └── codeowner_scan.py              # Queries the GitHub API for CODEOWNERS metadata
├── sql/
│   ├── active_repos_pr.sql
│   ├── active_repos_pr_top_2000.sql
│   ├── mergelatency_reviewlatency_final.sql
│   ├── owner_band_summary.sql
├── output/
│   ├── active_repos.csv
│   ├── active_repos_top2000.csv
│   ├── codeowners_meta.csv
│   ├── msr-final-results.csv (Hosted on Zenodo)
│   ├── owner_band_summary.csv (Hosted on Zenodo)
└── figs/
    └── merge_time_vs_owners.png

Additional Files  
* `README.md` — project documentation  
* `requirements.txt` — Python dependencies  

---

Data Overview

All output files produced from the SQL queries are stored in the output/ directory.
These datasets were generated using the workflow described below and form the basis for the results reported in the accompanying analysis.

File	Source	Description
active_repos.csv	active_repos_pr.sql	All repositories with ≥ 1 PR event (Jan 2024 – Sep 2025)
active_repos_top2000.csv	active_repos_pr_top_2000.sql	The 2 000 most active repositories
codeowners_meta.csv	codeowner_scan.py	CODEOWNERS presence, creation date, and number of owners
msr-final-results.csv	mergelatency_reviewlatency_final.sql	PR-level dataset (GZ - Hosted on Zenodo - Link already provided)
owner_band_summary.csv	owner_band_summary.sql	Summary of governed PRs grouped by ownership size

---

Quickstart — Reproducing the Analysis

Environment
All SQL queries were executed directly on the BigQuery console (standard SQL dialect).
Resulting CSVs were exported locally into the output/ directory for validation and visualization.
No proprietary or restricted APIs were used.

---

Step 1 — Identify Active Repositories

Find repositories with at least one PR event between January 2024 and September 2025.
Query: active_repos_pr.sql
Output: active_repos.csv — list of active repositories and their PR event counts.

Identified approximately 76 316 repositories.

---

Step 2 — Select the Top 2 000 Repositories

Filter the most active repositories by PR count.
Query: active_repos_pr_top_2000.sql
Output: active_repos_top2000.csv — top 2 000 repositories (repo_name, pr_events).

---

Step 3 — Detect CODEOWNERS Metadata

Query the GitHub API to detect whether each repository contains a CODEOWNERS file, when it was created, and how many owners are listed.

export GITHUB_TOKEN="ghp_yourTokenHere"
python code/scripts/codeowner_scan.py code/output/active_repos_top2000.csv code/output/codeowners_meta.csv

Output: codeowners_meta.csv
Fields: repo_name, has_codeowners, codeowners_created_at, owners_count

---

Step 4 — Upload CODEOWNERS Metadata to BigQuery

Upload codeowners_meta.csv to your BigQuery project for analysis.

repo_name STRING,
has_codeowners BOOLEAN,
codeowners_created_at TIMESTAMP,
owners_count INTEGER

Destination example: YOUR_PROJECT.msr2026.codeowners_meta

---

Step 5 — Compute PR Latency Metrics and Join

Calculate merge latency and first-review latency for each PR.
Query: mergelatency_reviewlatency_final.sql

Outputs

• pr_merge_latency_sample — per merged PR (repo_name, pr_number, pr_open_ts, pr_merge_ts, hours_to_merge)
• pr_review_latency_sample — first review timestamp (repo_name, pr_number, first_review_any_ts)
• msr_final_sample — joined dataset including CODEOWNERS metadata

Exported File: msr-final-results.csv

---

Step 6 — Summarize Governed PRs by Ownership Band

Aggregate PRs only where ownership existed before the first review, grouped by the number of owners.
Query: owner_band_summary.sql
Output: owner_band_summary.csv

Ownership Band	PRs	Repositories	Median First Review (h)	Median Merge (h)	Merge IQR (h)
1 Owner	57 255	139	3	48	158
2 Owners	31 545	73	5	62	160
3–6 Owners	110 926	178	9	79	267
7–10 Owners	71 386	88	7	70	232
> 10 Owners	253 860	200	11	110	340

This summary demonstrates that as ownership groups expand, review times lengthen and merge predictability decreases.

---

Results Overview

The analysis reveals a clear pattern: smaller ownership groups review and merge code significantly faster, while larger groups experience longer delays and higher variability.
This figure visualizes how merge latency scales with ownership size across all governed repositories.

Figure: Median merge time (hours) increases as ownership groups expand, reflecting added coordination overhead.

---

Reproducibility

All steps are deterministic and reproducible using only public GitHub data from the GitHub Archive.
Generated CSVs are included for verification, secondary analysis, and transparency.

---

Data Availability

The full dataset supporting this analysis is archived on Zenodo:

📦 Zenodo DOI: 10.5281/zenodo.17375757

Contents:

• msr-final-results.csv.gz — Complete PR-level dataset
• owner_band_summary.csv — Aggregated metrics grouped by ownership band

All materials are released under a Creative Commons Attribution 4.0 International (CC BY 4.0) license.
Accompanying SQL queries and scripts are licensed under MIT in this GitHub repository.

---

Citation

If you use this dataset or replicate the analysis, please cite:

Dhawan, R. (2025). CODEOWNERS & Review Accountability (2024–2025): Open-source pull request latency dataset.
Zenodo. https://doi.org/10.5281/zenodo.17375757

---