PQC-Corpus

PQC-Corpus is the development history of 179 public GitHub repositories about post-quantum cryptography (PQC): their commits with full patches, their issues and pull requests with discussions and reviews, and the repositories themselves. It was collected on 21–24 September 2026.

File	One line is	Rows	Size
commits/	a commit, with its full patch	114,361	about 28 GB
commits_filtered/	the filtered set of commits	47,729	about 14 GB
issues.jsonl	an issue, with its comments and closing pull requests	5,713	26 MB
issues_filtered.jsonl	the filtered set of issues	5,687	26 MB
pull_requests.jsonl	a pull request, with its commits, closed issues, comments and reviews	11,919	107 MB
repositories.jsonl	a repository, with its corpus and GitHub metadata	179	80 KB

The commit folders hold parts of about 1 GB (part-00000.jsonl, ...); read them in order as one file. provenance/frozen_observations_2026-09-29.jsonl holds the 29 September 2026 observations the commits were mined with (default branch, merged pull request, PQC terms in linked pull requests), so the commit export can be rerun on the frozen collection.

Corpora

Every record has repo (owner/name) and corpus:

Corpus	Repositories	What was collected
A	81	every commit, issue and pull request; at least 14.8% of the repository's pull requests and issues name a PQC term
B	87	larger, general-purpose projects: only issues and pull requests whose title or body names a PQC term, and commits whose message names one or that belong to such a pull request
C	11	repositories with no issues or pull requests; every commit

The 90 search terms include scheme names (ML-KEM, Kyber, Dilithium, SPHINCS+, ...), standards (FIPS 203–206) and generic terms (post-quantum, liboqs).

Filtered sets

The full sets are everything collected. The filtered sets are reduced by fixed, scripted rules, with every removal listed in commits_filtered_report/ and issues_filtered_report/:

• Commits: only work that reached the default branch or a merged pull request; each change once (rebase, squash and merge copies removed); without history a fork inherited, commits with no PQC term in repositories not about PQC, and housekeeping commits (CI, lock files, repository metadata, bots).
• Issues: without the 26 issues opened by bot accounts.

Using the data

from datasets import load_dataset

issues = load_dataset("pqc-datasets/pqc-corpus", "issues_filtered", split="train")
commits = load_dataset("pqc-datasets/pqc-corpus", "commits_filtered", split="train", streaming=True)

Join on repo together with sha (commits) or number (issues, pull requests): SHAs and numbers repeat across repositories. A commit's pr_numbers names the pull requests that contain it; a pull request's closes_issues and an issue's closed_by_prs link the two. SCHEMA.md describes every field, and schemas/ holds a JSON Schema per file. manifest.json lists every file's rows and SHA-256; check the download with sha256sum -c SHA256SUMS.

Limitations

• Selection is by keyword. A record in the dataset is not necessarily about PQC, and corpus B misses PQC work that names no term in a title, body or message.
• Observations are as of collection. Later edits, deletions and renames on GitHub are not reflected. Of 131,978 distinct links checked on 28 October 2026, 131,949 point to the same object; one repository and two issues have since been deleted, and facebook/folly now redirects to meta/folly.
• Counts reflect the selection policy, not adoption. The corpora were selected differently, so pooled counts are not a measure of ecosystem-wide PQC activity.

Personal data

Usernames and Git author names are kept as published on GitHub. E-mail addresses are not included as fields; commits carry only author_key, a keyed hash of the author's address made with a secret that is not published, so that copies of one commit can be matched without a known address being testable against the data. Addresses written inside messages and patches, such as Signed-off-by lines, are unchanged. Use the data for research on software development, not to profile individuals.

Licence

The dataset's selection, structure and documentation are licensed under CC BY 4.0. The content of the records was written by the contributors of the source repositories and remains under those repositories' licences; see LICENSE.

Citation

If you use PQC-Corpus, please cite it. The DOI 10.5281/zenodo.23784229 always resolves to the latest version; version 1.0.0 is 10.5281/zenodo.23785041.

@software{pqc_corpus_2026,
  author    = {Saha Dip, Souhardya and Fakorede, Moshood A. and Farooq, Umar},
  title     = {{PQC-Corpus: A Dataset of Post-Quantum Cryptography Development Artifacts on GitHub}},
  year      = {2026},
  publisher = {Zenodo},
  version   = {1.0.0},
  doi       = {10.5281/zenodo.23784229},
  url       = {https://doi.org/10.5281/zenodo.23784229}
}