The Dataset
Most code corpora flatten software into disconnected files or final snapshots. Codebase Collection keeps each project coherent: source, tests, documentation, configuration, commit topology, technical messages, timestamps, branches, tags, merges, reversions, and multi-file changes across 178 real-world codebases.
The result supports repository-level training, long-context code understanding, patch and commit reasoning, test generation, code review, refactoring, regression analysis, and the construction of realistic coding-agent evaluations from authentic history.
Coherent current repository snapshots.
Unique reachable historical text.
178 cloneable Git bundles plus evidence.
Exact Composition
Start with the corpus, not the showcase sample. These measurements define the release buyers receive and distinguish unique Git objects from per-repository totals.
| Metric | Exact value | Definition |
|---|---|---|
| Repositories | 178 | Complete retained codebases after empty and duplicate removal |
| Commits | 16,555 / 15,238 | Summed per repository / unique sanitized commit objects |
| Merge commits | 868 | Summed across delivered repositories |
| Heads and tags | 548 | Approved refs retained in prerequisite-free bundles |
| Current snapshots | 23,633 files | 1,379,283 measured code lines |
| Historical text | 61,454 blobs | 47,783,501 lines across unique reachable text |
| Pull requests | 464 / 444 merged | Aggregate source-inventory metadata; PR bodies are not claimed as delivered |
| Contributors | 99 | Unique retained contributor display names |
Download, Clone, Inspect
The free sample is one complete Solidity settlement and security repository, not a synthetic record or a folder of selected snippets. Download the archive, verify its checksums, clone the bundled history, and inspect the same package conventions used by the collection.
Solidity Settlement & Security
119 files · 18,497 code lines · 90 test paths · 98 JavaScript files · 13 Solidity files · 23 commits
- 01Download
solidity-settlement-security-git-history-sample.zip - 02Extract
unzip solidity-settlement-security-git-history-sample.zip -d codebase-sample - 03Verify
(cd codebase-sample && shasum -a 256 -c SHA256SUMS) - 04Clone
git clone codebase-sample/history/repository.bundle repo - 05Inspect
git -C repo log --all --graph
Static review parsed 98 of 98 JavaScript files successfully and found zero unresolved relative JavaScript or Solidity imports. The exact Hardhat dependency tree and lockfile were unavailable, so no runtime test-pass claim is made.
Construction & Sanitization
The build began with 214 mirrored repositories and 19,011 commits. Redaction policy 2026-07-18.19 removed principal identity, contributor contact details, credentials, sensitive records, private source paths, command-center material, and narrowly unsafe payloads while retaining useful technical context and authentic topology.
- Rewrite and reconcile all 19,011 commits across 575 source refs.
- Remove 13 genuinely empty repositories and 23 conservatively supported duplicate copies.
- Package 178 prerequisite-free Git bundles under stable pseudonymous IDs.
- Generate exact manifests, checksums, data cards, and public audit summaries.
- Verify a fresh extraction and submit the exact sealed bytes for independent round-eight review. The review returned PASS with zero unresolved supported blockers.
The corpus is pseudonymized, not anonymous. Safe company, product, project, and ordinary contributor-display-name context remains because it contributes to model utility. Receipt of evaluation material does not grant rights to the full corpus; executed transaction terms govern use.
Release Verification
19,011 source and target commits with zero structural mismatches.
Hashes, refs, git fsck, and git bundle verification passed.
Sanitized content and secret scans returned zero findings.
Round eight reviewed the exact policy-2026-07-18.19 sealed bytes with zero unresolved supported blockers.
Boundaries
- Pull-request counts are aggregate source-inventory metadata; the release does not claim delivery of PR bodies or review threads.
- The free sample is deliberately test-heavy and spans 23 commits over nine days; it demonstrates coherence more than long-horizon collaboration.
- Sample runtime tests were not executed because the exact declared dependency environment was not reproduced.
- Rights, licenses, provenance, and retained company context remain buyer diligence fields governed by transaction terms.