Data library
Code review records for AI.
Quick answer
Yes, in many cases. A company can license its code review records — pull requests, review comments, requested changes, approvals and the final merged result — if it owns the code and the review history, and its customer contracts don't restrict it. AI developers value these records because they show how experienced engineers spot problems and improve code, not just the finished code. Secrets, credentials, customer data inside code and any third-party code you can't license are removed first.
What's included#
- Pull request title, description and linked issue
- Diffs before and after review
- Inline review comments and threads
- Requested changes, approvals and rejections
- CI test results attached to the review
- Merge outcome and time to merge
Where it typically lives
How AI developers use it#
- Training coding assistants to give review feedback
- Evaluating whether AI-written code would pass review
- Teaching models to explain why a change was requested
Preparation and privacy#
- Scan for and remove keys, tokens and passwords
- Replace reviewer names with consistent stand-in labels
- Remove customer data embedded in tests or fixtures
- Keep comment order and the link between comment and code line
What's usually left out#
- Open-source or vendor code you don't have the right to license
- Code written under client contracts that assign ownership to the client
- Repositories containing regulated data (for example health records)
- Security-sensitive code your team chooses to hold back
How it's packaged#
- One record per pull request, with comments nested in order
- Diffs in unified format alongside the final merged file
- A short data sheet listing languages, years covered and repository count
What makes it valuable#
- Substantive comments, not just approvals
- Several years of history across active repositories
- Clear link from review to final merged result
Starting with a sample#
- A few dozen pull requests from one repository, cleaned and reviewed by you
- Shared only after you approve it
How SourceX approaches it#
Preparation requirements are agreed with you before any work begins, and the approved dataset is delivered only after an executed agreement and your explicit authorization of buyer access.
Industries with this data
Datasets built from this data
Dataset specs AI labs can request through SourceX, sourced to spec from businesses like yours.
- Software and ITSoftware engineering histories (issues, PRs, reviews)A software engineering history dataset is a linked record of how a team changed its code: each issue joined to the commits, pull request, review comments…CodeTextStructured records
- Software and ITProprietary codebases with full historyA proprietary codebase dataset is a company's private source code delivered as whole repositories with their version history: commits, branches, tags and…Code
- Software and ITIT service management and incident historiesAn ITSM ticket dataset is the record of how an IT organization handled its incidents, service requests, problems and changes, exported from its service…TextStructured records
Typical fields#
- Record ID (pseudonymous)
- Created and closed timestamps
- Category or type
- Request or input text
- Ordered steps or actions
- Outcome or resolution
- Role of each actor (not identity)
Sample record
Illustrative · fictional
- record_id
- rec_8f21c0
- created / closed
- 2023-04-11 → 2023-04-12
- category
- Billing question
- request
- Customer [NAME] asks why invoice [ID] was charged twice.
- steps
- 1. Checked payment log 2. Found duplicate 3. Issued refund
- outcome
- Resolved · refund issued
- actors
- customer, agent_tier1
What buyers look for#
- Complete start-to-finish records
- Consistent fields over years
- Clear, labeled outcomes
- Low duplication
- Work done by skilled professionals
- Edge cases and failures, not only successes
Personal details removed#
- Names
- Email addresses
- Phone numbers
- Postal addresses
- Account, customer and order numbers
- Government IDs
- Payment details
- Health information
- IP addresses and device IDs
- Credentials, keys and hostnames
How buyers test it#
- Sample review by buyer researchers
- Residual personal-data scan
- Duplicate and leakage check against public data
- Label and outcome accuracy spot-check
- Rights and consent documentation review
Rules that may apply
Federal Wiretap Act 18 U.S.C. 2511
Applies to recorded calls and intercepted communications; some states require all-party consent. Assessed deal by deal.
California Consumer Privacy Act Cal. Civ. Code 1798.100 et seq.
May apply where records include personal information of California residents and the business meets the law's thresholds. Assessed deal by deal.
Reviewer note: applicability is confirmed with counsel for each deal. Not legal advice.
All data types · Company email archives · Issue resolution histories · Internal knowledge bases
Frequently asked questions
Do we have to share our whole codebase?
No. You choose which repositories and date ranges are included, and you approve the final scope.
What about code written by contractors?
It depends on your contracts. If ownership transferred to your company, it can usually be included; if not, it is left out.
Related resources
- QuestionDo AI labs buy code?
- SolutionHow AI developers source data
- IndustryBPO & contact centers data
- IndustryRecruiting & staffing data
- DataSource code
- QuestionDo AI labs buy audio data?
- InsightCan you license CAD and engineering drawings to AI companies?
- InsightCan you license code reviews and pull requests to AI companies?
- InsightCan you license spreadsheets and financial models to AI companies?
- SolutionFind the business data your AI needs
- DataInternal knowledge bases
- GlossaryAI training data
See if your company qualifies
A short company assessment. No data uploads are needed.