Why is code review history useful for training AI?
It pairs a piece of code with an expert's written judgment about it. Researchers built CodeReviewer, a model for review work, from "a large-scale dataset of real-world code changes and code reviews from open-source projects in nine of the most popular programming languages", and tested it on estimating change quality, writing review comments and refining code (Li et al., 2022, arXiv). Public projects already supply that kind of data. Your private reviews cover systems, conventions and failure modes that nobody has published.
What does a code review record contain?
Each review ties a diff to comments on specific lines, the author's replies, the follow-up commits and a final approve or reject decision. The comments are often the clearest written reasoning in an engineering team: why a change is unsafe, slow or hard to maintain, and what should replace it. Linked to the issue that started it and the commits that finished it, a review shows a full task from problem to accepted fix.
What does code review history mean for what you get paid and control?
The rights questions mirror those for the code itself. GitHub's terms tell users "You own Your Content" (GitHub Terms of Service), and under US copyright law, comments an employee writes within the scope of the job are works made for hire (17 U.S.C. § 101) whose rights belong to the employer unless both sides agreed otherwise in a signed written instrument (17 U.S.C. § 201(b)). Reviews of open-source or third-party code can bring that code's licence terms with them, so check those with counsel; this is not legal advice. Every comment also carries a reviewer's name. fearofgod.ai de-identifies the data before it goes to buyers. Your company can keep ownership of its data. This is agreed case by case. Your company can be paid more than once, through renewals and repeat licences.