Base and head: the branch a pull request targets, and the change itself.

01Definition

Two versions of the same codebase.

Every pull request has a base, the branch it wants to merge into (often main), and a head, the branch carrying the proposed change. The diff a reviewer reads is the difference between them.

Testing only the head answers “do the tests pass with this change?”. Testing both answers a better question: “what did this change do to the tests?”

• Base: the target branch
• Head: the change
• The diff is between them

02Why compare

A failure that was already there is not this change's fault.

If a test fails on head, it matters whether it also failed on base. A test that passed before and fails after is a regression caused by the change. A test that fails on both was already broken. A test that passed on base and no longer exists on head may have been deleted to make the run green.

Comparing the two runs turns a single pass or fail into a statement about the change itself.

• Regressions are pass → fail
• Pre-existing failures are not regressions
• Deleted tests show up

03In Verify

Built: the same tests on base and head.

Verify checks out base and head, runs the project's own install and tests on both, and compares them. When a change breaks a test that passed before, the verdict is reject and the comment names the broken tests. When a change edits existing tests, Verify also runs the base's version of those tests against the change.

Built and running on TraceAxiom's own pull requests; not installable yet.