What is a regression? Definition in software

A regression is behavior that used to work and now doesn't — a defect introduced by a recent change that has broken something previously verified.

The term is precise on purpose. A new bug is the absence of a fix; a regression is the loss of one. The distinction matters because the response is different: a new bug goes through the usual prioritization, a regression often blocks the release that introduced it until it is fixed or reverted.

Catching them earlier

Regression tests, smoke tests, and visual diff tools all exist to catch regressions before users do. The unwritten cost of every regression that escapes to production is a small amount of trust spent — users notice that a thing they used yesterday no longer works, and confidence in the product erodes a little.

For teams reviewing a staging build before each release, a regression-focused pass is one of the cheapest investments available. Capture the broken state, name the previous behavior, and ship the comparison to whoever owns the change. The markdown artifact gives a coding agent enough context to find the offending commit and propose a revert or fix.