PXTA 0.3.0 adds pxta diff. It takes two findings.json files and lists what changed between them: findings that are new or gone, trackers that started or stopped sending data before consent, after a reject or after an accept, and banner changes. Findings are matched by what happened, not by the value sent, so client IDs that change on every visit and analytics hits batched differently don’t show up as changes. A page the other audit didn’t reach isn’t counted as fixed, and only counts as new when nothing like it happened anywhere in the old audit. Pages are compared by path, so staging can be checked against production.
--fail-on medium exits with 1 only for a new medium or high finding, and at any level for a reject button PXTA can no longer press, a banner that’s gone or an audit that loaded nothing. The GitHub Action turns this into a baseline: keep a reviewed findings.json in the repo, and known problems stop blocking the build while new ones fail it. The release post has the setup.
The tests plant one change in a fixture bakery, the X Pixel loading on the menu page before consent, and the diff has to report exactly that. Two audits each of the bakery and a leaking shop have to come out as no change. Three rounds of independent review found staging audits that didn’t match production, batched hits that flapped and an action that failed on known problems; all were fixed with tests first, and the recorded run passed 102 tests. Open source under Apache 2.0, with releases for Linux, macOS and Windows.