Opening Compare
Compare takes two requests — most often a failing attempt and its last known-good counterpart — and shows what's actually different between them: status, headers, timing, and body, structurally diffed rather than compared as flat text, so a reordered JSON key doesn't read as a change when the content is identical.
You'll reach it two ways: directly from an incident's "Compare With Last Successful Request," or manually after replaying a request with something changed.
The summary comes first
Above the raw diff, Compare leads with a plain-language summary of what changed. The detail is still there if you want it, but the summary is the point — most of the time it's the whole answer.
Reading the conclusion
Where the evidence supports it, Compare offers a narrow read on whether your change is what actually explains the difference in outcome. This is always hedged rather than stated as certainty, and it's explicit about the difference between "the status code changed" (measured) and "removing that header caused it" (an inference from the evidence, not a proven fact) — the same discipline Hollowport applies throughout, not just here.
Replay ×N and Compare together
If a single comparison isn't conclusive — the same change sometimes reproduces the failure and sometimes doesn't — run Replay ×N first to establish whether the outcome is actually consistent, then compare a representative pair. A one-off comparison of an inconsistent result tells you less than it looks like it does.