Reporting Flake
A powerful feature in the Happo review process is the ability to report a diff as flaky. To use this feature, click the "Report flake" button (an exclamation mark in a circle) next to a diff's reject and accept buttons, or choose "Report flake" from the "…" overflow menu at the end of that row. Once reported, both change to "Not flake" so you can undo it.
Reporting a diff as flaky will instruct Happo to ignore that specific diff now and in the future. It should be used with a little bit of caution as it might hide issues that should be caught.
If your account only lets selected users review, only those reviewers can report a diff as flaky or undo a report, since an ignored diff can approve a comparison.
Examples of where it is okay to report a diff as flaky:
- ✅ When there's a diff that you can't attribute to the changes made.
- ✅ When there are flaky rendering-related issues not caused by your code.
Examples of where you should avoid reporting a diff as flaky:
- ❌ You've made a UI update causing diffs (simply accept the report instead).
- ❌ You believe the diffs are caused by a bug in Happo (reach out to us instead).
If the underlying cause of the flake is not addressed, similar flaky diffs will eventually return once the baseline changes. After reporting flake, it is good practice to follow up with a deeper fix.
Detected flakes in new examples
An example that's new in a comparison (marked Added) has no baseline to be compared with yet. If it's flaky, you'd otherwise only find out later, from a diff in some unrelated change. So Happo renders each added example a few more times, one of them as if it were another date (iOS Safari can't change its date, so there the example is only rendered again as it was). If a render comes out different from the one in the report, the example gets a Detected flake notice. A difference counts only if a comparison would show it as a diff: your project's compare threshold and the diffs you've ignored apply here too.
- Compare renders shows the renders side by side, as a diff or with a swipe, so you can see what changed.
- When the notice says the example came out different on another date, it most likely shows the current date, or something worked out from it. See Dates and timestamps for how to fix that.
- While Happo is still rendering the example again, the notice says Checking for flakes. This doesn't hold up the comparison: you can review it in the meantime, unless approval is blocked on flaky added examples (below).
Fix the cause before accepting a flaky example, or it becomes a baseline that brings spurious diffs into later comparisons. Spurious diffs has tips for the most common causes.
To make sure nobody does, set blockApproval.flakyAddedSnapshots in your Happo config. A comparison with a flaky added example then can't be approved until you fix it and run Happo again, and approval waits while Happo is still checking the added examples. Approving through the API or Happo's MCP server is refused the same way.
The extra renders don't count toward your snapshot usage or quota: only the snapshots in your reports do. A comparison that adds many examples, such as a project's first, can take a while to check.
Global effect
Diffs reported as flaky apply to all future and past Happo reports. You may still need to manually approve or reject.
Technical details
What a diff is
A diff is a visual difference Happo found between the baseline and the new run for a specific snapshot—for example, "Button / default / Chrome". Happo compares the two rendered images using your project's deep-compare thresholds (compareThreshold and ignoreThreshold, and any other compare settings you've configured). If the difference is above that threshold, it's shown as a diff. Each diff corresponds to one component, variant, and target combination.
How Happo identifies snapshots
Happo identifies each snapshot by what it is: component name, variant, target (browser/viewport), and a hash of the image data. When you report a diff as flake, Happo remembers that specific before/after pair. So the same flake is ignored everywhere it appears--in past, current, and future comparisons--as long as it's the same baseline and new snapshot. If your baseline or your rendered output changes later, Happo may show a new diff for that same component/variant/target, and that new diff won't be covered by the flake you reported earlier.
What happens when you report a diff as flake
When you click Report flake:
-
On the comparison page
That diff is moved out of the list of diffs (in this comparison and in any other comparison where the same snapshot pair appears), so you may see fewer diffs, or none. You can still find it by clicking the Ignored count above the list, along with who reported it and when.

Each ignored diff says who ignored it: User for a flake a person reported on the comparison page, API for one reported through the API, MCP for one reported through Happo's MCP server, and Automated for one Happo ignored itself, most often because the project's deep-compare thresholds (
compareThresholdandignoreThreshold) let the diff through. Flakes reported before this was recorded show as User. You can show any of these on their own by typingignored:user,ignored:api,ignored:mcporignored:automatedin the filter box. API responses carry the same information in the ignored diff'signoredByfield. -
Build status
Your CI build status (e.g. the GitHub check) for that comparison is updated automatically if there are no longer any diffs remaining in the comparison.
-
Why flake can seem to come back
Because flake is tied to that exact before/after pair, a new run that produces a different snapshot (e.g. after a baseline update or a change in rendering) can show a new diff that isn't covered by the flake you reported. Fixing the underlying cause of the flake is the only way to prevent it from reappearing over time.