Skip to main content
Three quiet failure modes account for most tests that “did not work” even though the code was right. The results page checks for all three and shows a health chip next to any goal that qualifies.

​
Silent

Chip: “Silent, no fires yet.” The goal has been defined for the test but has never fired on any variant. Almost always one of:
  • Selector typo. The click goal targets .hero_cta but the button is .hero-cta. The preview modal’s QA overlay shows which selectors matched what on the current page.
  • The goal-firing page is not covered by the experiment’s Location. A conversion goal for /checkout on an experiment that only runs on /pricing looks silent. Widen the Location or move the goal.
  • The goal really has not fired yet. For a low-frequency conversion on a low-traffic page, silence is normal for the first few days.
The chip stays until the first fire. It does not auto-hide.

​
One-sided

Chip: “Firing on one arm only.” Every fire so far has been on the same variant. Two explanations:
  • The selector matches something the other variant hid or replaced. Common for goals defined against elements the variant deletes or moves. Adjust the goal to match a stable element, or add a variant-independent hook.
  • A real, huge effect on the goal. Very early in an experiment, one-sided fires can be a genuine early signal, but if you are more than a hundred fires in and every one is on one arm, the more-likely cause is the first one.

​
Insufficient data

Chip: “Not enough visitors yet.” The goal has fired, but not on enough visitors for any engine to say anything meaningful. This is the honest version of “still gathering” at the goal level. See the confidence gate for the minimum-visitors threshold and what happens when a test falls below it.

​
Primary-goal-specific chips

The primary goal has two additional warnings:
  • “Primary goal silent.” No primary conversions yet on either arm. The test cannot possibly declare a winner.
  • “Primary goal firing on one variant only.” Almost certainly a selector or activation problem, not a huge effect.
Both block the “winner” verdict until they resolve, the page will not call the test if the primary looks broken.

​
When to act

  • Chip after a fresh publish, within the first hour or two: wait. Cache propagation, sample size building up, all of the above.
  • Chip that persists past a day of normal traffic: investigate. Open the preview modal, verify the goal fires when you click the target, and adjust.
The goal health chips are the difference between shipping a broken test to production and catching it before it costs you anything.