Low confidence analysis

Why this happens

The page didn't fully load, so Levri reduced confidence instead of guessing. Rather than fabricate scores from incomplete signal, the report is delivered with a flag that tells you to read the numbers as directional, not authoritative.

A low-confidence report is still useful — copy weaknesses, broken hierarchy, missing trust elements all surface. Expect the lift estimates and pillar scores to shift if you re-run on a cleaner version of the same page.

Common reasons

• The page didn't fully render before timeout
• Critical content reveals after a user interaction (tabs, accordions, scroll)
• An A/B test variant was served and Levri saw a partial fold
• Major sections couldn't be reliably identified

Improve it

1. 01
   Retry

   Ensure full load. A second fetch often hits a warmer cache and renders the full page.
2. 02
   Avoid dynamic pages

   Use static content where possible. Marketing and product pages render more reliably than dashboards or account pages.

When to ignore the flag

If the experiments listed in your report match what you already suspected — weak hero, hidden CTA, missing trust signals — the low-confidence flag doesn't change what to ship first. It matters most when you're comparing two pages or baselining a score over time, where you want full-confidence runs on both sides for the comparison to be fair.