Beta. The idea board is part of the Signals Beta. See the Signals
overview.
Ranked by a PXL score
Each idea carries a PXL score: one point for each of the plain questions the scan credited, and two for each evidence question it credited. Higher scores rise to the top. Ideas that tie are ordered by how much evidence backs them, and then by effort, lowest first. You do not have to read the whole list to find the good ones. Start at the top. The score is read against that scan’s own maximum, not a fixed scale. A scan with qualifying traffic and with GA4, the heatmap and customer voice all contributing reaches 12 points; a scan with less evidence reaches fewer, because the questions the missing source would have answered are not scorable on it. So one idea reads 7 of 12 and another reads 7 of 8, and both are honest. The arithmetic is in the evidence contract. ABTestly computes that score, and the ranking that follows from it. The AI answers eight yes or no questions about each idea, each with a short rationale. It does not hand back a score, a tier or a place on the list.Test ready, Needs evidence, Below threshold, or Unverified
Every idea the scan found wears one of four badges. This is the honesty framing from the overview, shown per idea. An idea you added by hand is the exception: it is labeled Your idea and carries no PXL score and no gate, because nothing scored it.
A Test ready badge is a reason to build the test, not a claim that it
wins. The threshold each badge is measured against is set by the
idea’s evidence tier, which says what the scan credited on that
one idea rather than how your data reached the scan, so two ideas on
the same scan can be held to different bars. A source can appear as
Used on the scan’s evidence receipt and be credited on no idea at
all. There are four tiers, and the exact number per tier is in the
evidence contract. When the data to back
an idea is not there, the idea says so rather than inventing a number.
Grouped into campaigns
Related ideas are grouped into campaigns, so a set of changes aimed at the same page or the same goal reads as one theme rather than a flat list. A campaign is a useful unit to plan around: approve the ones you believe in, dismiss the rest, and you have a quarter of testing roughed out in a few minutes.Triage: New, Approve, Dismiss
Every idea sits in one of three states. Triage is how you turn a raw scan into a plan.1
New
Fresh from the scan and not yet triaged. Your inbox of ideas to
look at.
2
Approve
You want to run this. Approving marks it as a keeper and sets it up
for Build this.
3
Dismiss
Not for you right now. Dismissing clears it off your active board
so the list stays focused on what you intend to run.
Add your own ideas
You are not limited to what the scan found. Add your own idea by hand, with your own hypothesis, and it lives on the same board, carries the same New, Approve and Dismiss states, and flows through the same Build this button. The AI does not have a monopoly on good tests, and your own knowledge of the site belongs on the board next to the machine’s. A hand added idea is labeled Your idea, and it carries no PXL score, no evidence tier and no readiness badge. Nothing scored it, so Signals claims nothing about it, and it never reads as Unverified either: that badge is about scores Signals cannot vouch for, and this idea has none.Search the board
When a board gets long, especially after several scans and a few funnel walks, search filters it by keyword so you can find the idea about the checkout button, or every idea mentioning the pricing page, without scrolling.Next
Build a test from an idea
Turn an approved idea into a real experiment draft.
Connect Google Analytics
One click, read only. Lets a scan read your funnel data and credit
it on the ideas it speaks to.