Ship what you meant.
Tell us the task that matters and how you’d know it worked. We’ll review it from the perspectives it was built for, and send you every screen, marked.
Three marks on the Team screen. The confirmation disappears after 3 seconds. The confirmation isn’t announced (WCAG 4.1.3). The invitation was sent twice.
- Gone after 3 secondsUsability · attempt 2, step 4
- Never announcedAccessibility · set by rule
- Sent twiceUsability · attempt 2, step 6
Every product is designed once. twice.
Once in the design file. Then again by everything that ships after it: the edge case, the copy change, the pull request nobody walked through. People use the second version, and the wrong turns they take there are the evidence.
- Opened Team
- Typed sam@demo-target.test
- Pressed Send invitation
- “Invitation sent” showed, then vanished after 3 seconds4.1.3No status role, so nothing was announced
- Looked for a confirmation. None on screen
- Pressed Send invitation againRetry
- Two invitations waiting for sam@demo-target.test
1 retry · 1 duplicate · completed, at a cost
- What you meant
- Sam gets the invitation, and the admin knows it was sent.
- What they met
- Two invitations went out, and the admin never saw the first one land.Attempt 2 as New administrator
The same screens, read by the whole team.
Five questions every product team asks about a task, answered on one review.
Product design
Is this what we designed?
Write the task in two sentences: what someone is trying to do, and how you’d know it worked. Every screen is held to it.
UX research
Whose experience are we checking?
Name the perspectives that matter. Each works through the task once. It isn’t a person; it shows your next study where to look.
Usability and accessibility
Where did it get harder, and who couldn’t get through?
Wrong turns, retries and dead ends are kept as evidence, with each step’s screen, accessibility tree and what a screen reader would announce. AI proposes issues; rules settle what a rule can.
Design engineering
What exactly do I change?
A person decides each issue. The ones you act on become a fix brief for Claude Code, Codex or Cursor, with the screens attached.
Product
Did it land, and will it stay?
A recheck runs the same task, looks for what you intended, and records whether it was seen, not seen, or inconclusive. Recordings replay on a schedule, so you’d see it if the next release undid it.
- Intended
- The admin knows the invitation was sent.
- Rechecked
- Seen: at step 6 the confirmation stays, and it’s announced.
When it’s worth a look.
A signup, onboarding or invitation change is about to ship.
Review it on staging, before customers meet it.
Ask for a reviewYou ship every day, often with a coding agent.
Review the tasks that matter after each release, and let the replays show what changed.
How it worksSupport keeps hearing about one screen.
Name the task and who’s struggling, and see the screen where it gets hard.
Read an example reviewA university sent an accessibility questionnaire.
Check your public pages free, then publish a conformance report where every row links its evidence.
Accessibility and the report
Some problems slow people down. Some stop them.
A button no keyboard can reach. A confirmation a screen reader never hears. Variantly finds these on the same steps as everything else, and your team decides them the same way.
Asked for a VPAT? The same captures set a conformance report where every row links its evidence. How the report works
What each kind of evidence can tell you.
Variantly sits beside your research, analytics and tests.
| Evidence | It shows you | It can’t show you |
|---|---|---|
| Research with people | Why people struggle, in their words | What changed in this week’s release |
| Product analytics | How many reached each step, and where they left | What was on the screen when they left |
| End-to-end tests | That the task still runs | Whether a person could tell what happened |
| Variantly | Every screen of a task from a named perspective, what got in the way, and a recheck | How many people would hit it. “Not sure yet” sends that to research |
Built to be checked.
A person decides every issue.
AI proposes. Nothing closes on its own.
Every issue keeps its screens.
And the steps that reached them, so anyone can disagree.
An attempt isn’t a person.
It shows where to look. Your research says how people feel.
Gaps are listed.
Every review says what it didn’t cover, so silence isn’t a pass.
Priced per product. No per-seat fee.
Everyone at your company signs in. Plans count attempts: one perspective working through one task, once. Compare plans
Starter
$100/month
$100 a month, billed yearly at $1,200
Task reviews with an included allowance of attempts, nightly replays, and the conformance report when billed yearly.
Team
$250/month
$250 a month, billed yearly at $3,000
Everything in Starter for up to five products, with a larger shared allowance.
Report only
$800 once
Billed once
A conformance report without a subscription: scoped and dated, with monthly rechecks for 12 months.
Plans open when accounts do. Until then, ask for a review of one task