Screenshot in, spec out

Experiment tools ask you to describe the problem. Describe it and you sand off the details that matter: the cramped mobile pay button, the empty state that apologises. Upload the screenshot.

One screen, one aim

Upload the screen you want to move. Write one sentence for the aim; that sentence is the job of this test.

Aims that work:

• More completed checkouts on this pay step.
• More teams activating after this empty state.

"Improve onboarding" and "increase engagement" are programmes. Write the spec as a single bet.

The spec you get

You get an experiment spec.

• Hypothesis. A sentence you could put on a flag, with the CRO mechanism cited so the team knows why the treatment exists.
• UI. ASCII plus a prompt on Lite. High-fidelity mockups on Starter and up.

Edit the copy or swap the layout. Treat the spec as a starting ticket.

Keep the loop tight

You upload a screen. Growth-trained agents try treatments against known wins and you get a spec. You decide what ships.

If the spec is wrong, you spent a generate. Run it again on a tighter aim, or take the ticket to your own flag and change the treatment.

Run it on a live screen

Pick a surface that is costing you completions: the pay step or the empty state. Drop the screenshot with the aim written in one sentence. Instrument the flag from the spec and ship.