Diagnose, drill, prove it changed
One cycle takes about two weeks. Then it repeats, with a roadmap built from what the last one revealed.
- 1
Have a real conversation
You chat naturally for a few minutes — work, food, weekend plans, whatever comes up. There is no test and no score, and the app never corrects you mid-conversation. Correcting people while they talk makes them talk less, and the whole point is to hear how you sound when you aren't being careful.
- 2
Get your habits named
Afterwards the transcript is analysed into named patterns rather than a list of typos — “uses present tense for past events”, “drops the before abstract nouns”, “fills every gap with um”. Each one carries a status, how hard it is to fix, how often it occurred, and real examples in your own words. If you told us your first language, the analysis also anticipates the interference it tends to cause.
- 3
Follow one roadmap
You pick what to work on, and the app builds a single tree sized to roughly two weeks at your chosen cadence. Patterns that occur more often get more stops. The easiest fixes come first, so you feel movement early. Each stop mixes written and spoken formats that ladder from recognising the right form to producing it under pressure.
- 4
Prove it changed
Clearing every stop at 100% unlocks a check-in conversation. That transcript is compared against the one that started the round, and you get a report split three ways: what improved, what still needs reps, and what was newly spotted. Then the next roadmap is built from that.
Why it's gated this way
There is no free-form chat and no way to skip ahead. Conversations exist at the two ends of a cycle so that each one is a measurement of something — the first diagnoses, the last one checks. If you could talk whenever you liked, neither transcript would mean anything, and neither would the report.