Reviews
Rounds of looking at the app as it is, from pictures. The app is written by agents that cannot run it, so CI photographs every screen from invented sample data, and each round here is somebody reading those pictures: first the agent that made them, then a panel of synthetic users, each with a purpose, a level of knowledge and only the screens they would see.
A finding is where a user's expectation and the next picture disagree, or where a user stopped. Each finding names the screen, who found it, the session that owns that screen, and a status:
| Status | Meaning |
|---|---|
| open | Written down, with an issue under issues/open/ if it is worth fixing |
| fixed | Changed in the app, and seen to be changed in a later round's pictures |
| not a fault | An artefact of how the pictures are made, or a decision that stands. It says which |
A synthetic user reads a picture. It cannot feel a control that is too small, hear nothing, or drive a car, and it has been wrong about what a tap would do. The owner on a device remains the test that counts; a round is a cheap way to find what to look at.
Rounds
| Round | Date | Pictures | Findings |
|---|---|---|---|
| Round 2 | 14 November 2026 | 33 pictures per set, at commit 4338ea2 on work/screens (dev up to 6475ff7) | Of round 1's 15 open findings: 3 fixed, 6 fixed in part, 6 unchanged. Two users changed their minds. 12 new findings; issues 039 to 041 |
| Round 1 | 13 November 2026 | 28 pictures per set, iPhone and iPad, light and dark, at commit fb0c548 on work/screens | 21, of which 15 open and 6 not a fault; ten issues opened (029 to 038) |