What this prototype argues
A tester has two jobs on two screens: find what is waiting, then work through one scenario. The queue is one design. The variants change only the runner, because that is where the tester spends the time.
- Accepted on 2026-10-10: variant B. The rejected layouts stay here, switchable from the bar at the foot.
- A · Steps beside the detail. The steps are a list on the left and the chosen step fills the right. The tester sees where they are in the whole scenario and the whole step at once.
- B · One step at a time (accepted). The tester runs the scenario in one browser window and the application in another. A narrow window shows one step, a progress bar and the two verdict buttons. A Pass moves on. A Fail stays, so the tester can write the note first.
- C · The document. The scenario reads top to bottom like the org page, with the controls under each step. It suits a finished scenario that someone reopens to read, and a reviewer who wants every note on one page.
- The queue is the front door. A notification in the bell and a row in the queue open the same scenario. The bell here is mock. No notification kind exists for a waiting test.
- The overall result is inferred, never set. Any FAIL makes the scenario FAILED. All PASS makes it PASSED. Anything else leaves it PENDING. The tester marks steps, not the scenario.
- Story and Task sit beside the steps. They are the docs that drive the scenario, so the tester need not leave to recover the context. The links in them open the linked doc.
- A finished scenario reopens read-only. A Retest button clears the steps and starts again. The build must decide whether a retest overwrites the result in the org file, where git keeps the history, or keeps each run.
- Nothing reads or writes a scenario yet. The BFF has no scenario store. ores.web cannot rely on compass, so the BFF parses the docs that compass creates and writes the results into them itself, behind an interface that a database store can replace later.
- The queue has no owner. A scenario names a target dialog and an environment and no tester. The queue shows every scenario in state PENDING. Assignment comes with the tester role.
- A browser cannot capture the screen. The step takes a pasted or dropped image. The BFF has no upload route that saves it beside the scenario doc.
- The web package renders no org text. Step instructions are org. A shared package parses it for the BFF and the web package.
- A deployed server has no work tree. The module reads and writes the checkout the BFF runs in. Whether it runs in production is open.