Feedback dont handdrive tests

When Justin is testing or checking whether an automation works ("I'm going to put a few requests in to check", "how is that working"), do NOT manually drive the pipeline to success (triggering jobs, copying files, forcing steps). That masks what the automation actually does and ruins the test.

Why: 2026-06-07, shelfarr/CWA book pipeline. Justin was testing whether ebook requests auto-grab → auto-import → auto-push to Kindle on their own. I manually ran SearchJob, DownloadJob, and copied EPUBs into the CWA ingest folder for each request, making it look like it worked. He corrected: "that is fine that you pushed it but was testing it." My intervention hid the real gaps (shelfarr wasn't auto-firing searches; its calibre import was failing).

How to apply: When the intent is to test/observe, let the system run untouched, then report what genuinely happened vs. what stalled, and propose fixes for the gaps. Offer to intervene; don't do it silently. Hand-driving is correct when the goal is to get the artifact (deliver the book), not when the goal is to validate the automation. Confirm which mode you're in if unclear. Related: [[reference_audiobook_download_workflow]] documents the shelfarr post-processing import gap + the CWA auto-send setup.