Plain-text research logs: what is actually worth keeping?

My research log used to be a pile of command output. Now each entry starts with a question, a local environment identifier, and the result I expected. The transcript comes after that.

Unexpected output gets a note before I change anything. Otherwise the record becomes a story about what eventually worked, and the failed assumptions disappear.

I use plain text because I can search it years later. No special app required. What is your minimum useful entry?

A link back to the exact input artifact, ideally with a checksum. “The file from yesterday” is not a durable identifier.

And redact before sharing. Logs can contain usernames, internal paths, and tokens even when the experiment itself is harmless.