About

Sleepytime Facts is made by Doug Eubanks in North Carolina.

I'm the IT manager at a software development company, where I spend my days on security and chasing cybercrime. It's work that rewards patience and a methodical streak. The same habits turn up here, in a show about staying calm.

I'm a North Carolina native, born in Durham and raised in a rural stretch east of Raleigh. I live here still, with my wife Laura and our two cats, Melon and Julianna, who supervise late-night projects closely and without much sympathy for deadlines. Away from work I push home automation gear until it complains, and I'm an active ham radio operator as K1DUG, a hobby that is mostly quiet technical problem-solving and talking to strangers a long way off. I like museums, volunteering, and learning things for no particular reason.

This started with something I noticed in myself and then in nearly everyone I asked: at night, the thinking doesn't stop on request. I wanted to make something plain and kind for that hour. True things, read gently, taking up the space between awake and asleep. Not perfect. Comfortable. If it gets you to sleep a little sooner, it's working.

How the episodes are made

AI is part of how this is written, and you deserve a straight answer about that rather than a line of small print at the bottom.

I don't write the sentences, so calling myself the author would be overselling it. What I do is choose what each episode is about, set the rules it has to follow, and decide whether it's good enough to publish.

Subjects are chosen by hand, one at a time. They're usually things I've wondered about and had to go and read up on. Nothing here comes off a trending-topics list or gets queued in bulk to fill a schedule.

Each episode is drafted by one model, then taken apart by several others whose only job is to find what's wrong with it. Using different models for that matters more than it sounds: a model asked to review its own writing will nearly always tell you it did fine.

The checking starts before a word of the episode exists. The outline has to hold up first. Every chapter has to be a genuinely separate subject, with nothing repeated to fill time and nothing that can't be described calmly. Subjects that fail that last rule don't become episodes, however interesting they are. A famous logic puzzle got cut from one episode because it can't be explained without an execution in it.

Then every chapter is read again, several times over, by models looking for different things. One checks facts: dates, figures, names, and any claim stated more confidently than its evidence allows. Another reads only for tone, because an episode meant for falling asleep has no business raising your pulse. No peril, no stakes, nothing with an edge to it. Others look for the tells of machine writing, for a chapter that has started talking about itself, for one that's padded, and for anything that would sound wrong read aloud.

Nothing is published on the strength of a first draft. A chapter that fails any of those checks is written again rather than smoothed over, and some go three rounds before they pass. One chapter covers one idea; nothing is stretched to reach a running time.

I listen to the finished episode before it goes out.

The narration is synthetic. One voice, kept slow on purpose. The same text becomes the audio, the transcript, and the ebook, so what you hear and what you read are never different.

None of this makes it flawless. The checking exists because the drafts genuinely do get things wrong: an anecdote that never happened, a date off by a century, a confident number with nothing behind it. Catching those is most of the work, and I'd rather say so than imply a machine wrote something perfect on the first pass. If you catch something I didn't, tell me and I'll fix it.