How compunct videos are made

Every compunct video is AI-generated. The script is written by a language model, the narration is a synthetic voice, and the imagery is generated. We say this plainly because the interesting part is not that AI made the video — it's what stops the AI from making things up.

The rule: researched or clearly not factual

Videos about the real world — current events, news in a topic you follow, anything with checkable facts — go through a research pipeline before a single line of script is written:

Retrieval. The system searches for and reads real, current sources on the topic.
Evidence pack. What it found is distilled into a set of sourced statements — the only raw material the scriptwriter is given.
Grounded script. The script is written against that evidence, not from the model's memory.
Claim gate. A separate check compares every factual claim in the finished script back to the evidence. Claims the sources don't support are removed or the video isn't made.
Visible sources. The sources used are listed on the video's page, so you can check our work.

Videos that aren't about checkable facts — explainers of well-established ideas, ambient pieces, things you asked for directly — skip the research step, and don't carry sources, which is itself the signal: if a compunct video makes claims about current events, it shows its sources.

Mistakes and corrections

This system reduces invented facts; it doesn't make errors impossible — a source can be wrong, and a check can miss. If you find a factual error in a compunct video, email curtis@compunct.com and a human (see about) will look at it.

Privacy of personal videos

Videos are made for individual people and are private by default. They are excluded from search engines, and a video someone chooses to share publicly never reveals why it was made for them.