Book & Article Distiller

Summaries compress. Distillation selects. A summary tries to represent the whole text in fewer words; a distillation asks what from this source would change how you think or act, then keeps only that. The difference matters because you can't link to a summary in a knowledge system. "Chapter 3 discussed feedback loops" is dead on arrival. "Feedback delays longer than 72 hours lose 80% of their corrective power" is a node you can connect to management notes, learning theory, deployment practice, and parenting.

Every claim runs four filters before it earns a card. Originality — is this genuinely new, or a platitude restated? "Focus on one thing" doesn't survive. Actionability — could someone change their behaviour based on it? Connective — does it link across domains, or does it only describe the author's company? Surprise — does it challenge something you'd otherwise assume? Ideas that confirm what you already believe are comfortable and low-value. Ideas that make you stop reading and reconsider — those are the ones worth a permanent home.

Each idea becomes a card with the claim stated in your understanding (not the author's words), the evidence type named honestly (RCT, case study, or "personal experience with no cited data"), and suggested connections to other notes. The output also includes a relationship map — which ideas build on each other, which are independent, which are in quiet tension the author didn't notice — and a section called what the author gets wrong, because extraction without evaluation is just transcription.

The skill is honest when a source is thin. If a 300-page book yields one idea after filtering, you get one card and a verdict explaining why. If it yields zero — all platitudes, all restatements of older work, all unsupported claims — the output says so directly and names the specific failure mode. That honesty is part of the deliverable. A second brain full of dutifully extracted nothing is worse than no extraction at all.