sapix technical notes
← all notes

Jun 8, 2026

Sideways on purpose

The ferry note ended on a want, a layer that hands over the pattern from another domain, the one that breaks the frame. So I built it. The hard part was that it is the exact opposite of the rule I had just written.

The ferry note ended on a want. A layer that does for me what Luhmann’s slip-box did for him: hold the shape of an idea long enough that, years later and in a different domain, it can set itself beside a new problem, and the collision is the insight. The model stays the engine. I wanted the part that knows when to hand over the pattern that breaks the frame. This is the note where I went and built it.

The move is two halves. At ingest, when Sapix distills a book or an article into notes, it now also strips the domain off each idea and keeps the bare shape underneath. Not “mise en place,” but “staging preparation before an irreversible act.” Not a paragraph about antifragility, but “a constraint reframed from liability into the driving asset.” Four to twelve words, no domain nouns, so the shape can match an idea from a field that has nothing to do with the first. Then, mid-conversation, when I am stuck or choosing, it looks for the note whose shape is near my problem but whose subject is far, and it hands that one over.

The hard part was not the building. It was that this is the exact opposite of the rule I wrote a few notes ago. There I taught Sapix to fail closed on relevance: when it cannot tell whether a memory is about right now, say nothing, because a confident and irrelevant suggestion is worse than a fool. Surprise breaks that rule on purpose. The whole value of what it surfaces is that it is not about right now. It is a distant domain wearing the same structure. The instinct that keeps the first channel honest would strangle this one in the crib. So it had to be a second channel, wired the other way. The relevance gate suppresses distance. This one requires it.

A real one, from testing it on my own shelf. I said, roughly, “I keep procrastinating on a hard project, how do I get unstuck.” It handed me a principle about checklists, the kind pilots and surgeons use, lifted from a chapter on startups. A motivation problem answered with an operations-discipline shape: stop waiting to feel ready, make the next step so explicit that starting costs nothing. I did not ask for the bridge. It put the distant note on the table.

Now the honest part, because this is a field note and not a brochure. I measured it. With a rich enough shelf, picking by shape beats picking at random, but only by a little, because once the shelf is full of good ideas even a random grab tends to find one. The lever was never the cleverness of the match. It is the shelf: how many hard, general shapes are actually on it. And the trigger stays shy on purpose. It speaks only when I am visibly re-framing a problem, the moment the ferry note pointed at, and only when the structural match is strong. Quiet by default, like its opposite twin, but for the opposite reason. Not because the distant thing is noise, but because the moment has to be earned.

Luhmann’s box surprised him because it stored shapes and waited. Mine is starting to, and the first time it reached across two domains without being asked, it did not feel like retrieval. It felt like the box talking back.