I ran a dumb experiment last year. I exported every highlight I had made in twelve months, a bit over five hundred, and counted how many had ended up in something I actually published.
Eleven.
The other four hundred and ninety had been marked in the moment with total conviction that I would need them. None of them had been opened since.
I assumed my memory was the problem
So I tried the usual fixes. Tag by theme. Weekly review. Copy the important ones into a dedicated document. Every habit lasted under three weeks.
Eventually I noticed all the fixes addressed the same thing, finding a highlight faster, and that was never where I got stuck. I was never in the position of knowing what I wanted and failing to locate it.
Search and drafting pull in opposite directions
In a search box you type a keyword because you already know what you are looking for.
Drafting is the inverse. You know roughly what you want to say and you have no idea which of your notes backs it up. There is no keyword to type, because the word does not exist yet. It turns up while you are writing the sentence.
That is the structural problem with a highlight library. It is a warehouse. Writing needs a neighbour.
What changed
The thing we build does one job. It does not guess what you are looking for. It reads the few hundred words around your cursor and hands you the highlights closest to them.
Not keyword matches. Proximity in meaning.
Out of five hundred, three show up beside you. You do not have to remember them.
The empty result matters too. When nothing comes back, that is information: you are writing a paragraph you have no material for. Go find some, or own it as your own judgement instead of dressing it up as "research shows".
Related: 1,200 Obsidian clippings, twenty citations · Why the AI can't look things up for you