Experiment 40
Does my second brain pay rent
I save a lot: videos, articles, voice memos, all summarized and searchable. The uncomfortable question is whether any of it ever comes back, or whether I built an expensive shelf.
I went looking for evidence in 56 working sessions, arranged on a ladder from weakest proof to strongest. Weakest rung: a note merely exists in the system (671 of those). Stronger: a note was actually opened during real work. Strongest: a note was explicitly cited as the source of something I built.
The answer was yes, with a twist. Knowledge does come back: 45 sessions reached into old notes by choice, and 26 of those fed directly into real work outside the vault. But all of that evidence arrived through a logging channel I never designed for the purpose, a side effect of another tool recording which files got read. The channel I had designed as proof, citations in my end-of-session write-ups, held exactly one citation across all 56. The ritual asked what I produced and never asked what I consulted, so nobody wrote it down.
One line added to the ritual template fixed it. The lesson: if you want to know whether a loop closes, you have to instrument it, because the evidence will not volunteer itself.
Rungs and instruments
Two corrections drive every number. Counting at the observation level undercounts: a session reads a note in one observation and edits code in another, and at that level 49 of 55 read-and-modified events are the vault housekeeping itself. And ritual reads had to be separated from elective ones: ytk sessions are instructed to load wiki and project context at start, so the signal that means captured knowledge was wanted is specifically a read under sources. Pooling ritual with elective inflates reuse roughly threefold, 125 sessions against 45.
The full ladder, weakest claim to strongest: 671 notes embedded; 153 read inside sessions at all; 57 elective source-note reads; 55 served to a genuine search in a 15-day window; 1 cited in a session brief. Across all 56 briefs there are 2 genuine wikilinks, both to the same LinkedIn draft. The rung the design named as strongest evidence is the emptiest, because briefs describe work produced and nothing in the ritual asks what was consulted.
The calendar view adds the uncomfortable timestamp: the 45 elective-read sessions concentrate in the last five weeks, coinciding with the study and library features and heavier agent use. The loop closing is a recent habit, not a steady property of the system. And every bit of this evidence arrived through an accidental instrument: another tool's log of which files got read.
Consequences for the event vocabulary: the citation rung stays empty unless the brief ritual asks one question it never asked, a one-line template change that converts the emptiest rung into the strongest at zero cost; and elective-versus-ritual has to be a first-class dimension of any read event, or the threefold inflation comes back.