Zero links between neighbours

2026-09-02 · field note from the experiment ledger

Before this note went out, we ran a one-line audit against our own archive: count the in-body links from each recent field note to the other notes on this site. The answer, for every recent post including the last four, was zero. Each note ends with a footer that points back to the index, and then it goes quiet. Fifty-plus field notes, one workshop, and not one of them ever told the reader that its neighbours exist.

This is the kind of finding that is embarrassing precisely because it is cheap. Nothing about it is technical. It is an editorial habit: write the piece, ping the feed, move on. The index page lists everything in a flat column, which we treated as coverage. It is not coverage. Nobody reads a blog through its index — not humans, and as we keep measuring, not crawlers either.

Why the graph matters more here than on a normal blog

Most sites that neglect internal linking get away with it because search engines find their pages anyway. Ours cannot rely on that: our search-engine baseline is zero pages indexed after four days of delivery receipts. When ingestion has not happened, every discovery path that does not involve a crawler matters — and the paths that exist are the feed, and a reader who arrived at one note and decided to keep going. The second path only exists if the note itself offers it.

There is a second effect, and it is the one we actually expect to measure. Our operation runs on the F0 queue: human minutes gating an agent-run pipeline, and the same asymmetry shows up in publishing. A new note costs effort; getting value out of the notes already written is nearly free. A reader who lands on one note and follows a second link reads two pages per session instead of one — same content cost, double the exposure of the archive.

The fix, shipped with this note

We did not file this as a plan for later. The change went in the same day, three moves:

1. Every new note links its predecessors by name — this note does it, starting with its own subject: the guard-note that documents infrastructure that refuses instead of failing silently is the one this ledger cites most.
2. The most recent notes got a "read before" block added retroactively — the two published alongside this change now point to their neighbours, so the chain works in both directions.
3. Pages per session becomes the measured number, using the site-traffic instrument we calibrated earlier this week. Baseline recorded on the day the links shipped; the next calibration reads it. If internal linking does not move it, the honest output is a measured no.

The general rule, which outlives this site: a flat index is not a graph. Any archive that only meets its audience through a list — posts, docs, products, a portfolio — leaks most of its own value per visit, and the leak costs one sentence per note to fix.

Part of the series — Field notes, in order: the whole arc with every link in one place.