Is the universe becoming a graph?

An isolated caliber is one this corpus records no relative for — never one that has none. 9,750 of 12,269 calibers carry no lineage link here. That is what the sources say, not what horology did: the archive most of this corpus is drawn from records a family block on a minority of its pages. No caption, filter or count on this site may state the second thing, and the falling share above is a statement about our coverage, not about watchmaking.

Lineage edges are counted in every lane and at every status. The number here is the same one the landing universe draws — all 1,695 of them, including the inbox staging lane. This paragraph used to continue “a far smaller set has been reviewed and accepted”, and pointed at coverage as the answer to “what anyone has checked”. That question was retired on 2026-08-29: nobody was ever going to review 154,620 claims, so the answer would have stayed “almost none” forever while the corpus said something else. What is left is one question, and it is the one that was always doing the work — what does the corpus record, and what is behind each record? Every edge counted here carries a source url, a retrieval date and a verbatim quote, and clicking any of them shows all three.

The staging lane is a place, not a caveat. A claim in inbox/ has not been through a promotion step into corpus/, and gate G3 plus a schema CHECK make it structurally impossible for one to assert anything else — that guard is what allows an unattended pipeline to run at all, and it did not move. What changed is that the site stopped DRAWING the lane: staged edges are no longer dotted, staged values no longer hatched, and the default family profile no longer hides them. The effect on this page is none at all, because both metrics above were always lane-blind; the effect on the universe and the family pages was to stop showing a fifth of what the corpus holds.

3 of these 34 rows are archaeology, not a nightly log. Nothing was running to record history, so the earlier points were reconstructed from static export trees this repository still had on disk — each one carrying its own corpus_sha, its own compile timestamp and its own counts. Where an older payload predates the field that names the connected count, the number was taken from the edge list it ships instead, which is the same measurement and is asserted to be so on every tree that carries both. Rows are labelled with the tree they came from. Nothing was interpolated, smoothed or invented, and there is no point here for a day the corpus was not compiled.

A row is a corpus state, not a night. The x axis is evenly spaced because rounds are, and clock time is not: three of the first four states were compiled inside thirty-six hours. The schedule exists but is installed disabled — the owner arms it — so until it is armed the spacing is still rounds, and reading the slope as a rate would be reading a ratio of two things that were never rates.

A night that adds nothing is a normal night, not a broken one. Most of what the loop does is confirm that eighteen catalogues still say what they said yesterday, which costs a few dozen conditional requests and produces no rows at all. The block above reports a zero the same way it reports a thousand, because a pipeline that only appears in this page when it has news is a pipeline whose silence you cannot read.

Nothing here relaxes a gate. The ledger is written after make lint passes, so a corpus that fails the integrity gates never reaches this page at all — it aborts the round instead. Growth that arrived by lowering the standard would not show up as growth; it would show up as a build that stopped.

The real progress bar is the picture. Every new lineage edge lifts a caliber out of the reservoir and into a family, so the universe itself densifies as these numbers move. This page exists to say by how much, and in which direction, on a morning when the picture looks about the same as it did yesterday.