Every number in this series came from one place: a commit graph, read by the Excavator. A commit graph is an extraordinary record — dated, attributed, kept in full. It is also a particular record, with edges. Measurement is itself a kind of labor, and like all labor it leaves things out.
The survival ratio — the spine of essay No. 01 — divides the lines still standing by the lines ever inserted. It should never exceed one. Across the seventy-one areas the Excavator measured, thirty-seven of them report survival above one hundred percent.
why the needle pinsWhere the count loses its meaning
It happens wherever lines arrive at HEAD by a route the insertion count cannot see. When seventeen audio files are relocated into the public tree in a single commit, or a page is generated and rebuilt from source, the lines are present but were never "inserted" inside that area's window. The instrument divides survivors by a number that is too small, and the needle pins past one.
The figure is not a survival rate any more. It is the sound of a tool reaching the edge of its own definition — the place where a line was written here stops being a question the commit graph can answer.
a full accountingWhat the graph keeps, what it never had
Set the two columns side by side. The left is everything these essays could stand on. The right is everything they could not.
Recovered from the graph
- when each line was written, and whether it survived
- which files moved together, and which were renamed
- the days work landed, and the days it didn't
- what was begun, committed, and later deleted
Never recorded at all
- why any line was written the way it was
- the idea abandoned before it was ever committed
- the hours of thinking between two commits
- which changes were painful and which were easy
The graph is a cast of the labor, not the labor. It keeps the shape and loses the heat.
three places it went blindFailures, kept on the record
Essay No. 01's bibliography records 1,377 insertions and zero deletions — yet only 720 lines survive. A generated rebuild rewrote the rest without a single deletion ever reaching the log. The dig can count deletions. It cannot see a regeneration that quietly supersedes.
This very analysis first ran against a shallow clone. With the deep past missing, CLAUDE.md's origin resolved to the repository's squashed import — a commit whose subject was a stray merge line — and the tool nearly quoted that merge as the file's first authored words. Full history corrected it. The dig is only as deep as the history it is handed; a truncated past yields a confident, wrong origin.
The workhorse of essay No. 05 was touched in twenty-seven commits, and the companion finder returned nothing — no co-change pair at all. Not because it travels alone, but because its companions are whole features, not files, and the instrument was built to look for files. Absence of evidence, shaped by the question asked.
None of this voids the other nine essays. The numbers in them are real, and where the instrument breaks, it breaks loudly and in public — which is the only honest way for an instrument to break. But it does set the terms. Every reading in this series is a reading through a particular tool, with a particular reach, run on a particular day against whatever history it was given.
Bowker and Star's point was that infrastructure goes invisible when it works. The same is true of measurement. You stop seeing the ruler and start trusting the length. This essay is the ruler, held up to the light.
The series excavated the labor a finished thing hides. The excavator is a finished thing too — and it hides its own labor, and its own limits, exactly as well.
The tool is another stratum. Measured here, in turn, and just as partial as everything it measured.