Strata
Ten excavations of the work a finished thing hides
A repository is a dig site. The deployed website is the surface, the part meant to be seen. Beneath it lie the strata: the abandoned attempts, the rework, the files dragged along, the quiet maintenance, the renamings. All the labor that the finished feature is designed to erase.
This series excavates that labor. It is not a changelog and not a postmortem. Each essay
takes one kind of buried work, recovers it from the commit history with the Excavator, and
reads it. The numbers in every essay are real, produced by excavator.js against the
repository and recorded in excavations.json. The method is the argument: a version
controlled repository is one of the few places where the production of a thing is kept in
full, dated and attributed, so the labor can be dug back up after the surface has closed
over it.
The through line is Bowker and Star's observation that infrastructure becomes invisible exactly when it works. Success erases its own history. These ten cuts run against that erasure.
No. 01 · The Surface and the Strata Thesis. What you see of a feature is the smallest part of it. The polish is built to hide the production. The labor it surfaces. Rework: the lines written, rewritten, and discarded to leave the few that ship. The evidence. Insertions written versus lines surviving at HEAD. The survival ratio.
No. 02 · Gestation Thesis. The longest stretch of a feature's life is the quiet before it works at all. The labor it surfaces. The gap between the first commit that reaches for a thing and the commit where it finally stands up. The evidence. First touch to ship date, against the calendar of everything else shipping around it.
No. 03 · The False Starts Thesis. The discarded version is not waste. It is the reasoning that the surviving version inherited. The labor it surfaces. Dead ends: whole files written and deleted before the final form. The evidence. Files that appear and disappear in an area's history and never reach HEAD. The commit messages that killed them.
No. 04 · The Companions Thesis. Nothing ships alone. Every feature reaches sideways into files it did not announce. The labor it surfaces. Co-change: the work that rippled outward to make one visible thing possible. The evidence. The files most often committed alongside the feature, ranked.
No. 05 · The Workhorse Thesis. The most important file in a repository is usually the one no one thinks about. The labor it surfaces. A single file's whole life, told by the Biographer. The evidence. One file across every rename, every companion, every silent year, from birth to current weight.
No. 06 · Aftercare Thesis. Shipping is not the end of the work. It is the start of the part no one celebrates. The labor it surfaces. Maintenance: the commits that arrive after most of the feature was already written. The evidence. Commits to an area after eighty percent of its churn was done, and what they were named.
No. 07 · The Renamings Thesis. To rename a file is to reclassify it, and reclassification is never only cosmetic. The labor it surfaces. Identity changes, the small torques where a scheme and a reality are pulled back into alignment. The evidence. Every rename and relocation, drawn from the same history the Classification Lab reads.
No. 08 · The Fragmented Sitting Thesis. Software is not built in heroic sprints. It is built in scattered hours, against interruption. The labor it surfaces. Attention as the scarce resource: how many separate days a feature actually took. The evidence. Distinct calendar days with a commit to the area, against its raw commit count.
No. 09 · The Quiet Ones Thesis. Some features go silent because they failed. A rarer kind goes silent because it was finished. The labor it surfaces. Doneness: the clean, complete thing that asks for nothing further. The evidence. Shipped, present, low churn, no dead ends. The signal the museum reads, here read as praise.
No. 10 · The Dig Itself Thesis. This series was excavated by a program reading the same history it describes. The tool is another stratum. The labor it surfaces. The reflexive turn: the act of measurement as labor, and what it can and cannot recover. The evidence. The Excavator's own limits. What a commit records, and what it never will.
Each essay is a standalone piece for jakelawrence.xyz. The Excavator regenerates the evidence
for any repository: REPO_PATH=/path/to/clone node excavator.js. Point it at the real history
and the numbers in these essays become yours.
All ten cuts are authored against this repository's full-history run (31 May 2026),
recorded in runs/digest.json, as interactive scrollytelling surfaces in ../surfaces/:
| # | Surface | Real subject |
|---|---|---|
| 01 | essay-01-interactive.html | the bibliography born + finished in a day, 720 of 1,377 lines surviving |
| 02 | essay-02-gestation.html | src/app/ukraine — scaffolded 25 Apr, a 12-day silence, whole by mid-May |
| 03 | essay-03-the-false-starts.html | src/app — 98 dead ends, the abandoned Ukraine login among them |
| 04 | essay-04-the-companions.html | the universal companions — package.json ×31, CLAUDE.md ×18 |
| 05 | essay-05-the-workhorse.html | CLAUDE.md — 27 commits, almost none about it; it only grows |
| 06 | essay-06-aftercare.html | scripts/enrichment — 6 repairs after the 80% mark, next-day bug fixes |
| 07 | essay-07-the-renamings.html | 956 events; the Weight of Salt audio relocated manuscript → public |
| 08 | essay-08-the-fragmented-sitting.html | src/app — a commit on 64 of 110 days, a real per-day heatmap |
| 09 | essay-09-the-quiet-ones.html | .claude — finished in 3 commits, 99% intact, silent because done |
| 10 | essay-10-the-dig-itself.html | the reflexive cut — 37 of 71 areas read survival over 100% |
If you point the Excavator at a different clone, re-author the prose to the new data — the numbers are real or they are nothing.