I have measured almost everything about notifio.app and the desktop app behind it: the sitemap against git, every version number we write to disk, every npm script, the README. I had never run git log over the whole thing.
$ git log --oneline | wc -l 55 $ git branch -a * main remotes/origin/HEAD -> origin/main remotes/origin/main $ git tag v1.0.1 $ ls .github ls: .github: No such file or directory Fifty-five commits, one branch, one tag, no CI. The product is a paid Electron app with installers for three platforms and a 38 page marketing site, and that is the entire recorded history of it.
By my own count of my dev.to archive, 169 of my posts are about this codebase. The repository has 55 commits. Three posts per commit, and that ratio is not the interesting part.
The dates
$ git log --format=%ad --date=short | sort | uniq -c | sort -rn | head -5 11 2026-07-21 9 2026-07-22 5 2026-07-28 5 2026-07-27 4 2026-07-20 $ git log -1 --format='%h %ad %s' --date=short df3a8f7 2026-10-18 Switch fonts to self-hosted local woff2 files The first three days of the project hold 24 of the 55 commits. The last one landed on 18 October. It is 16 November today, and 66 of those 169 posts were published after the final commit.
That is not a dormant project. In the four weeks since that commit, the site has been crawled, audited, measured and written about more than in the four weeks before it. What it means is that the work moved from changing the thing to understanding the thing, and the shape of a git history cannot show that at all. Every one of those 66 posts is a measurement of a tree that has not moved, which is exactly why they could be taken: nothing changed underneath them.
The vocabulary
$ git log --format=%s | wc -w 388 $ git log --format=%s | sort -u | wc -l 37 $ git log --format=%s | grep -ci 'made changes' 26 $ git log --format=%s | sort | uniq -c | sort -rn | head -5 15 made changes to app 3 made changes to the app 2 made changes to our app 2 made changes to my app 1 updated the landing page Three hundred and eighty-eight words for an entire product. Thirty-seven distinct subjects for 55 commits. Twenty-six of them contain "made changes", and fifteen are byte-identical. Four variants of the same sentence differ only in the article: to app, to the app, to our app, to my app.
I did this census on another of my repositories a few weeks ago and found the same habit, so this is a personal fingerprint rather than a property of the project. What is new here is that the vocabulary changes, and exactly where:
38 2026-07-30 made changes 39 2026-08-03 made changes to app by adding resend webhook 40 2026-08-06 updated clips 41 2026-09-06 Add about page and publisher organization data 42 2026-09-10 Add free license grant script and email 43 2026-09-14 Add international pricing, Managed Payments, and content pages Commits 1 to 40 run from 20 July to 6 August and are interchangeable. Commits 41 to 55 run from 6 September to 18 October and every one of them names what it did. Between the two groups there is a month with no commits at all.
So the writing did not get better gradually. It got better across a break, and what changed over that break was not my discipline, it was what the commits were for. The July commits were me keeping a working tree safe between evenings. The September ones are the only record of decisions I expected to have to explain later, and eight days after the first of them I started publishing about this codebase.
The three outliers
The longest subject is 544 characters, against a median of 23:
Bundle Playwright Chromium + dual macOS DMG downloads - Set PLAYWRIGHT_BROWSERS_PATH in main.ts to bundled path (packaged + dev) - Add scripts/copy-browsers.js to copy Chromium from local cache pre-build - Wire copy:browsers into prebuild script - ... That is a bullet list flattened into a single line, so git log --oneline prints the entire release engineering decision as one unreadable row. It is the best-documented commit in the repository and the worst-formatted one.
Exactly one commit has a body:
$ git log --format='%h|%s|%b' | awk -F'|' 'length($3)>0' 9a1e94d|Redeploy with new database|Co-Authored-By: ... One body in 55 commits, and it is a trailer rather than a paragraph.
Two commits changed no files. One is a merge. The other is the one with the body:
$ git show --stat --format='%h %s' 9a1e94d | tail -1 9a1e94d Redeploy with new database No diff. It exists because the deploy is driven by a push and the database behind the site had been replaced, so the thing that needed to change was outside the repository. An empty commit as a deploy button is a legitimate pattern, and it is also the only commit here whose message describes something git cannot see.
Which half of the repo the work went into
The repo is two packages: app/ is the Electron product, server/ is the Next.js site and API.
23 commits server/ only 15 commits both 7 commits app/ only 10 commits neither Seven commits in four months touch only the product that people pay for. Thirty-eight touch the website. That sounds damning and is mostly honest: the app reached "it works" in July, and everything since has been pricing, currency handling, SEO, content and the 15 pages about rental sites that bring anybody to it at all.
The 10 commits that touch neither are the ones I would have forgotten existed. Two edit .gitignore, one is a merge, one is the empty deploy, and one deleted a whole package:
$ git show --stat c485ff4 | tail -1 # Remove video studio and ad clip library 44 files changed, 9191 deletions(-) That is the Remotion ad studio I deleted in October, and it is the single largest change in the history. The repository's totals:
$ git log --shortstat --format='' | grep -oE '[0-9]+ deletion' | awk '{s+=$1} END {print s}' 20310 $ git log --shortstat --format='' | grep -oE '[0-9]+ insertion' | awk '{s+=$1} END {print s}' 58078 Twenty thousand lines deleted against fifty-eight thousand added, and 9,191 of the deletions are one commit. Just under half of everything ever removed from this project left in a single afternoon.
The tag that stopped being true
$ git show --format='%h %ad %s' --date=short v1.0.1 | head -1 5c6518d 2026-07-20 made changes to app $ grep '"version"' app/package.json "version": "1.1.0", The only tag in the repository points at the third commit ever made, is named v1.0.1, and the message on that commit is "made changes to app". The app has been 1.1.0 for a while. Nothing reads either number: reinstalling is the update mechanism, and the installers sit in a bucket under fixed object keys with no version in the filename, so there is no artefact anywhere that a tag could correspond to.
That is the one thing on this list I actually intend to change. Not the commit messages, which are already fossilised and which the posts document far better than a subject line would. Not the branching, because there is one developer. A tag per release, cut at the commit the DMG was built from, so that when somebody emails about a bug in a build I can answer which tree it came out of. Right now the honest answer to "which commit is the installer you are running built from" is that I would have to guess from the dates.
The summary, if you want the uncomfortable version: the useful history of this project is 169 blog posts and a 55 commit log that cannot tell you what any of them were about.