I Mined 13 Years of My Git History. It Knew Me Better Than I Did.

The commits per year

2014-17  · 18 total                      ← employed; work owned by others
2018     ▊▊▊ 604                       ← independence begins
2019     ▊▊ 382
2020     ▊▊▊▊▊▊▊▊▊▊▊▊ 2,647
2021     ▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊ 3,845
2022     ▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊ 4,735
2023     ▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊ 5,749   ← what I thought was my peak
2024     ▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊ 4,178
2025     ▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊ 4,070
2026     ▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊ 4,875       ← by July 21

The surprise is the last bar. 4,875 commits in 202 days is about 24 per day — a full-year pace of ~8,800, more than 50% above my best year ever. Two things changed in 2026: I took a full-time role on top of freelance work, and I started using AI coding tools seriously.

I "felt" more productive this year. I did not know the number. If you want to measure what AI tooling actually did to your output, your git log already ran the experiment. Commit count is a rough proxy — commits are not equal in size — but a jump this large at the same commit style is hard to explain away.

The clock does not lie

Commits by hour of day, all 13 years:

00–02  ▊▊▊▊▊▊▊▊▊▊▊▊▊▊ 3,420
03–05  ▊▊▊▊▊▊▊▊ 1,986
06–08  ▊▊▊▊▊▊▊ 1,851           ← my only quiet hours
09–11  ▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊ 3,998
12–14  ▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊ 4,638
15–17  ▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊ 5,968  ← peak
18–20  ▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊ 5,002
21–23  ▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊▊ 4,240

Three facts I could not have told you before, and cannot unsee now:

• Almost one commit in four happened between 22:00 and 05:00. 24% of my life's work is night work.
• 25% happened on weekends — and Sunday (4,197) out-produces Saturday (3,782). My weakest day of the whole week is Monday.
• My longest streak is 57 consecutive days with at least one commit. Two months without one single day off. My busiest day ever: 130 commits.

Here is the uncomfortable part. If you asked me, I would have said I protect my weekends and try not to work at night. I have written plans that say exactly that. The data says I have never lived that way, not for one year out of thirteen. When your self-image and your commit timestamps disagree, the timestamps win. Either the plan changes or the behavior does — pretending both are true was only possible before I looked.

Your commit messages are a vocabulary test

I took the first word of all 31,103 commit messages and ranked them:

I always described myself as a builder. My git log says my number one verb, by more than double, is fix. "Add" — the builder's verb — comes fourth.

At first this felt like an insult. Then I looked at it differently: most of my income for years came from keeping running systems alive, not from greenfield builds. The log did not say I am a bad builder. It said the market has been paying me to be a repairer, and I never noticed that was my actual product. (The 840 public reverts I choose to read as honesty.)

The graveyard and the garden

Of my 128 repos, 100 are dead — no commit in the last 90 days. There it is: the classic developer graveyard. Abandoned experiments, weekend ideas, a batch of half-finished apps, forks I touched twice.

I almost stopped there, feeling appropriately guilty. Then I looked at which 28 repos are alive, and the guilt turned into the most useful finding of the whole exercise:

• One client codebase: committed for 7.8 years, still active this week
• Its sister project: 7.7 years
• Four more client systems between 6 and 7 years each
• An open-source case-management app I still maintain: 6.1 years

Every long-lived repo is a system someone depends on. Every dead repo is an experiment nobody did. So the honest reading is not "I abandon things" — it is "I abandon experiments cheaply, and I do not abandon systems that people rely on." Side projects die fast; client work lives for the better part of a decade.

I had been writing proposals describing myself with the same generic words everyone uses. Meanwhile the single most documented behavior in my history — provable with timestamps — is that I stay. For a client with a 10-year-old production app, that is worth more than any list of technologies. I just never thought to sell it, because I never saw it.

Run this on yourself

Local repos are not enough — pull your whole GitHub account as bare, blob-less clones. Full commit history, almost no disk (my 137 repos: 77 MB):

mkdir git-archive && cd git-archive
gh repo list YOUR_USER --limit 300 --json name --jq '.[].name' | while read -r name; do
  git clone --bare --filter=blob:none "[email protected]:YOUR_USER/$name.git" "$name.git"
done

Then extract commits from every repo:

for d in */; do
  git -C "$d" log --all --author='[email protected]' \
    --pretty='%ad|%s' --date='format:%Y-%m-%d %H %u' 2>/dev/null
done > all-commits.txt

Two traps that will silently ruin your numbers:

• You have more identities than you think. Run git log --all --format='%ae' | sort | uniq -c across repos and read the list by hand. I found four: my main email, the GitHub noreply address (2,299 hidden commits), an old employer address — and a typo domain, gmai.com, quietly holding 129 commits. Also check for colleagues with names similar to yours before you grep loosely.
• Timestamps are local time of the machine that committed. Good enough for patterns, do not over-read single hours.

Then it is just counting: commits per year, per hour, per weekday, first word of each message, days since last commit per repo. Any scripting language does it in 50 lines — or paste the file into an AI assistant and ask for exactly the charts above.

What to actually look for

The charts are fun. The value is in the disagreements:

• Where does the data contradict your self-description? Mine: "builder" vs. 7,201 fixes; "protects weekends" vs. 25% weekend commits.
• Which repos refuse to die? The dead ones are noise. The survivors are your real positioning — whatever they have in common is the thing people actually pay you for.
• Where does your archive start, and why? Mine starts fully five years after my career did. The empty years tell a story too: that work was never mine.
• What changed and when? Output jumping 50%, night work creeping in, a repo going quiet — the log dates every turning point you lived through without noticing.

Twenty minutes of shell scripting gave me a more honest review of my career than any yearly reflection I ever wrote by hand. You already own the data. It has been recording you the whole time.

Some details are generalized so clients and codebases cannot be identified.