Dev Stats

Anyone can say they have been programming for twenty years. This is the measurement: an exploratory analysis of every commit in my GitHub account, read straight from git history rather than estimated.

Every repository was cloned with git clone --bare --filter=blob:none — a partial clone that pulls the commit graph and directory trees but no file contents. 269 repositories, 71 MB, 17 seconds, zero failures.

Twenty years, no gaps

The first commit is dated 6 January 2006 and there has not been an empty year since 2012. Counts below are authored commits, original repositories, merge commits excluded — 3,253 of the 28,103 commits processed.

The 2019 spike (966 commits) is almost entirely the blog's automated static generation, not hand-written work.

The inflection was October 2025

Monthly commits from January 2024 onward. The step up in October 2025 has held for twelve months, which makes it a new regime rather than a spike.

Process improved in steps, not ramps

Share of commits following the Conventional Commits specification (feat / fix / docs / test / ci). Two discrete jumps: in 2025 messages stopped being labels and became descriptions; in 2026 the convention became the default.

Years run newest first; the top row is the chart's highlighted benchmark.

The last twelve months, day by day

Every other figure on this page is frozen at the analysis date. This one is live: the calendar is fetched from GitHub in your browser when the page loads, so it always shows the trailing year as it stands today.

What I build vs. what I study

The same six topics, measured twice: as a share of my original repositories, and as a share of the repositories I fork to read. Forks run one to two years ahead of production — today's forks predict the code of 2028.

Python, with Markdown as a second language

File-type composition across the 9,692 files in the original repositories (9,634 of them classified below). Much of the JavaScript and HTML is compiled output rather than written code, so the real profile is close to monolingual Python.

Working with Spec-Driven Design leaves a measurable trace in the file tree: the specs are committed alongside the code.

A portfolio and a reading library

Just over half of the account is forks. Those are not abandoned copies: they are the reading list, which is why they are counted separately from the work.

The code gets written on Sundays

One bar, split by day of the week from Monday to Sunday. Saturday and Sunday together take more of it than Monday, Tuesday and Wednesday combined.

A sustainable rhythm rather than an all-nighter habit: the work is episodic and concentrated — 397 active days, a median of 3 commits on an active day, and a maximum of 126 in one.

From repository to release

Not every repository is meant to ship, and the funnel says how many actually do: how many are original work, how many show real branch and PR flow, how many carry tagged releases, and how many end up installable.

Sprints and marathons

Each point is a repository: horizontally, the days between its first and last commit; vertically, how many commits it holds. The upper left is compressed sprints; the right is the long-running projects. Hover a point for its name.

Not plotted: ecrespo.github.io, with 1,451 commits over 2,609 days, most of them the blog's automated static generation rather than hand-written work. On the same axes it would squash every other repository into one corner.

Two projects the numbers single out