Last year we filled four engineering seats. Three of them are great. The fourth taught us that our hiring process was lying to us, and that we had been listening.
Resumes are good at telling you what someone has done. They are terrible at telling you whether someone can fix a real bug, in a real codebase, against a real deadline. We were spending eight hours per candidate to learn things a forty-five-minute exercise would have shown us in week one. The cost was not the hours. The cost was the people we did not see because they did not have the right kind of LinkedIn.
So we deleted the funnel. We replaced it with slug-or-die.
The brief: there is a bug in Brigada\Support\Slug::make(). The function drops non-ASCII characters. Përshëndetje Botë becomes p-rsh-ndetje-bot. Привет Мир becomes the empty string. The function powers brigada.dev's journal URLs — we want a real slug for every title we'll ever publish, in every language we'll ever publish in. Open an issue with your fixed src/Slug.php. Endrit, our skeptical engineer running on claude-opus-4-7, comments back within thirty seconds.
No fork. No PR. One issue per attempt. Three attempts per GitHub account. Account must be at least ninety days old, because the world has bots and we do not. The contest closes Friday 2026-05-15 23:59 Europe/Tirane. Top 5 get an interview DM Saturday morning. If fewer than five pass, every passer gets one.
The judge has two layers. Layer 1 is a sealed Docker sandbox running a hidden test suite — mechanical, deterministic, you pass or you do not. Layer 2 is Endrit, default REJECT, who reads the code on merit. Sweet-talking does not work. Production-quality code earns approval. The Endrit system prompt is locked and its SHA256 is published in the repo before the contest opened, so nobody — including us — can move the goalposts mid-flight.
We are not telling you how the leaderboard ranks. The scoring metric, and the direction it cuts, are sealed until the post-mortem. After Friday closes we will publish the metric, the full hidden test suite, the Endrit system prompt in clear text, and the top 5 winning diffs with annotations. If you trust us, you submit now. If you do not, you wait until Saturday and check our math.
We expect this experiment to misfire in some interesting way. We are running it anyway. The alternative is another quarter of resumes that look good and interviews that do not predict the work. We would rather be wrong in public on a Friday than wrong in private for six months.
brigada.dev/join. Live leaderboard. Closes Fri 23:59 Europe/Tirane.