Why this log exists
A short note on what this blog is for and what kind of writing will land here.
This is a working log, not a polished publication. The papers go through peer review; this is where the rest goes: the things I figured out the hard way, the dead ends worth remembering, and the small decisions that never make it into an abstract.
What lands here
Mostly three kinds of posts.
Build notes. When a system finally works, there is usually one detail that made the difference. A calibration step, a data format quirk, a deployment trick on constrained hardware. Those are easy to forget and expensive to rediscover, so I write them down.
Research detours. Ideas I tried that did not beat the baseline. Negative results almost never get published, but they save time for anyone walking the same path. If a metaheuristic underperforms a simple random forest on my data, that is worth saying out loud.
Field reality. A lot of my work has to survive contact with the real world: limited compute, messy sensor data, thin bandwidth, and budgets that do not stretch. The gap between a benchmark and a thing that runs on a Jetson at the coast is where most of the interesting problems live.
What does not land here
No hot takes, no tutorials I am not qualified to write, no reposting other people's work as my own. If a topic deserves rigor, it goes into a paper with a reproducible repository, not a blog post.
That is the whole charter. Short, honest, and useful to at least one person, even if that person is me six months from now.