A gate without a spend ceiling is not a gate
My pipeline could tell you whether the work was good. It could not tell you that the answer had cost four hours and eight failed attempts to reach.
Read this entry →Working log · started August 2026
What I am building, what broke, and what the evidence actually supports. Entries are dated and finished when they are useful, not when they are polished. There is no schedule, and I would rather write nothing than write filler.
My pipeline could tell you whether the work was good. It could not tell you that the answer had cost four hours and eight failed attempts to reach.
Read this entry →Swap the harness and the benchmark moves three points. Swap the model and it moves twenty-five. So the harness barely matters — unless you look at what the benchmark never measures.
Read this entry →Three deploys failed. The site kept answering with 200. I only found out because I went looking for something else entirely.
Read this entry →