Flavio F. Contreras-Torres @flaviocontreras.link · 11/09/2026The failure is the policy working — not breaking. 000
Flavio F. Contreras-Torres @flaviocontreras.link · 21/08/2026Comparing classifiers is easy. Keeping the comparison reconstructable is harder. An F1 score is a claim. Without the split, the seed, and the preprocessing that produced it, no one else can test that claim. 000
Flavio F. Contreras-Torres @flaviocontreras.link · 16/08/2026Building research software in public has stages, each raising the bar. A script works on your machine. On GitHub it needs docs. When another project depends on it, it needs a stable contract. Peer review finds what you missed. Somewhere along the way, the side hobby becomes part of the research. 020
