Build notes and data releases
The decisions behind JobAIRisk, the corrections we publish, what the corpus can support, and what it cannot predict.
A title can provide a useful baseline. It cannot describe your week, predict an employer decision, or tell you which work to strengthen next. This is the product reasoning behind the free-first flow.
Read the builder's noteThe first classifier over-scored physical and in-person work. We kept the mistake visible, rebuilt the task values against an anchored rubric, and documented what r2 still does not solve.
Read the release story