You change a prompt, switch to a new model, or tweak your retriever. You rerun your evals and accuracy goes from 82% to 80%. Did you break something? Most eval tools show you the two numbers and leave ...
An optional knowledge base of alignment examples (agent-training/knowledge_base/) obtained from a training stage on bug-free models helps the alignment step.
Some results have been hidden because they may be inaccessible to you
Show inaccessible results