Evals & safety2026-07
LangSmith Evaluation System
How Solerno measures agents in production: traces, datasets, judges, and a loop back into the next run.
Agents without evals are demos. This system wires LangSmith into the way we ship: golden datasets, pairwise judges, regression gates, and traces that become the next training or prompt change.
It is the measurement layer behind client work and internal research. FigmaBench is the public cousin. This is the private one.