Instead of relying solely on span kind – a limitation shared by most tools – you can now create dedicated interaction types and define exactly which spans should be routed to them. This gives you full control over which calculated properties, auto-annotation rules, and evaluation configurations apply to each specific component in your system.
By separating spans that previously fell under the same umbrella, teams gain clearer insights, more accurate scoring, and greater operational efficiency. You can recalculate only what matters, avoid unnecessary noise, and reduce evaluation costs, while keeping your traces exactly as they are. With Deepchecks, the same spans and the same data now deliver a far more precise understanding of how your system behaves.
Yaron Friedman
