Figure 15.2: One agent run drawn as a span waterfall
Figure 15.2 One agent run drawn as a span waterfall. Each bar is a span: where it starts
is when it happened, how long it runs is how long it took, and the indentation of its
row records who called whom—the tool spans nest under the model call that asked for
them, and everything nests under the root run. Every span carries the same record, the
attributes bracketed over the first model call: the rendered prompt, the tokens, the
cost, the latency. The bug is not the final answer but the first wrong span—here the
long database call that erred, in accent—and a waterfall makes it the thing your eye
lands on. The timings and ordering are illustrative.
