Plots several per-request latency samples under each API endpoint as a single distribution series, so the tight, fast spread of /profile reads very differently from the wide, slow tail of /checkout.