One line per selected variant across weekly runs. Click a legend entry to toggle; drag the slider to zoom the time range. Same-color lines share the chosen category; dash style separates variants within a category. A hollow amber marker flags a low-confidence point whose replicates did not converge; a solid red marker flags a suspect point whose throughput is beaten by a larger payload (a measurement or config artifact, not a real result). A diamond marks the run where that library changed version, so a step in the line reads as a release rather than as noise -- hollow when the point is otherwise trusted, amber or red when it also carries one of the flags above. The version is on every tooltip row either way.