Load a span trace. Get the critical path, concurrency signal, failures, and a diagnosis you can carry into code review.
Each span needs id,start_ms,end_ms. Optional: parent_id,name,kind,status.
Analysis appears here after a valid trace.
The source sample has root zero to two hundred sixty, retrieval A twenty to one hundred twenty, retrieval B thirty to one hundred thirty, and generation one hundred forty to two hundred sixty milliseconds. These interval bars share two pixels per millisecond and a common zero. The outer window is maximum end minus minimum start, two hundred sixty. It is an entered fixture, not a measured live request or a dependency-derived critical path. Durations two hundred sixty, one hundred, one hundred and one hundred twenty sum to five hundred eighty milliseconds. Subtracting the two hundred sixty window gives three hundred twenty, and three hundred twenty divided by five hundred eighty rounds to fifty five percent. Bars share one pixel per millisecond. Because parent and child durations are both included, the displayed overlap is a source excess-sum metric, not an interval-union measure of actual concurrency or verified critical path. The sample has four spans and one explicitly marked error, generation. Four equal boxes count spans and the highlighted box counts that supplied error. Validation requires a nonempty array with identifiers and finite ordered timestamps. A missing error label is not proof of correctness. Invalid input keeps the previous valid report and timeline; reset clears the report and disables export but preserves existing visible metrics. Export contains the complete report and physical generation timestamp; no model or tool is executed.