Compare
Select a baseline and current analyzed snapshot through the host's controls. Verify labels/direction: deltas are current minus baseline. Compare summary, per-class histogram, suspect classifications, and all four waste categories.
Comparable captures
Keep workload, capture/GC conditions, JVM/layout, analyzer version/backend, and options comparable. File creation time is not necessarily capture time. IDs are not stable cross-snapshot identity.
| Label | Actual meaning |
|---|---|
| New | Class/candidate appears in the compared output |
| Resolved | Previously reported candidate is absent under the current detector |
| Persisted | Candidate appears in both outputs |
| Grew / shrank | Positive / negative measured aggregate delta |
Resolved is not proof a leak was fixed or a specific byte count freed. Persisted is not proof a continuing leak. Thresholds, workloads, and capture conditions can change classifications; unchanged aggregates can hide object turnover.
Waste and exports
Total estimated waste combines duplicate, empty, over-allocation, and boxed categories. A consistent illustrative change: -40 KiB strings, -5 KiB empty, +10 KiB over-allocation, -1 KiB boxed gives -36 KiB total. This is arithmetic, not a benchmark or guaranteed reclaim.
Exports/downloads include names and can include paths. Review before sharing. Host adapters can compare compact snapshots across isolated processes; native compare requires analyzed states, while the CLI releases the baseline graph before loading current.
See API result fields and Timeline for trend investigation. Comparison controls and export destinations differ by host.