Create comparable capture conditions
A before-and-after capture is most useful when the task, browser state and network conditions are deliberately controlled. Warm caches, an existing connection, a different account, changed response data or a shorter capture window can change timing even when application behavior has not improved. Record these conditions separately without placing sensitive notes in a shared HAR.
Load the baseline as the primary capture and the later run as the optional second capture. The included example demonstrates a failed response becoming successful and a shorter wait. Its data is generated locally so you can exercise matching, selection and exports before opening any real account traffic.
Understand the matching rule
The workbench pairs requests by method, origin and path. Query strings are ignored because they often contain session or tracking values. This can also merge different logical resources behind the same endpoint. Repeated matching keys are paired in capture order and flagged rather than presented as unique proof of identity.
Compare response status, duration, waiting time and captured body sizes together. An error response can be faster because it contains less data, while a successful retry can take longer for a legitimate reason. Requests absent from the later capture and newly observed requests are counted separately. Filtering the primary export narrows the requests included in its comparison report.
Retain useful evidence with clear limits
Use the matching table to identify candidates for a closer review, then inspect the related local request details. A descriptive timing difference is a lead, not a causal diagnosis or a statistically reliable benchmark. Repeated controlled runs and application-side evidence are needed when the result affects a production decision.
The exported comparison uses the same endpoint-redaction policy as the reviewed package. The second original capture is never bundled. When you retain original paths or additional fields, review the resulting comparison as well as the HAR: a separate report can reveal context even without credentials. Preserve private originals for your own investigation and send only the reviewed files needed to explain the issue.