A HAR file records browser network activity and can help explain a failed request or a slow interaction. It can also contain credentials, cookies and private responses. Treat the original export as potentially sensitive evidence, not as an ordinary attachment that is safe to send anywhere.
ToolsOnDuty HAR Network Analyzer inspects supported HAR data in your browser. This guide uses the tool's built-in, synthetic two-request example rather than private production traffic. Its actual results show why request duration, recorded span and known transfer size must be interpreted separately.
Capture only what you need, and review it before sharing
If you need to collect your own HAR, open your browser's developer tools and Network panel, clear unrelated entries, reproduce the authorized problem and export the relevant network log. Names of export commands vary by browser. Use Preserve log only when navigation would otherwise remove evidence you actually need.
Check the export settings and contents. Even an export labeled sanitized may retain sensitive URLs, request bodies or response data. Do not assume every token is a query parameter. Paths, hostnames, identifiers and timing metadata can identify customers or internal systems.
For troubleshooting, a test account and a minimal reproduction are often safer than a full session from a real user's account. Follow your organization's handling rules, and revoke or rotate exposed credentials when appropriate. The analyzer does not turn an unrestricted production export into safe public data.
Load the HAR locally
Open HAR Network Analyzer. Select a UTF-8 HAR or JSON file through Open HAR, or paste the original JSON into HAR JSON and click Analyze HAR. The workspace supports up to 10 MiB and 10,000 HTTP(S) requests.
For this walkthrough, click Try example. The sample includes a successful GET request taking 160 milliseconds and a POST request returning HTTP 503 after 400 milliseconds. The POST starts 100 milliseconds after the GET begins. The example is synthetic data shipped with the tool, not a measurement of ToolsOnDuty's production performance.
Read the summary without adding overlapping requests
The example reports two requests and a 500.0 millisecond recorded span. Span starts at the earliest request start and ends at the latest request end. The second request starts at 100 milliseconds and ends at 500 milliseconds, which explains the span. Adding 160 and 400 to get 560 milliseconds would ignore overlap.
Recorded span is not full page-load time. It depends on which requests were captured and does not include every rendering task, script execution, interaction delay or event needed for a complete performance measurement.
The GET has 2,048 bytes of response body and 320 bytes of headers, giving 2,368 known transfer bytes. The POST has an unknown header size, so its transfer remains unknown instead of treating missing data as zero. The overview shows one unknown transfer size and one server error. Decoded content size is a different measure and should not be substituted for wire transfer.
Isolate the failing request
The filtered view has one matching request, but the overview still describes the complete captured dataset. Likewise, the downloadable summary includes all analyzed requests, not just the current filter. Keep those scopes separate when preparing a report.
- Set Status to HTTP errors. Only the sample POST returning 503 should remain in the matching-request list.
- Use Slowest first under Sort when comparing several requests, or Largest known transfer for size investigation.
- Expand the POST request to read its recorded timing phases, MIME type and transfer information.
- Keep Show full URLs in results unchecked unless you have a specific, authorized need to reveal the original values.
Interpret timings as clues, not a diagnosis
For the sample POST, send is 2 milliseconds, wait is 390 milliseconds and receive is 8 milliseconds. Waiting dominates the recorded 400 milliseconds. That is a useful starting point for investigation, but it does not prove a slow database or identify the responsible service.
Correlate the request with authorized server logs, request IDs and the reproduction context. Network latency, queues, upstream services and server processing can all affect what the browser observes. A 503 indicates a server-side failure response, but the HAR alone does not establish its cause.
Missing timing phases stay Unknown. SSL time overlaps connection time and should not be added again to inflate a total. A status-0 request means no HTTP response was recorded; it does not by itself distinguish cancellation, connection failure, browser blocking or another interruption.
Know what redaction does and does not cover
Result URLs hide URL credentials, query values and fragments by default. However, the raw JSON input above the results remains unredacted. Do not include that input area in a screenshot if it contains secrets. Headers, cookies and response bodies are not displayed in the analysis.
Download redacted summary (all requests) produces a reduced report that omits headers, cookies and bodies and redacts URL values, even if you enabled full URLs in the on-screen results. It is not a cleaned copy of every field in the original HAR.
Review the summary before sharing. Hostnames, paths and other remaining metadata can still be sensitive. A customer identifier embedded in a path is not made anonymous just because the query string was redacted. Never attach the original HAR merely because the summary looked acceptable.
Troubleshoot incomplete evidence
- JSON cannot be read: verify a valid UTF-8 HAR structure rather than an HTML error page saved with a .har extension.
- An expected request is missing: repeat the authorized capture with the Network panel open and check navigation or logging settings.
- Transfer or timing is Unknown: the source export may not provide that field. Do not replace missing values with invented measurements.
- The file exceeds limits: capture a smaller reproduction or use appropriate local development tooling for larger logs.
- The page still feels slow despite quick requests: investigate rendering, scripts and user interactions with other performance tools.
Try the related tools
Apply the steps from this guide directly in your browser.
Frequently asked questions
Is recorded span the same as page-load time?
No. It covers the earliest captured request start through the latest captured request end. It is not a complete browser performance or page-load measurement.
Why is a transfer size unknown when body size is present?
The tool uses a known transfer measurement or known headers plus body. Missing header data can prevent a reliable total, so it stays unknown.
Is the redacted summary guaranteed safe to share?
No. It omits several sensitive fields and redacts URL values, but hostnames, paths and remaining metadata still need review. It is not a complete original-HAR sanitizer.
Does a long wait phase prove the database is slow?
No. It is a clue to investigate with authorized server-side evidence. The browser's recording alone cannot identify the internal cause.
