Free tool

HAR File Analyzer: Find Failed Requests

Analyze a HAR locally to find failed and slow requests, read timing phases, and copy a safe summary into a reviewed bug report without replaying traffic.

How do you analyze a HAR file to find the first failed request?

A HAR file analyzer reads a browser network recording and helps you compare request methods, HTTP statuses, and timings. Use it as a quick HAR checker: analyse a recording to find the failed or slow request. Start with the first failed request in the captured sequence, then check surrounding requests and the page behavior. The first failure you see is evidence to investigate; the HAR alone cannot establish the root cause.

Choose a HAR 1.2 file or load a synthetic example below. This analyzer accepts files up to 20 MiB, 10,000 entries, and 32 levels of nesting. It parses in a browser worker, does not upload the file, and never replays its requests.

Maximum 20 MiB. Files with more than 10,000 requests or nesting deeper than 32 are rejected.

Builder example: compare a successful read with a slow HTTP 500 submission. Agency / QA example: distinguish a missing resource from a request with no recorded response. Both are synthetic captures.

Collect reports on your own site with ContextCapture

ContextCapture adds a bug report widget to your website. Testers send a screenshot, their environment and their notes from the page, reports save to your inbox and, when the site’s destination is GitHub, open as issues in the repository you pick.

Use the same evidence habits on your own site: testers report from the live page, and each report saves to your inbox.

Read a synthetic four-request trace

Imagine a checkout attempt with these four HAR entries in capture order. The statuses and times below are illustrative, not a customer result or a file loaded by the example buttons.

#1 GET · 200 · 45 ms

The page request received a successful HTTP response. That does not prove the checkout interaction worked.

#2 POST · 500 · 1,400 ms

This is the first observed failed response. Record its request number and the action that triggered it; inspect server logs before assigning a cause.

#3 GET · 302 · 30 ms

A redirect points to another location. Check the following request and expected destination before calling it a problem.

#4 POST · status 0 · unknown

No HTTP response was recorded. The capture alone cannot say whether a request was blocked, cancelled, or lost elsewhere.

Illustration of network request rows, one highlighted for inspection, leading to a bug-report document
Illustration: inspect a request in context, then write down only the evidence you have reviewed.

Interpret statuses and timing without guessing

The Failed filter includes 4xx and 5xx responses plus status 0. HTTP 3xx responses are redirects, not automatic failures; 4xx means a client error response and 5xx a server error response. Use the MDN HTTP status reference to interpret a specific code, and the HTTP status codes for bug reports guide to record one without guessing its cause.

Slowest sorts by recorded total time. Expand a request to compare blocked, DNS, connect, SSL, send, wait, and receive phases. A long wait includes network latency and server preparation time, so it does not by itself identify a slow server. SSL may be included in connect; do not add all phases to calculate a new total. Missing or negative fields appear as unknown, not zero. Chrome explains these phases in its Network panel reference.

Request numbers retain the HAR entry order, but that order alone does not prove one request caused another. Compare the captured action, timestamps, initiators in DevTools, and any relevant application logs before describing a sequence as causal.

When the failed request is missing

Check the capture boundary before changing the diagnosis

If the page failed but the expected request is absent, confirm Network recording was active before the action. Clear the log, turn on Preserve log for a navigation, reproduce once, and compare the Network panel with the new HAR. Chrome notes that DevTools records while its Network panel is open and that Preserve log keeps requests across page loads. An absent row in this analyzer is evidence about the capture, not proof that the app never attempted the action.

If a request appears with status 0, record the browser-visible error and request timing separately. Do not convert the missing HTTP response into a 4xx or 5xx diagnosis. See Chrome’s recording and Preserve log instructions.

Capture, review, and hand off the finding

  1. In Chrome DevTools, open Network, clear earlier requests, reproduce once, and export the HAR. Use Preserve log if the failure spans navigation. Chrome documents the recording and sanitized export controls.
  2. Load the HAR here. Compare All, Failed, and Slowest; note the first observed failed request, method, status, total and phase timing, plus the exact action and expected result. The copied summary includes up to ten slowest requests and omits URLs, headers, and bodies.
  3. Review the summary before sharing. If the team needs a HAR attachment, process the original file with the local HAR sanitizer, then inspect every retained hostname and path. A sanitized export can still reveal private names or IDs there.
  4. Put the reviewed finding beside clear reproduction steps in the bug-report generator. The sample report shows the level of context a developer needs.

Updated October 21, 2026. Method: this page describes the analyzer’s current local parser, filters, and copied summary; status and timing guidance follows the linked Chrome and MDN references.

Examples for your team

Builder example

A builder compares checkout requests by status and timing to identify the first observed failed response.

Start your 14-day free trial

Agency / QA example

A QA tester groups slow requests from a staging session and shares only reviewed findings with the client delivery team.

Start your 14-day free trial

Capture the next report with ContextCapture

Start a 14-day free trial. The beta plan is $7.99 per month for unlimited projects and reports, with up to 5 team members.