HAR File Analyser

How this reads your output

The export is parsed and summarised around the distinction the waterfall makes you squint at: whether a request was slow or merely waiting. A request blocked for 800 ms behind the browser's connection limit is not slow — it is queued, and a faster server does nothing for it. Blocked time is reported separately for that reason, and the slowest requests are broken into time spent waiting for the server against time spent queued, because those have completely different fixes.

What the results mean

Time spent queued

Blocked time — waiting for a connection rather than for a response. High queueing means too many requests, not a slow origin.

Waiting for the server

Time to first byte for that request. This is server-side work, and the one figure a CDN cannot improve for uncached responses.

Transferred bytes

What crossed the network. Compressed, so smaller than the decompressed size developer tools also show.

Third-party origins

Requests to domains other than the page's own. Usually the traffic you control least and can optimise least.

Common problems and fixes

One request dominates the waterfall

Check whether its time is waiting or blocked. Waiting is the server; blocked means it was queued behind others and the fix is fewer or earlier requests.

The totals do not match developer tools

Developer tools may be showing decompressed size while a HAR records transferred bytes. Both are correct and they answer different questions.

Nothing looks slow but the page feels slow

A HAR ends where the network does. Rendering, layout and script execution happen after the last byte and are invisible here — that is a performance profile rather than a HAR.

Frequently asked questions

More in DevOps

• JSON Formatter / ValidatorFormat and validate JSON — with the error pinned to a line and column.
• JWT DecoderDecode a JWT to read its header and claims — safely, in your browser.