Research Guides
Separate First Byte From Final Document Timing
A performance report can show that a browser heard from a server sooner without showing that the finished page arrived sooner. Before announcing a first-byte improvement, identify what the tool counted as the first response and keep a separate timing for the final document.
This worksheet is for comparing website navigation measurements. It does not label a particular site as fast or slow, establish a ranking effect or describe a Colquhoun.shop performance feature.
1. Name the timing boundary
Google's web.dev guide to Time to First Byte explains that navigation TTFB can include redirects, connection setup and the wait before a response starts. It is not a direct measure of application execution time. An informational response such as HTTP 103 Early Hints can also be the first response a tool reports.
The guide distinguishes responseStart from finalResponseHeadersStart, which marks the start of the final response headers where available. Check the browser version and the measurement tool's definition. If the final-header field is not available or does not provide a usable observation, record that limit; do not copy the first-byte value into it.
2. Reproduce a fictional before-and-after
Suppose a navigation test before a change records no informational response, final document headers at 420 milliseconds, and the response ending at 620 milliseconds. After a change, the test records Early Hints at 150 milliseconds, final headers at 420 milliseconds and response end at 620 milliseconds.
For a tool that reports the first observed response, the reported start moved 270 milliseconds earlier. Neither the final headers nor the response end moved in this fixture. “The browser received an informational response earlier” is supported. “The finished document arrived 270 milliseconds sooner” is not.
The earlier hint might still allow useful work to start earlier. To make that claim, measure the relevant later event, such as a requested asset starting sooner or a measured display milestone changing. The unchanged final-header timing does not rule out those benefits, just as an earlier hint does not prove them.
3. Keep a navigation timing record
- Navigation: initial and final URL, with any redirects.
- Environment: browser and version, test tool, connection conditions and location.
- Cache and reuse: whether the run started with warm or cold conditions as defined by the test plan.
- Response boundaries: first reported start, informational response if observed, final headers if available and response end.
- Reader-facing event: the later outcome actually measured, or explicitly untested.
Keep values on the same timeline and in the same units. A duration from a command-line client and a timestamp relative to browser navigation can have different boundaries even when both are called first byte. Do not graph them as one comparable series without reconciling the definitions.
4. Compare like with like
Use the same entry URL, browser version and run conditions for both samples. A newly removed redirect, a reused connection or a warmed cache might change the measurement without isolating what happened at the origin. Preserve the individual runs and note which variables actually changed.
W3C's Navigation Timing specification defines browser navigation performance entries. When using that interface, save the actual field names alongside the values so the next reviewer can tell which boundary you measured. Keep a separate note for any server instrumentation used to investigate processing time.
5. Publish the narrow finding
A clear review note is: “In the sampled browser, the first response moved from 420 to 150 milliseconds after informational hints were introduced. The final headers and response end were unchanged at 420 and 620 milliseconds. Visual loading was not measured.” Report different outcomes separately instead of compressing them into a single speed claim.