Research Guides
Check Server Timing Before Reporting Zero Duration
A performance export can report a server metric of zero even when the server did not provide any duration. Before calling that result instantaneous, check the original timing header and whether the measurement was actually supplied.
This worksheet examines a single measurement boundary in website performance research. It does not diagnose a particular website, promise a ranking change or describe a Colquhoun.shop feature.
1. Separate an existing metric from its duration
W3C's Server Timing specification describes named metrics a response can report. The duration parameter, dur, is optional. In the browser interface's duration getter, both an absent parameter and a parameter that fails the specified number parsing yield zero.
That means the presence of a named metric and a displayed zero do not independently prove a measured zero. The specification recommends milliseconds, but cannot enforce what the producer intended. For an unfamiliar metric, verify the emitter's definition and unit rather than assuming its name explains them.
2. Use three synthetic response cases
Imagine a fictional article server produces one metric named render. Compare three variants on the same test route:
Server-Timing: render: the metric appears without a reported duration.Server-Timing: render;dur=0: the producer expressly supplied zero; its precision or rounding still needs interpretation.Server-Timing: render;dur=18.4: the producer supplied a numeric duration. Under a documented millisecond convention, the reported interval is 18.4 milliseconds.
The first two variants may both become zero through the browser interface. Preserve their distinction in an export: “duration not supplied” is not the same claim as “producer reported zero.” If the producer sends a value that fails parsing, record invalid metadata as a third condition rather than silently promoting it to a measurement.
3. Keep the raw and derived observations together
- Response: sampled URL, test time and which response in the navigation carried the metric.
- Header observation: metric name and whether its duration was absent, present and parseable, or invalid.
- Browser observation: exposed entry name and duration, or that the entry was unavailable.
- Contract: emitter, units and documented start and end of the measured interval.
- Report decision: measured, reported zero, absent, invalid or unable to verify.
Use a controlled test or authorized observation of the response. Do not expose private timing details, credentials or internal identifiers in a public example. Synthetic metric names and harmless durations are enough to show the distinction.
4. Avoid turning one server metric into a page-speed score
A server can report selected pieces of work without reporting everything on the request path. Two named intervals can overlap or one can contain another, so do not add them unless their definitions justify it. Likewise, a reported render duration is not a complete navigation time and does not include every possible delay before the reader sees content.
Keep browser navigation timing in a separate column. A reader-facing improvement requires evidence of the reader-facing event, not simply a smaller number from one server component.
5. Publish a narrow conclusion
For the fictional fixture, a useful note reads: “The first response exposed a render metric but did not report its duration. The interface showed zero. We excluded that value from the duration comparison and kept the sample marked as unmeasured.”
Retest the export with the explicit-zero and positive-duration cases before closing the finding. That confirms what the measured data can support without inventing a faster server from a default value.