How we measure
Every number on this site is either measured by us on the live page or labelled as coming from somewhere else. This page says which is which.
What we measure and why
Each measurement run produces six kinds of evidence: a screenshot, Core Web Vitals, transferred page weight and request count, the fonts used by rendered text, an area-weighted colour palette, and detected platform technology.
These are facts we can observe from the live page rather than claims copied from a marketplace listing. Price, license, version requirements and publication dates remain source-supplied facts and are labelled separately.
How each measurement works
The table below is the measurement contract. Runs start with a fresh page load; the viewport, network profile and cache state belong to the stored run. If that context or the measured date is missing, the public page must omit the figure rather than imply a comparable result.
| Evidence | Tool | Viewport | Network and cache |
|---|---|---|---|
| Screenshot | Chromium capture | 1440×900 and 390×844 | Native profile · cold load |
| LCP and CLS | Lighthouse | 390×844 | Simulated mobile · cold load |
| Page weight and requests | Chromium network log | 1440×900 | Native profile · cold load |
| Fonts | CSSOM and computed styles | Both capture viewports | Same cold run |
| Colour palette | Area-weighted image sampler | Both screenshots | Same capture |
| Platform technology | Headers, scripts and DOM signatures | Both capture viewports | Same cold run |
When we measure again
Every published figure carries its measurement date. High-traffic entries are refreshed first; the remainder rotate through a six-month queue. We do not blanket-crawl the whole catalogue on a fixed daily schedule.
A demo can be offline, behind a login, or block automated clients. When a run cannot complete we keep the dated earlier result or omit the field; an empty value means unknown, never zero.
Platform detection and confidence
Technology is detected from response headers, scripts, fonts and markup signatures on the live page. It is never accepted as self-reported fact. Several matching signals produce high confidence; one weak or ambiguous signature produces low confidence and is presented as such.
Palettes are sampled from the rendered page and weighted by occupied area. This prevents a repeated near-black background pixel from overwhelming the colours that actually define the design.
Challenge a measurement
If a figure or platform detection is wrong, use the Takedown and corrections route and include the entry URL plus the value you believe is wrong. We will run it again and publish the replacement with its own date; we do not silently overwrite measurement history.
Worked example
The example below is one real atlas entry shown with the same dated measurements used on its detail page. It demonstrates the shape of the evidence rather than substituting an invented average.
bohdan.designMeasured 29 Aug 20268Pages4Breakpoints340 KBPage weight