Web Standards
The W3C HTML WG "200 Sites"
How a working group learned to write a specification by reading the web instead of imagining it.
How a working group learned to write a specification by reading the web instead of imagining it.
Somewhere in the long committee years that produced HTML5, someone made a decision that sounds obvious in retrospect and was quietly radical at the time: before deciding what the language should be, go and measure what it actually was. The result was a corpus of real-world pages, sampled from popular sites and pulled apart tag by tag. In the folklore it is remembered as the "200 sites" study, though the honest number was larger and messier. What mattered was the premise. The specification would answer to evidence.
Standards bodies had spent a decade writing rules that browsers ignored and authors never followed. The great insight of the sampling effort was that a specification is not legislation handed down to a compliant populace; it is a description of a living system that already behaves however it behaves. If ninety-something percent of pages opened with a malformed doctype, then a language that refused to parse them was not being principled. It was being useless.
So the working group crawled. It counted elements, attributes, class names, script patterns, the shape of forms, the depth of nesting. And the data was humbling in the way real data always is.
<font> tags survived long past their obituaries. The spec could deprecate them; it could not pretend they were gone.The web is the largest deployed body of markup in history, and almost none of it validates. A specification that ignores this is a specification about a different web.
The lesson was not "anything goes." It was subtler: describe the common case generously, define the edge cases exactly, and stop moralizing. The new elements earned their place because authors were already reaching for those concepts. The parsing algorithm became rigorous because the input was chaotic. Strictness moved from the author-facing rules, where it had failed, into the browser-facing rules, where it could finally succeed.
There is a wider craft principle hiding in the 200 sites, and it outlasts any particular tag. When you design a system that others must adopt, your intentions are worth very little and their behavior is worth everything. You can argue with a colleague. You cannot argue with a crawl of a million pages.
If you want to understand any standard, ask what evidence it answered to. HTML5 answered to the web that existed, warts and spacer-GIFs and all. That is why it stuck. And it is a decent test to apply to the next standard someone tries to sell you: did anyone go and look first?