It's okay to make mistakes, that's how you learn and improve. Being conservative has drawbacks of its own. Id argue we need more parties involved earlier in the process rather than just time.
It's a weird balancing act. On the other hand, waiting for everyone to agree on everything means that the spec will take a decade or two for everyone to come together, and then all the additional time for everyone to actively support it.
AJAX is a decent example. Microsoft's Outlook Web Access team implemented XMLHTTP as an activex thing for IE 5 and soon the rest of the vendors adopted it as a standard thing as XmlHttpRequest objects.
In fact, I suspect the list of things that exist in browsers because one vendor thought it was a good idea and everyone hopped on board is far, far longer than those designed by committee. Often times, the initially released version is not exactly the same that everyone standardized on, but they all get to build on the real-world consequences of it.
I happen to like the TC39 process https://tc39.es/process-document/ which requires two live implementations with use in the wild for something to get into the final stage and become an official part of the specification. It is obviously harder for something like a network stack than a JavaScript engine to get real world use and feedback, but it has helped to keep a lot of the crazier vendor specific features at bay.
It's okay to make mistakes, but its not okay to ignore the broad consensus that HTTP2 was TERRIBLY designed and then admit it 10 years later as if it was unknowable. We knew it was bad.
AJAX is a decent example. Microsoft's Outlook Web Access team implemented XMLHTTP as an activex thing for IE 5 and soon the rest of the vendors adopted it as a standard thing as XmlHttpRequest objects.
In fact, I suspect the list of things that exist in browsers because one vendor thought it was a good idea and everyone hopped on board is far, far longer than those designed by committee. Often times, the initially released version is not exactly the same that everyone standardized on, but they all get to build on the real-world consequences of it.
I happen to like the TC39 process https://tc39.es/process-document/ which requires two live implementations with use in the wild for something to get into the final stage and become an official part of the specification. It is obviously harder for something like a network stack than a JavaScript engine to get real world use and feedback, but it has helped to keep a lot of the crazier vendor specific features at bay.