@tenderlove I thought it was time to first content byte, because otherwise the metric makes no sense in an http/2 context.

@tenderlove I did that for a frontend that was server rendered react and it was slow because of shit ton of backend api calls.

The server would reply immediately with the html, its head and a couple of elements of the body, just enough to get the browser to start fetching css, fonts, …
The rest would come later as the API would start replying.

This required some amount of work to get compression and buffering correct. We had to move the the compression near that code just so we could force a flush of its buffer. This also required a cache-control: no-transform to make sure the reverse proxies would not play funny.

@tenderlove Pitchfork/Unicorn does that by default.

Not to game any benchmarks though, but to check the client is still there.