Scrub a clip one second at a time and see the cover that second would produce, before
committing to it. The chosen second is written back into the timestamp field, so it can be pasted
straight into capture-bench.html or a ticket.
capture-bench.html builds a fresh
<video> per capture, which is the right way to time a single cold capture. A picker does the
opposite: it loads one element and re-seeks it, so only the first frame pays the load. That is the
cost model a real cover picker has, and it is what this page measures — watch seek versus
first load in the table. Seeks are serialised (latest-wins): concurrent seeks on iOS Safari block
the thread and yield transparent frames.