It's right in the article :P

web.dev/articles/opt...

Anyways, the reason Michael Crichton probably thought T. Rex's vision was based on movement was because he didn't know the Opit was above the Brale but *behind* the ⨡arl

`import defer` will get you some of what you're looking for. The file is still loaded, but the synchronous parts of the module (and submodules) won't execute until first property access of the module namespace object.

github.com/tc39/proposa...

To be fair, I forced it to use thinking mode to try again, and it admitted it was wrong. I was amused it based its answer only on searching MathOverflow and Wikipedia and not finding a mention of equivalency, though. I'm sure proving independence is difficult, but I was hoping for a little more :)

I had a funny one when I saw a claim about π + e and the Riemann hypothesis on the math subreddit that I hadn't heard. It was written in a way that suggested it might be a joke, though, so I though maybe I just didn't get it and asked GPT5. The first line is so intriguing! And then just nonsense

small multiples! en.wikipedia.org/wiki/Small_m...

Looks fun, any shareable information about how "Create variants" is constrained? Is "give me more that looks like this" enough or does it need to go deeper?

I vaguely remember the Wario game being ok? And Mario Tennis being so basic that it might as well have been pong. And yet, yes, the kid still in me who saved up allowance to rent one from Blockbuster for a weekend really wants one

scheduler!

Great answers!

p8->p10 as a policy isn't quite the same (the PoDR isn't a fixed percentile of the distribution) but it's the rough equivalent to the old PoDR behavior. The p10 values also have the most manual tweaks, e.g. the actual p8 CLS is 0, but the p10 used is 0.1 to better recognize what CWV considers "good"

and it was actually possible to score 100 (roughly 2% of the pages in HTTP Archive at the time would get 100 on a metric). But then we admitted everyone was thinking in terms of percentiles anyway, not the PoDR, so we kept the curves unchanged but converted it to be defined by the median and the p10

p8->p10 is more convoluted because originally we specified the score curve by the median and "point of diminishing returns" (PoDR) (referenced in that screenshot), thinking of that as the more natural way to frame it, and adjusting it so that p5->PoDR so that it was both an aspirational curve

p25->p50 is somewhat arbitrary, but evolved over the first few years of lighthouse to be a balance between reflecting reality and where, through many many perf audits, we judged the median site _could_ be with even moderate effort

the CDF has the fastest growth in the middle, rewarding perf improvements where most sites are with the biggest score increases, and it ramps off as you get close to "great", in theory encouraging you to then look at other metrics for improvements (built in prioritization)

Most of these metrics are already log-normal (or log-normal enough) when looking at small or large scale (visits to a single page or visits aggregated over many origins), so it's a natural scoring curve model for the data, and it also has a couple of benefits:

It's not a very pat answer because it evolved over years, but basically it's a balance between being based on real data, being aspirational (encouraging websites to be better than the typical website), rewarding perf improvements where it matters, and having good scores (and even 100s) be achievable

youtube.com/watch?v=dt8O...

youtube.com/@JPerm does really good explanations, I've found. I've never tried the megaminx, but looks like he has at least one video on it: youtube.com/watch?v=oVRo...