signalsmith

I've submitted a comment on the site but I'll post it here as well:

I think this looks neat. However, last time I implemented something like this, I ran across a few elements/attributes that didn't behave in the simple way I expected.

I can't remember the full list (and can't find my old code), but some things like the "class" attribute, and the "type" attribute of <input> don't like being dynamically changed, at least in all browsers. I think something strange happened with tables as well, and maybe <title>.

So, I ended up needing some special handling for these attributes/elements, in some cases creating a new node with the right attributes and swapping it in.

I don't see anything like this in the Emerj code, so I wonder if it has the same awkward corner-cases. I might go digging to see how DiffDom or DOMDiff or whatever it's called handles it...

The level-4 modifiers are actually useful - I don't see their inclusion as a failure of a standard.

Ambiguity was going to be possible either way - here's a Level 2 example:

    /{+a}/{+b}
            ->  "/foo/bar/baz"

By "reversible" I meant it's possible to find something that reconstructs to the original URL (http://bit.ly/1H27rwD), much like you did.

Um - shouldn't this have the [2012] tag for historical posts?

I mean, I love URI templates. They're neat, they're reversible, but they're not exactly new, right?

I was told a story by a friend who works for a CDN (big one that handles lots of online video):

"Hey - look at this, I think someone's trying to DDOS us."

> "A hundred thousand requests a second? Neat - is it affecting our latency?"

"Um... no, not really."

> "Give me a shout if they hit a million a second."

As others have mentioned, the JS for the crypto is fetched from the server every time, which is a weak spot because the server could give you duff crypto code.

I've looked at similar things before, and I became very interested in Data URLs. Here's an example that PGP-encrypts files and uploads them to DropBox: https://bitbucket.org/geraintluff/encrypted-drop

The idea is that you can fit a very small implementation of SHA-256 as inline JavaScript in an HTML page. This means that you can produce a 1500-character Data URL which loads, SHA-256 verifies and executes the external resources needed to bootstrap your app.

I started a whole module loader based on this concept, so you could do fancier things like async/parallel loading, adding more verification methods (e.g. public key, once that module is loaded) and so on: https://bitbucket.org/geraintluff/caution.js

This isn't a technological change - they're just publicly stating that members of the community are being harassed (especially women), and that's not OK.

So it's not de-anonymisation, it's calling out bad behaviour that is often ignored (because it's nicer to pretend it's not an issue).

I can't +1 this enough.

As a mathematician, that article was a little embarrassing to wade through.

While there are valid criticisms of the way statistics are sometimes misused in science, pretty much every one of them comes from lack of understanding about how statistical models work - scientists reaching for a familiar test and following a formula they were taught. I can't blame them too much - understanding the true purpose and nature of statistical models is HARD (my recommended step one: become a Bayesian). What we need is for more people to recognise when they don't have that understanding and work with somebody who does.

What Briggs seems to have done, though, is decided that because HE doesn't understand statistical inference and modelling, that statistics are bunk. Taking a simplistic definition of "trend" like "the second-half average is higher than the first" and turning that into a boolean yes/no answer is the statistical equivalent of being an anti-vaccer.

The most frustrating thing, though, is that all the alternative definitions of "trend" he defines can actually be expressed as statistical models! The issue is that when you express these definitions/tests for "trend" as models, you see that the statements each model makes about the underlying system are very problematic.

TL;DR - Briggs doesn't understand statistical modelling, and has therefore concluded that his home-rolled tests are just as good.