comntr

That's what I'm trying to test. Last time I've got 800 mostly meaningless comments.

The idea is to stop numerous zergs: you can't ban them one by one, but a little bump on their way can do the job.

To avoid annoying more serious commenters, the captcha should be as simple as it can be. So asking to add 321+625 would be too much, while a question like 2+4 doesn't need any thinking, so I chose two-digit numbers.

This is a small social experiment. I don't know how it'll go.

The captcha is basically a question to add two 2-digit numbers, e.g. 23+45. The question is returned as an SVG picture.

The hypothesis is that the need to engage brain even for 3 seconds is enough to stop 99% of the spammers, so the quality of comments should shoot through the roof.

The way it works is each comment has a sha1 hash and from that hash we can derive the 2 numbers. No state needed.

Sources of the service: https://github.com/comntr/captcha

And just rendering to canvas may be very tricky if the captcha is animated with css, i.e. it moves a bit and different parts of it appear at different times.

The catch is that the text will be represented as small geometric svg shapes, so the spammer will need to first render the svg to png and then run text recognition tools. But in that svg we can easily add some css animations that make sure the entire image is never rendered, so spammers will need to run the entire browser to take screenshots and will need to assemble the image from multiple frames.

That's true when we talk about 10M monthly pageviews, but I doubt that this little extension will reach such popularity levels. If this somehow happens, by that time there will be a way to enable 3rd party captchas for any page.

Yes, this is how it works:

comntr.github.io#http://foobar.com/

I think, "comntr" is more like "integrated 3rd party" that can be run like a self hosted solution.

Can you elaborate on those one word answer questions?

That's right, I haven't.

The puzzle will be sent as an SVG, obviously.

Em.. "off-the-shelf OCR" sounds neat, but anyone who knows such words isn't an average spammer. The goal of basic SVG puzzles is to block 99% of the spammers who just type dumb comments on keyboards. The rest 1% can be taken care of by human mods.

TBH, I don't like the reCAPTCHA-like solutions. They are just annoying from my personal experience and if they rely on any 3rd party service, I'll give them a hard pass for this reason alone. My approach is to use trivial SVG-style captchas with adjustable complexity, e.g. instead of asking "23+34", we can ask "log(32)/log(2)" and effectively filter out everyone except people familiar with math, or "md5(2615), first 7 hex digits" and let in only people familiar with cryptography. Forcing users to detect birds and crosswalks will just make them upset, IMHO.

Done: https://github.com/comntr/comntr.github.io/commit/af35

You can git clone and launch your web client too.

Or rather time-delay moderation, as it's easier to implement. Comments are added to the server as usual, but the web client shows them only after 1 hour.

I think most of the spammers can be deterred by a simple puzzle, like 23+47. If we want to raise the bar, we make the puzzle more and more complex. Obviously, the puzzle is returned as an svg picture where the letters are "rendered" with little squares. My point is that 99% of the spammers out there are lazy and won't be able to pass this little test vs someone who's written a thoughtful comment and can definitely add 23 to 47.

That's right. My first attempt was to use IPFS or DAT. Figured out it's not quite possible, but we can get very close to that, in theory. Imagine the extension or the iframe could run a ipfs.js or dat.js that would discover all the http servers with comments via DHT: servers that want to participate, publish a unique key to the DHT and the web clients discover this key and then the IP addresses of the servers. In practice, this doesn't quite work because DHTs are based on the assumption that any node can quickly ping (with a UDP packet) any other node and thus perform the DHT discovery using the Kademlia algorithm in log(N) steps. But in the web, the only way to "ping" someone is to set up a p2p connection with WebRTC: this not only needs a signaling relay, but also implies a multi step exchange with SDPs and has other costly overhead. And I haven't even approached the Symmetric NAT problem. This is why ipfs.js hogs CPU, allocates a 1 GB and keeps 4-8 sockets always open (they aren't even p2p now, but rather web sockets to some relay, for perf reasons).

Yea, I missed the obvious one!

Also, it's already possible to have your own comntr server: you just need to git clone the comntr/http-server repo, npm install & npm start it, and tell the iframe to use your server with ?srv=https://foobar.com:42751. This would be an "on-prem" solution.

My first post about the web extension idea got some interest (and almost 150 stars on github!), so I've made the next logical step: an <iframe> that renders the comntr.github.io page and effectively adds comments to your page. The spam problem is partially addressed by the filters: the iframe.src can have a special ?filter=[...] param that can hide some of the comments you don't like. Although it's unlikely one can bypass the security model around those filters (but you are welcome to try!), advanced spammers can still post a lot of garbage to those comments as they can easily generate new ed25519 keys.

Easily. Open source merely means that the source can be seen by anyone. It doesn't mean that readers of the source have any say in how to run the browser.

I'd expect some ML stuff, like training some models, etc. Such comments in raw form hardly have much value. What adtech usually pays for is data pinned to user ids.

Comments can be seen like this: https://comntr.github.io#http://contoso.com/ The trick is that the #... part of the URL isn't sent to the server. This request downloads JS which computes SHA1 and gets comments by that hash from the data server.

The space of possible reasonable URLs is way too big to brute force. If your point is that someone can just grab the set of existing HN URLs and get their heashes, then I don't see what this achieves. Someone who knows the URLs can just download all the comments for them anyway. In other words, is there an example where a sophisticated state of the art hash has advantage over MD5 in our case?

Looking at comments for example.com, I understand why they want human moderators.

But if you click on it, it will have access to all your data? There is no in between: either no access until you click or full access when you click.

This can be done already:

javascript:location.href='https://comntr.github.io/#'+location.href

Ah, right, I forgot to add basic html escaping!

Comments under example.com already give a good idea what it will look like when everyone will start posting comments in public places.

Not right now. But I'm hoping to come up with an equivalent of captcha that doesn't need 3rd parties. That way any comment will need a few seconds of the commenter's time (not CPU time).

Isn't bookmarklet just a script that runs on the context of the current page when you click on that bookmark? If so, the script has all the access to your data. It can steal your auth cookie or some keys from the local storage and send them to some server.

Yes, they can be.