As for the latter comment....not sure what your implication is. Yes, bot/spam mitigation is whackamole, but there are consequences for not playing the game of whackamole. Luckily residential proxies are few and far between so far, but they will grow in popularity. When they do, and I can't get by with the occasional individual residential IP ban, we'll come up with other methods to handle.

Luckily the signal is strong with vulnerability scanning, which makes it pretty easy to automate. The only reason to put up whole ASN mitigation (captcha/turnstile, outright bans) is just efficiency. Nothing stopping individual IP banning. The scrapers are the tricky ones, since they more easily hide in legit traffic. However legit traffic has patterns that scrapers do not emulate (at least for a service like ours with millions of pieces of user generated content that's easily walkable), so you can still pull out the signal. It's just a little trickier.

Definitely a continual arms race though.

We collect signals that help determine good vs bad networks. For example, large amounts of requests to .php endpoints, large amounts of empty accounts from the same /24 subnet, etc etc. All these signals let us automatically determine risk, and then put up a challenge. Authenticated users never see the challenge even if they are on a risky network (VPN 99.9% of the time), unless the network has been identified as 100% malicious, then it gets a full block.

Here's a small snapshot of the dashboard:

https://cos.ridewithgps.com/screenshots/6a7c54d0-12Aug26-358...

This was probably a total of 3-4 days of work, spread out over a couple months of iterative claude led hacking. I didn't know exactly what to build, but had some of the key architectural ideas in my head. Opus+Faable made easy work of it all, and ended up guiding some really slick improvements for performance.

I would say this has dropped about 20% of all traffic to our service, though it turns out turnstile is a massive target for bots, so replacing that with something custom is next on the list.

It helps that we have something closer to a lifestyle business, where I can ask for a brief paragraph about your relationship to the outdoors and cycling, but that just means I had 500 slop cover letters gushing about cycling. The three that made it through were short concise honest and linked to real world activities they did.

Good luck, it’s a hard problem , and very very adversarial. You have true scam level applications from North Korea and India, and you have unqualified people trying to appear qualified. Sprinkled in are unqualified people who would be a good hire because of raw capability, and qualified people who are looking to do bare minimum.

Instead, I hittup ebay, got six used gen3 processors, found a "good deal" on a couple tb of new ram (still insanely expensive), and came out with the same overall horsepower for a total of $20k instead of $110k.

I know this is about consumer desktop, but seeing the comments about upgrading old hardware caused me to chime in. This is happening in the production/enterprise level in some segments.

If you have a logging stack, you can easily find crawler/bot patterns, then flag candidate IP subnets for blocking.

It's definitely whackamole though. We are experimenting with blocking based on risk databases, which run between $2k and $10k a year depending on provider. These map IP ranges to booleans like is_vpn, is_tor, etc, and also contain ASN information. Slightly suspicious crawling behavior or keyword flagging combined with a hit in that DB, and you have a high confidence block.

All this stuff is now easy to homeroll with claude. Before it would have been a major PITA.

Instead of using integers IDs for the hexes, we created an encoded version of the ID that has the property that removing a character gets you the containing parent of the cell. This means we can do basic containment queries by querying with a low resolution hex (short string) as a prefix query. If a gps track goes through this larger parent cell, the track will have hexes with the same prefix. You don’t get perfect control of distances because hexes have varying diameters (or rather the approximation, since they aren’t circles they are hexes), but in practice and at scale for a product that doesn’t require high precision, it’s very effective.

I think at the end of this year we’ll have about 6tb of these hex sets in a four node 8 process ES cluster. Performance is pretty good. Also acts as our full text search. Half the time we want a geo search we also want keyword / filtering / etc on the metadata of these trips.

Pretty fun system to build, and the concept works with a wide variety of data stores. Felt like a total hack job but it has stood the test of time.

Thanks uber, h3 is a great library!

IMHO git bisect is slower, especially depending on the reload/hot-reload/compile/whatever process your actual app is using.

https://github.com/kingcu/screendrop

this is a common problem in any complex codebase that has a culture of using third party dependencies to solve small problems. you see this conversation all the time with modern frontend development and the resulting dependency tree you get with npm etc....