Patrick Boueri
@patrickboueri
RecSys/ML. Chicago Architecture boat tours on the side. Lebanese American. Lots of opinions, reskeets are reskeets. Sporadic personal site: Building try it out at
Ok it will actually take quite a bit of time. It’s a really thorough study! The core data collection is 28k survey results of which 15k were thrown out. It does give a good glimpse into awareness and is probably really good source material for @mu.social and @bsky.app
I think the moat will be trust, customer service, and community. we are getting closer to thing that matters - the craft was just a way to get there
All nicely packaged like this and everything. There's your data model
Used to live right next door: they always fed the pigeons and bless them for it.
@attie.ai they say copying is the sincerest form of flattery - but also reads a little scared to me too
If @localfirstconf.com is successful this opinion will seem quaint- and that it will very much matter where data is stored. https://www.supremecourt.gov/opinions/25pdf/25-112_0am4.pdf
When is the new Ronan Park opening? Just like it’s been done for a minute.
Shocking - www.anthropic.com/research/eco... Claude code usage is higher on the weekends?!? Talk about latent demand.
It was a heavy import on an idle machine with a lazy network filesystem -- yikes. This fixed it
Love a good viz - numbers are reductive and you have to watch the game to know what happened but also look at the flatline in Q4
Interesting i did not know about 2 -- i see it's in the playground! Thanks for exposing all the knobs foryou.club/playground?u...
I built something like this but i wound up needing way beefier machines to make it useful so I stopped indexing posts.
Feed Local is this year's healthiest diet. Very excited to see where this leads
Im pretty sure this is because google ai overviews… which is also not great
I know there are other things to worry about - but they count the height of the radio antenna for the empire state but not sears tower. This feels like a Chicago snub and i dont like it. Sears tower was tallest in world from 1973 to 1998
@nbartley.bsky.social A LOTTTT of short hash name accounts -- something to filter
Here's the inverse @enflux.bsky.social 70% of accounts have followed fewer than 10 accounts since 2025 and 40% are functionally dormant / inert. I think it means I have to rerun the analysis and purge these inert accounts, and dive into who they are. I'll have the time to do so in a couple days
It’s not really bimodal but its a huge problem 70% of accounts have less than 10 follows. I think the key to understanding what is going on is looking at those accounts. Are they bots - or people who got not traction?
Made me feel warm and fuzzy Shouldn’t a new follow also be active? Instead of any old follow. Presumably theres a lot of overlap. It’s weird people are churning so much im gonna randomly sample and see whats up
One thing i want to put more time into is projecting out DAUs or WAUs based on current rates and see where we wind up -- then create a sensitivity analysis on top of it to see where the biggest changes can be driven from. For example if you 2x retention from 5% to 10% the user base will grow 50%
One pet theory of mine is that people post and get no engagement - so why would they stick around. Most signups basically do nothing. But of those that post -- earning a like makes a big difference. Or maybe feed quality is too low to start and you churn It's bad lots of sign ups dont follow ppl
There IS a committed base of highly active users which is why the place feels so lively -- but I personally would love to see more growth because it just makes everything else easier. I know I know -- growth for growths sake is dumb. But if the network grows we have more choices and paths ahead
So I've been digging into bluesky retention data to see what is preventing growth of the ecosystem. Using likes/posts/follows you can see who is active or dormant on ATProto. The headline is that only 5% of users are retained at 28d and 10% of current users churn. That's not pretty