pippellia
Pip the social graph guy

I have an idea of making s client that:
- shows you who followed/unfollowed you the past week/month...
- shows you the inactive or least interacted among your follows

2025-04-17 15:03:35

How to use Pagerank to filter out spam or gate access to a premium community.
https://vertexlab.io/blog/pagerank_as_filter/
(I won't make a nostr long form because there is no Latex)

2025-04-17 11:10:37

- reply

suggestion taken 🫡

2025-04-16 10:06:37

- reply

yeah, but Vertex doesn't crawl for all events, so I had to work with what I have.

2025-04-16 09:33:52

- reply

it's an heuristic. Not perfect but it's something

2025-04-15 07:36:25

- reply

@npub18lz…ugm3 if you are interested, I would be very happy to come on the show. Just sayiiing

2025-04-12 10:47:39

what's the easiest way to accept bitcoin/lightning payments on a website?

2025-04-10 13:59:11

- reply

wouldn't change anything, the bots will just do the same thing.

2025-04-10 13:01:03

- reply

I think it's a combination of these institutions being old monoliths + as you pointed out, that they avoid any potential liability.

2025-04-09 09:08:31

- reply

It's a bit of a shame that no one in the pod explained exactly how the 2-of-3 shamir secret sharing works, since it's so simple!
So, here is my contribution:
First, let's call S our secret. S is a number or can be represented as such.
We choose a random line that pass through the point (0,S). The lines with this property are the following:
y = a * x + S (where a is a coefficient)
So we choose a at random, and we have generated a line that pass through our secret S. If you know the line, you know the secret by simply substituting x = 0 into the equation.
Now, we want to split this knowledge into 3 pieces. The way to do it is to choose 3 more points on the line, different from (0,S). For example we choose:
P1 = (1, a+S)
P2 = (2, 2a+S)
P3 = (3, 3a+S)
(check that these points are on the line)
Now, how many lines go through two distinct points?
That's easy, only one! So if we know any 2 of these 3 points, we can reconstruct S.
Let's imagine we know P1 and P2.
Then, we can find S this way
2*P1 - P2 = 2*(1, a+S) - (2, 2a+S) = (2, 2a+2S) - (2, 2a+S)
= (2-2, 2a+2S-2a-S) = (0,S)
ta daaaa.
We have our secret back!
This is how a 2-of-3 SSS works. To increase the second number, you simply generate more points (P4, P5, ...). To increase the first number, you need to increase the degree of the polynomial.
wtf is a polynomial? The line we used:
y = a * x + S
is a polynomial, of the smallest degree, which is 1. The degree of a polynomial is the highest exponent of x.
This is a parabola for our secret.
y = a1 * x^2 + a2 * x + S
Note that there are two random coefficient now. With this parabola we can create 3-of-whatever SSS schemes for our secret S. This means we'll need at least 3 points to reconstruct it.
The number of points you need to reconstruct the secret is always the degree of the polynomial + 1. But why? This is left as an exercise for the reader.
https://en.wikipedia.org/wiki/Polynomial_interpolation

2025-04-09 08:07:52

no one in this podcast mentioned exactly how a 2-of-3 Shamir secret sharing works, which is a shame because it's super simple. Here is how:
First, the secret S is a number, or can be represented as such. Second, when you generate the shares for the secret S, you come up with a random line that pass throughout the point (0,S).
For example the line (a is random):
y = a*x + S
Then, you take three more points that are not the original point (0,S), for example:
P1 = (1, a+S)
P2 = (2,2a+S)
P3 = (3,3a+S)
(check this points are on the line)
How many lines go through 2 points? That's easy, only one. This means that with only two of these points, we can find the line, and then the secret.
For example, If we have the first two points P1 and P2, we can do:
2*P1 - P2 = (2-2, 2a+2S -2a -S)
= (0,S)
ta daaaa!
We've got the secret back. This is the basic for a 2-of-3 SSS.
To increase the second number, just generate more points at the shares creation.
To increase the number, you have to use polynomials of higher degrees, for example a parabola (3-of-whatever), a cubical (4-of-whatever), and so on.
https://fountain.fm/episode/gWYJ5PgxwzdnBjuNuzyc
#nevent1q…qwv9

2025-04-08 21:41:16

- reply

it's the nip05 reported in the kind:0

2025-04-08 13:29:15

- reply

different pubkeys

2025-04-08 10:32:10

- reply

One can easily use #rely to:
- maintain a whitelist of npubs
- ask the client to auth when connecting/when posting/ when requesting chats...
so yeah, basically a community, right?

2025-04-07 21:56:03

- reply

teach me this power master jedi

2025-04-07 07:39:01

- reply

well, I found many race conditions, and one weird bug where many clients unregistering all at once caused the blocking of all threads

2025-04-07 07:37:24

- reply

yes. random requests, random events, random filters, random disconnects...
I've found so many weird bugs, like the unregistration queue filling up causing all threads to be blocked

2025-04-06 22:29:01

- reply

noice. Btw recently I've learnt about random testing and I'm pretty proud of this
https://github.com/pippellia-btc/rely/tree/main/tests

2025-04-06 22:26:20

- reply

hey guys, maybe you can improve your processes by using Vertex and npub.world

2025-04-06 21:52:14

Search and pay is the ultimate paymente UX.
#nevent1q…pxfv

2025-04-04 21:29:35

- reply

yeah go is amazing, and one can really appreciate how much thought went into it by trying to build stuff and realised bit by bit why things are the way they are

2025-04-04 19:23:30

- reply

mmm, you gave me some things to think about. Thanks!

2025-04-04 18:38:36

- reply

yestr, collaboration and pollination of ideas is how we win

2025-04-04 16:11:07

- reply

I think you'll find it very simple :)
Let me know any feedback sir

2025-04-04 16:04:07

- reply

suggestion taken

2025-04-04 16:01:00

- reply

@nprofile…fngn I think you are conflating khatru and rely. When I say that rely took inspiration from khatru, I mean that I copied the ease of use (users supply their rely functions), but the inner working is completely different. completely.

2025-04-04 15:58:49

- reply

> i don't even understand how khatru based relays are still functioning the excessive amount of threading they use
yeah I know, I've been running Vertex on khatru for few months. Rely uses only 2 go-routines per websocket, one for reading, and one for writing. Then, one goroutine for dealing with client registration, unregistration and requests.
> also, yeah, not to forget, i totally flattened the concurrency of the event store, so, it also is a solution for that, excessive amounts of threads in each query were a component of the resource leak problem i found with khatru/eventstore design
Oh yeah, that part is gone in my design. Users supply their query method and that's it.
In fact, all your cool DB stuff can be used in the rely.

2025-04-04 15:52:49

- reply

fuck u 😂, it's a great name

2025-04-04 15:10:03

- reply

thanks! Well, for now REQs have multiple filters, so I'll keep it the way it is until the PR gets merged

2025-04-04 15:02:49

- reply

> relayer has better architecture and realy was forked from that base
can I ask why it's better?
> i don't like go-nostr at all, though i'm not saying that realy's libraries are exactly perfect
Yeah, I feel similarly about go-nostr. I am considering switching. Why don't you extract the library from realy, making its own repo? Other people might use it in that case
I'll take a closer look at realy sir. The first time I was just scared away by how big the repo was.

2025-04-04 14:52:22

- reply

really? where did you read it?
NIP01 still shows multiple filters per REQ

2025-04-04 14:33:54

- reply

@nprofile…fngn would love to know what you think

2025-04-04 14:25:07

💫 Introducing rely
A framework for building super custom relays you can rely on. Designed to be simple and stable. About 1000 lines of code.
https://m.primal.net/QECM.mp4

2025-04-01 22:10:58

- reply

followers can be faked easily, unless good filters are used.

2025-04-01 14:02:08

- reply

depends on what you are testing

2025-04-01 08:58:34

- reply

yes exactly, but the community already exists in the social graph. 99% of people will be invited to communities, won't be their founders. Founders will need to acquire reputation themselves, but these are the type of people that aren't afraid of doing the work.

2025-04-01 08:33:42

Try using npub.world. 99% spammers free
#nevent1q…lv2m

2025-03-31 12:02:02

- reply

Even though I am a bit skeptical about super apps in general, I wish you god-speed

2025-03-30 12:29:25

- reply

that's cool. I was thinking more about weekly report who followed/unfollowed you + recovery

2025-03-30 10:14:16

others interested? Might code it up
#nevent1q…du3c

2025-03-30 10:12:09

- reply

Last week one of my tests failed because I was using an old graph and you stopped following @nprofile…ucf3 so the response was inconsistent.

2025-03-30 09:45:23

- reply

grazie Daniele :)

2025-03-30 09:44:42

- reply

no idea, sorry

2025-03-30 09:30:10

- reply

We don't discriminat non-human npubs 🤪
If people follow it, then they probably find it valuable

2025-03-30 09:28:55

- reply

Depends on your definition of active.
Active could mean they changed their kind3, or they posted some kind1s. Anyway, at the moment no

2025-03-29 08:23:29

- reply

pretty brutal yes. There are ~300k reputable pubkeys in total

2025-03-28 18:45:42

- reply

If you are interested @nprofile…5czt I can point to you where that is in the code

2025-03-28 18:44:57

- reply

This algorithm.
https://en.wikipedia.org/wiki/PageRank
We use a slight variation but it's basically identical.

2025-03-28 18:44:08

- reply

An interesting idea to improve it is to add recommendation DVMs (like this one https://vertexlab.io/docs/nips/search-profiles-dvm/), so people can start with a high-quality feed.

2025-03-28 18:39:31

- reply

yes, keep the feedback coming guys, I'll consider it all :)

2025-03-28 18:38:10

- reply

Hey, thanks miljian!
For context I'll add here a response I just wrote. It's a bit convoluted I know, I should probably make a video explainer.
nevent1qvzqqqqqqypzpa5rapcrtaadfazwpwvvl0v4xlskg4df9nfcem7yevcaka2h7hhjqyt8wumn8ghj7mtvv44h2tnjv4skc7fwd3hkctcpzamhxue69uhhyetvv9ujumn0wd68ytnzv9hxgtcqyqscjdyme00uaezmgs3d4cg6p2eezfj365xm8qqeuy7gfvkm9jhcyupgfpd
> how exactly does somebody make the cut of being reputable? It has to be followed by at least one reputable
> how many real humans do you think you are missing in your reputable follower counts?
No idea tbh, but these users haven't gained a single reputable follower, which probably means they haven't posted anything / met anyone in meatspace that is already on Nostr.
One can find these people by their npubs, but not using the search DVM. This is a very opinionated decision that I think will make more and more sense in the future: I choose to crawl what people find interesting/valuable. and nothing else.

2025-03-28 18:22:07

- reply

> What’s a reputable follow?
great question!
some definitions related to our graph database:
1. a reputable node is one that passed our internal filters, and thus was added to the graph (point 4.)
2. an active node is one which contributes to the global Pagerank
3. a node get's promoted from inactive to active if it's Pagerank score surpasses a certain threshold.
4. to be added to the graph as an inactive node, one has to be followed by at least one active node
At the moment there are about 100k active nodes, and 300k reputable nodes.

2025-03-28 08:41:58

- reply

Looking forward to seeing it.

2025-03-28 08:13:38

- reply

> there is no point in doing sliding window spidering on events once you have the history, from then on just open a subscriptiion
I am not following sorry. My crawler does this:
- Firehose: subscribes for new kind3s and kind0s
When a pubkey that is in the database is "promoted" (has acquired enough reputation)
- Query: query directly for its own kind3s and kind0s.
The query starts when one of these two conditions is met:
- One minute has passed since the last (with at least one pubkey)
- There are 100 pubkeys that have been promoted
Normally the crawler in production promoted ~100 pubkeys/day.
However, when I do tests locally I re-do the entire crawl which obviously is much more intensive, but I'll make sure I'm not using your relay for that sir.

2025-03-27 09:29:47

- reply

this is amazing!

2025-03-26 15:03:06

Nostr is home to a lot of spam,
but now there is an effective way to deal with it.
https://m.primal.net/PtrC.mp4

2025-03-26 11:05:54

@nprofile…y38p SortProfiles (kind 5314) is already the #2 most used DVM.
And this is just the beginning
https://m.primal.net/Pteu.png

2025-03-26 07:31:28

- reply

yeah, that would be a nice add.
About 300k reputable pubkeys. Yeah I know, super low

2025-03-24 14:26:24

- reply

2, but it looks scary which is not great imo. (because of cold colours and the "scary" to me bolt).

2025-03-24 10:35:22

- reply

just found this note. You can use pagerank-powered search at https://npub.world/

2025-03-22 23:48:22

- reply

I'm confused and amused at the same time

2025-03-21 15:43:41

- reply

thanks sir

2025-03-21 15:43:27

- reply

no one is using nostr, YET

2025-03-21 11:17:17

- reply

wow, that was quick!

2025-03-21 11:16:37

Follower counts that make sense, live at npub.world 🌍
https://m.primal.net/PniX.mp4

2025-03-20 11:08:18

- reply

Although I understand the data saving of CRDTs, I think they can introduce a lot of pain points, so I have a slight preference for the current grok brain approach of big lists.

2025-03-20 10:58:10

- reply

I get the benefits of CRDTs, but it can create problems when some event in the change of diffs is not to be found. How would you deal with that case?

2025-03-20 08:03:49

- reply

It's not an API, but a search DVM. In practice you use them the same.
vertexlab.io
If you want to see it in practice, you can go to npub.world

2025-03-18 09:52:59

- reply

yes, that is true. Trusted introductions are a way to solve for this.
Invite someone into nostr and you start following him right away, or invite him into a community and that is already a signal if the community is reputable.

2025-03-18 08:26:32

@nprofile…jku2 fixes this
#nevent1q…cxrf

2025-03-14 09:45:46

I must give a massive shout out to @nprofile…pvsl and @nprofile…rp63 for designing and implementing npub.world's new design.
https://m.primal.net/PfSy.mp4

2025-03-13 12:33:59

- reply

I think a warning when last commit is older than 2 months or so.

2025-03-13 11:53:25

- reply

yeah, and for my math blog

2025-03-13 11:08:17

- reply

love there is latex

2025-03-13 08:26:54

- reply

awww, thank I bitch

2025-03-12 11:09:20

- reply

data vending machine NIP90. It's really very simple.
There is one event called request, that specifies what job you want to get done, and how.
And there is another event called response, where the result is provided.
If there are issues, a DVM error/fedback event is returned instead.

2025-03-11 14:40:34

- reply

It is a relay, yes, however the output is a new event kind which depends on the inputs of the request.
Yes, the same can technically be achieved by abusing a REQ filter, buuut it's different enough from the normal use that I am okay in using a new thing.

2025-03-11 14:21:00

- reply

it can be improved a lot by not downloading a bunch of events in the first place.
By using DVMs this burden can be moved to service providers, which benefit from better infra and economies of scale.
An example is using Vertex vs downloading 1000 kind:3s to compute WoT things.

2025-03-10 15:15:17

- reply

is that your real voice? If so, how didn't you start a podcast before 😄

2025-03-10 09:33:45

your ads are the best 🤣
https://fountain.fm/episode/nRt6WJSGWzQlWY7smAu5
#nevent1q…37hk

2025-03-06 08:37:02

- reply

lol we live in a simulation
https://m.primal.net/PVZy.jpg

2025-03-04 16:35:25

Someone is sending thousands of kind 5314 requests to @nprofile…y38p .
The relay is working like a charm, but nevertheless soon we are going to be more aggressive with rate-limiting.
Just letting you know:
YOU CAN JUST DM US, I KNOW IT'S SOUNDS CRAZY BUT IF YOU LIKE SOMETHING YOU CAN JUST ASK THE PEOPLE BEHIND IT AND TALK ABOUT THE PRICE.

2025-03-03 15:17:50

- reply

I am working on it. In the next refactor we'll parse only tags that follow the scheme:
"param", <key>, <value>
Where <key> is one of our parameters.

2025-03-01 10:19:53

- reply

yes this is something we definitely want to explore in the future.
One idea, specific for recommend follows, is to provide some kind3s (or their IDs), so the user can verify client side if and how he's connected to all the recommended npubs. We are talking about 2N +1 kind3s as the worst case where N is the number of recommendations.

2025-02-28 18:53:28

- reply

stunning

2025-02-28 18:52:23

- reply

yeah true. But on the other side, asking the user for approval and making him pay can be taking away from the experience. Maybe clients can pre-pay for new users, or all users, and monetizing somewhere else if they wish so.

2025-02-28 18:08:02

- reply

wooooow. GG guys

2025-02-28 18:07:22

- reply

we can provide the same service in other ways, but a DVM is very straightforward to integrate and it has the benefits of being a native nostr event

2025-02-28 18:06:22

- reply

this is a very interesting question. @nprofile…rp63 and I were so busy building that we haven't though out the payment aspect yet 😅

2025-02-28 18:05:08

- reply

Thanks mate!

2025-02-28 18:02:32

- reply

as Fran pointed our already, we already have a suite of services using our algos in multiple ways!
https://vertexlab.io/docs/nips/
Perhaps we can in the future start indexing and analyzing e-cash mints rating.

2025-02-28 17:59:00

- reply

I think coracle performs search client side, right @nprofile…92wl ?
So comparing it in speed and "breath" would be unfair.
https://m.primal.net/POth.mp4

2025-02-28 17:54:56

- reply

😎 🙏

2025-02-28 16:29:57

- reply

yeah I know, our Firehose is inspired by yours. Not the same because we want to aggressively filter spammers, and we use internal metrics for that.

2025-02-28 16:26:15

- reply

and thank you :)

2025-02-28 16:26:01

- reply

we are doing all ourselves.

2025-02-28 16:20:38

- reply

that's coming soon

2025-02-28 15:37:59

- reply

ahhhh, I'll add it so we can have some more commies indexed 😂

2025-02-28 15:37:21

- reply

ah dude, we have a daily limit of 30 free DVM requests per day. We should probably have a chat on how to extend that for olas users. cc @nprofile…rp63