straycat
david

neurologist and freedom tech maxi
Co-Founder @ NosFabrica
🍇 Grapevine, 🧠⚡️Brainstorm, Tapestry

Public Key

npub1u5njm6g5h5cpw4wy8xugu62e5s7f6fnysv0sj0z3a8rengt2zqhsxrldq3

NIP-05 Address

Profile Code

nprofile1qqsw2feday2t6vqh2hzrnwywd9v6g0yayejgx8cf83g7n3ue594pqtcpz3mhxue69uhhyetvv9ujuerpd46hxtnfduqs6amnwvaz7tmwdaejumr0ds9d69cw

Publishing to

relay.damus.io

nos.lol

Author Public Key

npub1u5njm6g5h5cpw4wy8xugu62e5s7f6fnysv0sj0z3a8rengt2zqhsxrldq3

Show more details

Last Notes

2026-04-11 00:28:59 UTC

- reply

Yup, for now. On our roadmap is to incorporate more signals.

2026-04-10 16:42:30 UTC

- reply

Correction: if fakes/spammers/bots/bad actors follow you, they don’t drag your score down; they have no effect whatsoever. Their follows are simply ignored.

2026-04-09 20:44:29 UTC

- reply

To integrate this into coracle or flotilla or any client, see https://brainstorm.world/developers. You'll be querying a relay with the search string and observer pubkey using NIP-50, not an api endpoint.
New users will by default be shown the "global" point of view (called the "House PoV"), which in this case means the PoV of NosFabrica. From a practical perspective, the global view will work perfectly well the vast majority of the time for the vast majority of users.
It's open source so you can run your own instance and define House (global) as hodlbod or flotilla or coracle or whatever pubkey you want.
To personalize scores, right now it's a one-time setup of 20-30 minutes but we're working on getting that down by at least an order of magnitude. And once it's set up, its' faster and more performant than whatever alternative you may be thinking about.

2026-04-09 20:08:16 UTC

- reply

Are you trying to decide between personalized WoT scores (coracle was one of the first to implement these) versus global PageRank via Vertex? What about personalized PageRank? I know (if I am not mistaken) that Vertex offers personal and global PR although who is going to pay for it in the long run is to me an unanswered question.

2026-04-09 06:01:47 UTC

- reply

Last I checked (last year, on nostr.band) there were ~ 45 million pubkeys. But I don’t have 45 million trusted assertions. I only keep as many TAs as I want, no more, which for now is on the order of 100k. One (of multiple) advantages is speed. For any user who doesn’t have a TA from my personal PoV, if I need a score, I calculate it after I know I need it. Alternatively, I pull the TA from the PoV of someone I trust, if the option exists; and it will be relatively easy to find one if so. This is the kind of solution that will be highly effective at reducing the number of TAs by many orders of magnitude.

2026-04-08 19:00:01 UTC

- reply

What are you going to do hodlbod when the state comes to you and tells you that @npub1kpt…nhmz is not on their list of approved WoT score providers and you need to exclude the vertex endpoint from your client? With NIP-85, the state won’t ask you to do that bc you can’t. It doesn’t work that way.

2026-04-08 18:49:29 UTC

- reply

Eventually we will be supporting the kind 0 solution in appendix 1 of nip-85. For now, the short list includes
Brainstorm.nosfabrica.com and others including @nprofile…zvyu — I’ll invite other SPs from the WoTathon chat group to advertise themselves in this thread.

2026-03-31 16:01:08 UTC

- reply

These are still processed manually, and I’ve been remiss in keeping up with it so haven’t triggered your scores yet.
But I’ll DM you the link to the new brainstorm version which triggers automatically. (Tell me if you get the DM, since DMs have really been hit or miss for a while)

2026-03-28 20:11:57 UTC

- reply

I hope so! 😂
I created an issue to add this as a new feature on our brainstorm server. Feel free to add comments or suggestions.
https://github.com/NosFabrica/brainstorm_server/issues/8

2026-03-28 18:21:03 UTC

- reply

I think there’s a good chance we’re going to do it — we just haven’t yet bc we’ve been more focused on making sure the main product is stable.
Has anyone else implemented appendix 1 yet, enough for you to have something to play with?

2026-03-28 14:28:35 UTC

- reply

Every Brainstorm user gets a personalized pubkey, like your own personalized Brainstorm Assistant, which publishes scores on your behalf. That’s the pubkey that’s in your 10040 personal preferences event.
We don’t currently publish a kind 0 profile for this pubkey but it’s on our roadmap. Probably as an opt-in feature. NIP-85 was recently updated to support WoT service provider discoverability guidance — see Appendix 1 of the merged NIP.
Suppose you don’t have a WoT SP and don’t know which ones are good. Following the recent NIP-85 update, you will be able to look through your follows, find who has a 10040 event, pull the WoT assistant pubkey for the “rank” metric (or perhaps for all available metrics), pull that pubkey’s kind 0, and extract the WoT SP website. You’d get a report that says something like “Looking for a WoT SP? N1 of your follows use WoT SP A at www.wot-sp-a.com, N2 use SP B, etc.”

2026-03-26 14:12:47 UTC

Less than an hour away!
#nevent1q…24p0

2026-03-24 04:45:13 UTC

- reply

Could do what @nprofile…54jr is doing: pull the whitelist of 100k+ pubkeys from Brainstorm and use that not only to delete old events but also to filter new events. I’d be curious to know how many GB of data that would prune away. Might want to use some more permissive settings to prune less aggressively.

2026-03-21 01:35:44 UTC

- reply

I had previously signed up at brainstorm.nosfabrica.com to have WoT scores calculated. Nostria was able to communicate with the brainstorm server directly and trigger recalculation.

2026-03-20 23:54:05 UTC

I just triggered recalculation of my personalized web of trust directly from the nostria.app/settings WoT page 🤯
And I can see the same scores in @npub16x7…a44x that I see in @npub142g…xrj0. Which means my WoT is not only personalized … it’s PORTABLE. 🤯🤯🤯
👏🏻@npub1zl3…jajh 👏🏻

2026-03-19 18:06:12 UTC

- reply

At @nprofile…59ml we are working on a tool that will be useful in exactly this scenario. Which, as we all know, is going to be a recurring problem!
Currently, the compromised account has a high WoT score using any of the currently-available WoT scoring algos. But GrapeRank has a remedy: NIP-56 reports. The GrapeRank score is currently visible to anyone who uses Brainstorm (still in alpha) on @nprofile…3a29 , @nprofile…hnuv , and any client that uses NIP-85 Trusted Assertions. A sufficient number of NIP-56 reports from trusted accounts will drive his GrapeRank score from 100 to 0.
We are currently tweaking the default GrapeRank parameters that determine how many NIP-56 reports it will take to counteract the almost 2k trusted follows. If you would like to assist us in this process, then issue a NIP-56 report (report type can be other, something bespoke like “compromised”, or whatever you think is appropriate) of the compromised account, which is this one: @nprofile…h8le
#WoT FTW!
🍇 🧠 ⚡️

2026-03-19 14:52:01 UTC

- reply

Correction: 11 am eastern, 3 pm UTC!!! (I forgot to take the recent time change into account)!

2026-03-19 14:21:42 UTC

Weekly #wotathon community call is less than an hour away! 😃
#nevent1q…yld4

2026-03-17 12:34:53 UTC

Reminder that our 18th #wotathon weekly call is 2 days away on Thursday, March 19, at the usual time: 4 PM UTC, 11 AM Eastern. We look forward to hearing updates on the progress each team is making!
Any specific topics you'd like to add to the docket? Reply below!
https://nosfabrica.com/wotathon

2026-03-12 13:27:07 UTC

#wotathon call starts in an hour and a half! Come one, come all! 💜 🫂
#nevent1q…ldlf

2026-03-11 15:20:54 UTC

One day away!
#nevent1q…er22

2026-03-11 06:27:38 UTC

@npub1gxd…5zfk - here (below) were my thoughts typed up immediately after you presented at last week’s #wotathon community call. Very important problem you’re focused on. I posted this to our WoT signal chat but then learned you’re not on it!
👇🏻👇🏻👇🏻
Sounds like nostr needs you to build a WoT Service Provider, but instead of publishing personalized scores of trusted pubkeys, you’ll publish personalized lists of trusted relays. As a user I want to sign up and have my relay lists managed by my community so I don’t have to have a PhD in relay management. Here’s the challenge: how do these lists of trusted relays get communicated from your service to nostr clients in a way so that clients can and will use them? NIP-85 Trusted Assertions would be one possible solution. Or maybe Trusted Lists, so you can publish lists of relays for me. As a user I won’t even need to know how many relays lists there are, if I don’t want to. But for this to work, you’ll need clients to integrate your trusted relay lists. To me, that is your first challenge.

2026-03-07 06:06:59 UTC

copyleft ded?
#nevent1q…xfpp

2026-03-05 21:30:04 UTC

- reply

I’m using p tags on my brainstorm concept graph prototype (I’ve gotten a year’s worth of work done on it in the past week) but we’re not using p tags yet on @npub1hea…vp8p Brainstorm. Eventually probably yes but not yet.
Mleku you should listen to the recording of today’s WoTathon community call once it’s posted at https://nosfabrica.com/wotathon. @npub10mt…27hf gave us a detailed overview of the neo4j data model of his Banyan nostr relay which uses tag nodes similarly to Orly.

2026-03-05 19:00:49 UTC

- reply

In my mind, it’s Orwellian when some centralized entity (the CCP, some big tech company, whatever) is telling us what to pay attention to under a system where we 1) don’t know how they decide what to feed us (or can only guess, at best) and 2) can’t get out from under their thumb. For whatever reason.
Consider that fact that even after we filter out the really obvious spam, there’s still a lot of content in the world that is non-spammy but nevertheless uninteresting to me. How do I find the stuff that’s interesting to me? There’s too much for me to comb through all of it by myself.
So that’s what WoT is for: not just to get rid of the spammy stuff, but also to help me find the specific stuff that I want. If I travel to some city on the opposite side of the world and want a list of local chiropractors, I want my extended trusted community to give me that list — not Yelp, not Google, not Grok, not the CCP. And if it’s far enough away, my follows probably don’t have the info I’m looking for. Their follows probably don’t either. But go enough hops away, in a way so I can filter out the bots and bad actors, there will be people who can answer the question I want to be answered.
If my only options are centralized entities using algos that are opaque to me (the status quo) then that’s Orwellian. If my trusted community can help me focus my attention where I want in a way that allows me to route around any of these big centralized entities, where I have ultimate control over how the routing works (I can route by hand or I can delegating the routing process to members of my trusted community), then it’s decentralized WoT, which is the opposite of Orwellian.

2026-03-05 16:00:31 UTC

We’re live!
#nevent1q…2wzm

2026-03-05 14:59:19 UTC

We start in one hour! 😃
#nevent1q…2wzm

2026-03-05 06:47:30 UTC

- reply

It’s going well and we’re having fun! Lots of very cool projects and teams still have over a month of building left 😄

2026-03-04 17:32:08 UTC

Less than a day away!
#nevent1q…3xcr

2026-02-26 16:04:08 UTC

We’re live!
#nevent1q…y75d

2026-02-26 15:29:33 UTC

- reply

Your scores are all calculated using updated follows/mutes/reports and published as TAs. You should be able to see scores on amethyst and nostria. Let me know if it works!

2026-02-26 15:15:04 UTC

45 minutes!
#nevent1q…y75d

2026-02-26 03:30:09 UTC

Tomorrow!!
#nevent1q…hz2r

2026-02-24 19:32:30 UTC

- reply

Methods like this can be useful, if they are sufficiently user-friendly that people actually use them. The main question that always comes to my mind: what’s the remedy if / when the passphrase itself becomes compromised?

2026-02-24 00:18:33 UTC

- reply

It is still on manual — I’ll resync the database and see if I can get it to work for you. (May take a day or two! Gotta run a script that i haven’t run in a while so gotta make sure it’s not broken)

2026-02-23 19:14:59 UTC

- reply

Who would want to harass people off nostr, and why? for the lulz?

2026-02-21 21:33:37 UTC

- reply

@nprofile…nfq6 can you see your GrapeRank trust scores on Amethyst and/or Nostria? You have a little over 121k TAs that should be showing up.

2026-02-21 00:49:45 UTC

- reply

Ah I didn’t know it was merged. Looks good.
Will we be the first Service Provider to publish kind 0 events for our personalized SP npubs? Or will someone beat us to it?
We really do need different names for Service Provider (Brainstorm) versus service provider (Vitor’s Brainstormer).

2026-02-20 23:35:06 UTC

- reply

@npub1gcx…nj5z we were discussing this recently

2026-02-20 23:17:18 UTC

- reply

@npub137w…ked7

2026-02-20 22:08:10 UTC

- reply

It’s a good question.
Right now the answer is no. We have considered giving the pubkey a persona, like “Alice’s Brainstorm Personal Assistant,” which would include creation of a kind 0 event. Probably would follow the user (Alice) and maybe the “official” Brainstorm npub. We’d use this account to DM Alice updates, like:
- your Grapevine has been calculated
- subscription fees are due
- a new client just integrated TAs so you can see your scores there
- new services added by Brainstorm
- alert that the nsec of a highly trusted account just got NIP-56 reported as compromised (in which case we’d publish an urgent update to that account’s TA)
Potentially, we could establish the convention that we should put the service provider’s url in the kind 0 event, and that would serve as an advertisement of sorts. Then the URL could have more info, like service provider name, thumbnail, etc. Of course, this would be optional: the user might choose to hide this information. In fact, I’ve been wondering whether the tags of 10040 events should be encrypted in the content, and maybe that should even be default behaviour. But if you want to share this info, then do it.
What do you think of this solution?

2026-02-20 16:13:02 UTC

- reply

As of right now, the answer is no. Although maybe I’m not sure what you mean by map. The 10040 itself provides a map, does it not? Because inside Alice’s10040 you find pk_service_provider

2026-02-20 02:09:59 UTC

- reply

I got it this time! 📨💪🏻😂

2026-02-20 01:25:06 UTC

- reply

I don’t see it!
Confirmed, DMs broken. 😑

2026-02-20 00:34:38 UTC

If anyone has DMd me in the last few weeks and I’m not responding, i think there’s a good chance I’m not getting your DMs.

2026-02-19 15:19:00 UTC

Community call in less than an hour!
#nevent1q…xr6n

2026-02-18 14:02:30 UTC

#wotathon weekly call is tomorrow!!
#nevent1q…6fc8

2026-02-17 23:35:10 UTC

- reply

@nprofile…x6kf has made some great strides on the NIP-85 front recently

2026-02-15 04:19:35 UTC

- reply

The bots are going to discover they need their own WoT.

2026-02-13 03:51:34 UTC

- reply

I suppose we could assert that wherever a cycle is discovered, A subset of B subset of C subset of A, then all nodes in the cycle should be merged into a single node. Then we have a DAG. Any cycle that’s not so reduced would be breaking the rules, and the graph db would be not fully “normalized” (not sure if that’s the best word here).
The reason for deviating from that and allowing cycles is UX. Suppose I develop the concept for cities. I want to arrange them by continent, and also by their official languages. And I want to visualize the tree of Sets in the page. On the left I arrange them by continent — cities in N America, cities in Europe, etc — and on the right I arrange them by language: cities that list English as an official language, cities that list French as an official language, etc. Any given city will be an element of more than one Set node.
So at the top I’d have the Superset of all cities, then underneath two subsets: Cities that have an official language and Cities that are on a continent. Those two set nodes are each subsets of the other. (In theory they don’t have to be — there could be a city that’s not on any continent — but in reality, to the best of my knowledge, they are.) But we still keep them as separate nodes because it’s just better UX.

2026-02-13 03:26:48 UTC

- reply

My guess (or at least my hope) is that you and I are coming at the same problem from two different angles. Like a theory of physics, where one person (me) starts with simple observations / physical assumptions, the other person (you) builds some abstract mathematical tools; at first these seem like totally unrelated ideas, and only later do they discover that the mathematical tools are needed to use (or even to state) the theory to its fullest extent.
I still haven’t connected the dots. What do elliptic curves have to do with class threads? Should we expect a class thread to be points on an EC through embedding space? Or something like that?

2026-02-13 03:06:12 UTC

- reply

I just realized what I said is untrue: tapestry theory does not require the sets and subsets within a concept to be acyclic.
The class_thread_propagation relationship type (one of the 3 main relationship types that define a class thread — the other two types being class_thread_initiation and class_thread_termination) can be interpreted as the statement that node A is a superset of node B (or conversely, node B is a subset of node A). And there is no rule that A and B cannot be subsets of each other, in which case they are equivalent (any element of one is an element of the other).
I could say, for example, that the superset of all dogs (in my database) has subsets A and B, that A is a subset of B, and B is a subset of A.
So I shouldn’t refer to it as a DAG. It’s directed, and a graph, and is probably acyclic in many cases, but definitely not always.

2026-02-13 01:19:48 UTC

My 30-min talk at the first-ever @npub1the…4dha Nostr Nights on Feb 2 about @npub1hea…vp8p and our mission to bring decentralized reputation and web of trust to humanity is now available!
https://bitcoinpark.com/nostr-nights/david-strayhorn-weaving-the-fabric-of-nostr.html

2026-02-13 00:25:22 UTC

- reply

So “if this is true” — if time is quantized rather than continuous — then we don’t need to worry about our bitcoins being looted by the first person with a sufficiently powerful quantum computer … because why exactly?
I am singularly unconvinced by this line of reasoning. It actually makes me worry that the smartest of us might have shifted gear from unraveling the delusions of the fiat narrative in favor of building new delusions of our own.

2026-02-13 00:08:54 UTC

- reply

btw I messaged you in chat with some examples of basic cypher queries that we’ll want to use for the knowledge graph. That, and it would be useful for Orly to support bolt+s (I suppose I should add this as a feature request to the repo)

2026-02-13 00:04:06 UTC

- reply

Reading you document …
Part 4: Concept Lattice
Seems closely related to the notion of a Concept, as defined by class threads in tapestry theory: a concept is the set of all nodes and edges traversed by the set of all class threads that emanate from a single node. So if the node is Widget, then an example of a class thread emanating from that node would be:. A second class thread might exist that starts and ends at the same nodes but traverses different subsets, eg: widget—>widgets—>round widgets—>my widget
So class threads create a DAG that organizes things into sets and subsets. Which I think a lattice does too, right?
The most important aspect of class threads is that the class header node (widget) and the superset node (widgets) must be distinct. The temptation is to merge them together into one node for the sake of simplicity, but that would be a mistake; you have to keep them separate if you want to integrate concepts the way that tapestry theory allows you to do. Seems like a mundane detail, but it’s important.

2026-02-12 16:01:43 UTC

We’re live!
https://zoom.us/j/91801428092?pwd=k5Gl2pYKX1X7GXPpiEdsEDaGIYsdvK.1
#nevent1q…ln29

2026-02-12 15:59:09 UTC

- reply

I don’t know the difference between a type and a category, or type theory and category theory, but I want to learn.
Here’s a concept for you. Right now, in Orly, any single neo4j node probably has lots of information stored in multiple properties. Basically an entire nostr event, for the Event nodes. I have an idea that any given node can be “exploded”, or unpacked, or whatever we call it, which means we take the information in the node and put it into the graph. Which means it doesn’t need to be in the node anymore. Example: instead of each Event node having a tags property, we have Tag nodes (this is all stuff you’ve already built, of course). But here’s the question: how far can we take this process? Can we end up with a graph that is nodes and edges with *no data* inside any given node or edge? Other than — perhaps — a uuid? Or maybe instead of a uuid, a locally unique id?
Suppose we call this state a “fully normalized” graph db. Will there be heuristics for when we desire full normalization and when we’ll want to break normalization, from a practical perspective? Like it’s more performant to retrieve tags when they’re packed as a property into an Event node than if we have to do a lot of cypher path queries?

2026-02-12 15:01:10 UTC

#wotathon weekly community call starts in one hour!
#nevent1q…vged

2026-02-12 14:59:51 UTC

- reply

The prototype brainstorm (which is klunky but we’re rebuilding it) calculates number of verified followers, muters, and reporters, where “verified” means rank score is above some cutoff. Right now nostria is showing just the rank score, but probably in the future nostria will ingest verified reporter count, which we could easily break down into report type, and show that number, and/or a red flag if it’s above some cutoff.
And these are all personalized scores, not global.

2026-02-11 21:49:54 UTC

- reply

In the tapestry model, new knowledge usually takes the form of new nodes and/or edges, so I think we’re all 3 aligned on that. Sometimes there can be deletions, updates and/or reorganizations, but additions would be the main method.
The core insight of tapestry model is that the vast majority of new knowledge comes from my trusted peers. The need to arrive at consensus / develop conventions with my peers over seemingly mundane but nevertheless vital details of semantics and ontology — like what’s the word we use to describe this writing instrument in my hand, or do we use created_at or createdAt for the timestamp in nostr events — is what dictates the architecture of the tapestry model.
The rules that govern the knowledge graph (class thread principle and maybe one or two other constraints) need to be as few and as frugal as possible. Isaac Newton didn’t come up with 10 thousand or 100 thousand or a million laws of motion, he came up with three. (Hamilton later replaced these 3 laws with a single unifying principle — an improvement, getting from 3 all the way down to 1.)
The tapestry method of arriving at social consensus only works if all peers adopt the same initial set of rules as a starting point, and that doesn’t happen if the set of rules are more complicated (or greater in number) than they have to be.
The class thread principle is a simple starting point. Only a handful of “canonical” node types (~ five) and “canonical” relationship types (also ~ five) are needed to get if off the ground. Once off the ground, an unlimited number of node types and relationship types can be added — usually learned from trusted peers. And the class thread principle allows concepts to be integrated vertically and horizontally. (So it’s not like you end up with a huge number of mostly disconnected sql tables. Class threads support multiple meaningful ways to weave disparate concepts together, even without adding any new node types or relationship types.)

2026-02-11 20:27:56 UTC

- reply

We’re gonna have to build the solution sooner rather than later.
And there’s still so much basic theory to work out. @npub1fjq…leku

2026-02-11 18:10:26 UTC

Don’t forget: next #wotathon community call is tomorrow!
#nevent1q…g3tu

2026-02-11 18:07:49 UTC

- reply

@nprofile…2dhd — I think your AI agent just volunteered you to lead a very interesting discussion topic on tomorrow’s #WoTathon call! 👍🏻

2026-02-11 17:55:06 UTC

interesting
#nevent1q…6qcc

2026-02-11 17:13:30 UTC

- reply

I’m thinking the class thread rule for LLMs could be something like the Einstein equation for 4-manifolds. In both cases, we have a continuous space, the shape of which is constrained by an equation.
An untrained LLM doesn’t follow the class thread rule, but a well trained one does.
So tapestry theory would have several uses:
1. How to read the graphical (compressed) representation of the embedding space directly
2. How to detect when training is incomplete: measure the extent to which the class thread rule is or is not fulfilled.
3. Corollary to 2: how to guide LLM training to those areas where the class thread rule is being poorly followed.

2026-02-11 16:59:27 UTC

- reply

I’ll be there — Czech lager it is 🍻
In the meantime I’ll work on answering your questions about the query algos we’ll be using so you can build them into Orly.
For starters, here’s the cypher query to return all class threads in the database:
MATCH classPath = (listHeader)-[:IS_THE_CONCEPT_FOR]->(superset:Superset)-[:IS_A_SUPERSET_OF *0..5]->(subset:Set)-[:HAS_ELEMENT]->(element)
RETURN classPath
If we want the list of all widgets, we’d add:
WHERE listHeader.uuid = (the list header for the list of widgets, which is an event id or a-tag) and use RETURN element
If we want all widgets made by Acme Corporation, we’d need to have previously organized the list of widgets into sets and subsets, one of which would have to be “the set of widgets made by acme corporation” and we’d add something like WHERE subset.uuid = (the node that represents the set of widgets made by acme corporation)
btw an alternative naming system for the three relationship types you see above are:
CLASS_THREAD_INITIATION =
IS_THE_CONCEPT_FOR
CLASS_THREAD_CONTINUATION =
IS_A_SUPERSET_OF
CLASS_THREAD_TERMINATION =
HAS_ELEMENT
Every class thread has one initiation and one termination edge, and any number (0 or above) continuation edges.
The class thread rule specifies, in effect, that you can’t make a Widget an item on the list of, say, Cats. And you can write algos to detect such category errors by looking at data formats. Widgets have a property called “brand”, Cats don’t. So if I have an item in my database on the list of Cats and a record it was made by Acme Corporation, the code throws a flag that the class thread rule has been broken and the error should be addressed.

2026-02-11 16:26:09 UTC

- reply

One way functions.
Always makes me think of diodes. One way valves for electric charges, made possible by the magic of quantum tunneling.
I don’t know where exactly to go next with this line of thinking, but this is how we connect cryptography to quantum mechanics. There’s no universe where this doesn’t work, one way or another.

2026-02-11 16:11:40 UTC

- reply

I’ve known about category theory and elliptic curves but I’ve never studied either one in depth. Or known they were connected. And I’ve never heard of the discrete log problem. I will need to do a deep dive.
But check this out:
https://prettygoodproject.substack.com/p/for-web-of-trust-to-succeed-it-must
For web of trust to succeed, it must first be paired with the universal language of mathematics

2026-02-11 16:03:23 UTC

- reply

I have long observed that many people in society are either unable or maybe simply unwilling to indulge in concepts that do not have names. If it doesn’t have a name — one that has been approved by the tribe, of course — then it’s not a “real” thing. Or so the thinking goes.

2026-02-11 15:54:19 UTC

- reply

We’re gonna have a lot to talk about!!

2026-02-11 15:47:59 UTC

- reply

Hypothesis: class threads (or something like them) are an emergent property of well-trained LLMs.
I have no idea what (if anything) this hypothesis has to do with elliptic curves. But maybe … an embedding space that is fully infused and governed by class threads will have some sort of global topological features that we can derive and detect empirically?

2026-02-11 15:39:21 UTC

- reply

I think I’m the same way as you Mleku - often I’ll have concepts that are well defined in my head but I don’t have the nomenclature for them. This can facilitate creative thinking but can inhibit communication with other people, for obvious reasons.
A relevant example: only recently did the nomenclature crystallize in my mind the distinction between connectionist AI (LLMs, neural networks, machine learning, etc) versus symbolic AI, aka GOFAI (good old fashioned AI). These two distinct concepts formed in my head as an undergrad in electrical engineering forever ago but didn’t have a name or resurface in my mind until mid or late 2025 when a friend asked me if I had ever heard of “symbolic AI.”
I don’t understand the math of connectionist AI, or the math of what you’re doing, well enough to connect what you and asyncmind are talking about to what I’m doing with the knowledge graph. But some of what y’all are discussing definitely catches my attention. I’m wondering whether continuous versus discrete (quantized) is the organizing principle. Connectionist AI deals with continuous spaces where we use tools like gradient descent to arrive at local minima. GOFAI, symbolic AI, and graphs are discrete. Could it be that the basic rules and features of the knowledge graph (most notably: class threads) are an emergent property of well-trained LLMs? I conjecture the answer is yes, as evidenced by things like the proximity of hypernyms (animal type) and hyponyms (dog, cat) in embedding space.
Suppose we want to take an embedding and compress it into a smaller file size. Could it be that a graph is the ideal way to represent the compressed file? If so, can we read straight from the graph without the need to decompress the graph and rebuild the embedding space? If so, then we have to know how to query the graph, which means we have to know the rules that organize and give structure to the graph, and the class threads rule seems like a great contender for the first (maybe the only) such rule.

2026-02-11 14:27:41 UTC

- reply

This is a very interesting discussion that I’m going to have to digest. I’m going to ponder whether what you’re doing with ECAI and what I’m doing with tapestry theory are complementary. Our Brainstorm product has two parts: the Grapevine, the prototype of which is what we’re currently building with you Mleku, and the knowledge graph (or sometimes I call it concept graph). The knowledge graph encodes not just mundane content, facts and information, but also ontologies, semantics, etc that we can in principle build individually but prefer to build collaboratively — which is what the grapevine is for. The knowledge graph supports queries like: give me the list of all widgets in my local graph database with characteristics A, B, and C. These are well defined, generalizable and extensible path queries that can be written using cypher. It requires the uuid for the node that represents the concept of widgets; if we don’t have that uuid, we can find it using the same cypher query.
Mleku are those the kinds of queries you’re talking about when you say you can optimize for them?

2026-02-10 22:14:28 UTC

- reply

It used to be that you had to go to settings and select one of 3 or 4 hardcoded relays but looks now like that’s been updated … ?

2026-02-09 21:46:20 UTC

- reply

Reasoning from first principles is always a good thing. But we have to recognize our cognitive limitations as individuals. I’d love to review every line of open source code that I’ll ever use but that’s never going to happen. The vast vast majority of what I know about the world is stuff that I’ve learned from other people. Topics like bitcoin I reason from first principles to the best of my ability, but I’ll never stop relying on trusted people to guide my thinking.
I don’t have a strong opinion on core vs knots. And LukeJr can be kinda crazy. But he’ll always have my respect based on a few of the positions he’s taken in the past. Among them: the March 2013 unintentional chain split. Read the bitcoin-dev channel, he knew pretty much instantly what happened and what needed to be done, faster than most (including Gavin).

2026-02-09 19:45:59 UTC

My clawdbot just named itself Arc. ⚡️

2026-02-08 01:40:09 UTC

- reply

Would it be feasible for a single instance to support multiple sources? So one tech-savvy dev could calculate WoT scores and publish TAs for family, friends, etc?

2026-02-06 23:14:44 UTC

- reply

Ok - I just created a new user called straycat’s shadow, followed myself (straycat), logged in at relatr.xyz/ta. I see Gigi listed at the top as the “source” which I guess means perspective, meaning that if Gigi follows Alice, then Alice’s score gets a big boost bc she’s only one hop away from
Gigi. But I want to make myself (the new account I just created) the source. Can I do that? Or do I need to run my own relatr server?

2026-02-06 21:07:13 UTC

- reply

I probably don’t understand relatr or CVM well enough so maybe I’ll start by asking some ELI5 questions. For starters: suppose I’m an average nostr user, not a dev, and I want relatr to calculate and publish TAs for me. My goal is to see the 0-100 trust scores (rank) show up when I log into Amethyst. How do I do that?

2026-02-06 19:08:37 UTC

- reply

Relatr has started publishing Trusted Assertions using NIP-85. I’m not sure how deep into it they’ve gotten or what pain points they’ve encountered ( @npub1gzu…a5ds ?). So far there are a handful of clients (Amethyst, NymRank.dev, NostrHub.io, Nostria, Sandbox, some others) with varying degrees of NIP-85 integration. So we’re still in the early stages of testing it all out.
Here’s the litmus test in my mind as of today (subject to change as I have discussions like this): suppose law enforcement goes to zapstore or some other client and demands that you integrate with Brainstorm and support GrapeRank but blacklist PageRank by Vertex. Or vice versa. Oh and you can’t tell your users that we ever spoke to you. I’d like you to tell them that it’s impossible to comply with their request because that’s just not how it works — the choice of algo and WoT service provider is completely out of your hands. Does ContextVM work that way? Or can it? I genuinely don’t know the answer to that.

2026-02-06 18:27:17 UTC

- reply

I’m not saying that isn’t your goal. I am saying, however, that when the client selects one or more WoT service providers for its users, that introduces a fatal flaw that we need to fix before it actually kills us, and NIP-85 fixes it.
When I first learned about bitcoin, I thought there’s no way we can fit every transaction for billions of people onto a single ledger. Turns out you can. It’s a challenge and we have to figure out how to deal with it, but it’s not a fatal flaw.
If I don’t have room on my personal NIP-85 relay for 8 billion kind 30382 events, then I won’t store that many. Only the people I want to keep track of. I can keep track of reputation in my head but my head is limited by Dunbar’s number, about 150 people. If I’m willing to fill my personal NIP-85 relay with a million Trusted Assertions, maybe 1 GB of data, then I’ve boosted my personal Dunbar’s number from 150 to a million. I’d call that a good start.

2026-02-06 17:22:04 UTC

- reply

Hard for a user to get a feel for what a PageRank score means, especially when there are many different ways to run it (different parameters, etc). If I as a user could be guaranteed to see the same PR score on zapstore as I see on amethyst and damus and primal and everywhere else, maybe over time I could get a feel for what score is a good score. And that can happen if we give users full control over how and where their scores are calculated. Direct client-to-vendor integration is contrary to that goal. But NIP-85 is designed with that goal in mind: giving users, not clients, control over which WoT service provider to use.

2026-02-06 17:14:22 UTC

- reply

I’ll let you and others know when the next iteration is ready to try out. It’s already better than my prototype but needs polishing.
GrapeRank has a very clear roadmap for contextual scores. At least two major categories of strategies. It’s only a matter of time.
Does PageRank have a path forward for contextual scores? If you’re pursuing one, I’d like to hear it. I’ll actually suggest a method: one of the categories of strategies for GR contextual scores is decentralized lists, which can be curated by any list of pubkeys:
- everyone (including bad actors, not ideal)
- your follows list (easy to code, not a bad choice)
- your follows plus their follows (wider reach but more chance for bad actors to sneak in)
- anyone with personalized GR score above a cutoff
- anyone with personalized PageRank score above a cutoff
@npub19ma…axkl

2026-02-06 17:03:20 UTC

- reply

If I search for John on NymRank I see 10 results.
Do you mean the 2nd result on npub.world for Odell? NymRank serves a different purpose from your typical keyword search. In addition to GrapeRank, it calculates an affinity score that dings the second result on your page, mostly bc the name isn’t an exact match.
I know this isn’t an exact side by side comparison. That will be coming soon.

2026-02-06 16:14:32 UTC

- reply

GrapeRank may be newer and more experimental than PageRank but in our hands, GR empirically does a better job than PR at its core function.
Search for Odell at npub.world (which uses PageRank, as of course you well know) to NymRank.dev, which uses GrapeRank. You’ll see a floodgate of impersonators that fool PageRank but are filtered out by GrapeRank using standard (default) GR parameters.
https://i.nostr.build/uPB15hsBNQBbyEiw.png
https://i.nostr.build/VTjVWLpD4Pg70DOm.png

2026-02-06 15:37:18 UTC

- reply

Once it’s available and the marketplace is developed, your users will probably already be signed up to Brainstorm or a competitor service that supports NIP-85 so they can use them on clients like Amethyst. Vertex could easily be one of those providers. It’s not like you have to sign up every time you want to use the scores.
Yes, PR is widely deployed, with well known and easily exploitable weaknesses like link farm attacks, the solutions to which are band-aids. This attack, btw, is extraordinarily cheaper and easier to execute in the context of nostr and follows than in the context of internet keyword search and hyperlinks.

2026-02-06 15:07:31 UTC

- reply

I’m not suggesting you send your users to my prototype! The prototype is only that, a prototype, not a product. But the product is coming. (Soon™️!) So if I were you or any dev who hasn’t been following this closely and doesn’t want to experiment with alpha (or pre-alpha) stuff, I’d wait until there’s at least one stable, user-facing WoT service provider who supports NIP-85, and when it’s ready (easy for users to sign up, good UX, all open sourced, etc), I’d be ready to integrate scores using this method.
Yes, I created GrapeRank. I have articles explaining what it is and how it works. My prototype is open source and the rebuild will also be open source, which means you can be your own WoT service provider. And our #wotathon is very likely to produce another team who will have a competing implementation of GrapeRank. The more the better.

2026-02-06 14:56:43 UTC

- reply

Is this a non-issue?
https://i.nostr.build/OF9C45jfOhTUQ8oO.png

2026-02-06 14:55:47 UTC

- reply

So the real result is the first result, and that’s the solution.
That solution works for Odell because he has the second (or maybe third) highest PR score in nostr.
What if you’re not an influencer? What if your job is to screen code for malware, I want to trust you and not an impersonator to do it, but you haven’t shotposted your way to 1000 followers. Your system as is will not protect them from impersonation.

2026-02-06 14:02:40 UTC

- reply

GrapeRank is the best tool to flag the moustaches and trucker hats!

2026-02-06 13:59:07 UTC

- reply

So step 1 is I gather all my files in a folder, then I point this tool at it? Is there a tool to help me gather them together? (Some are on nostr, some on github, some on substack, some on my local machine, etc)

2026-02-06 13:38:38 UTC

- reply

So how do you explain this many impersonators fooling your system?
https://i.nostr.build/kZMa332xMlgEhJc6.jpg

2026-02-06 13:20:46 UTC

- reply

That’s my prototype which I built on my own and is super slow. As of a few months ago I am with a team and some devs who are much more talented than I am!
Yes, more than one click when you include sign on, plus one to sign the 10040 event which needs to be done or else amethyst or zapstore or damus or wherever won’t know how to access your scores.
For the prototype, I need to trigger score calculation on the back end which I just did for you. In an hour or so your scores will be calculated, stored locally and available at https://straycat.brainstorm.social/profiles.html. (Our rebuilt version does it in ~ 5 minutes.) if you go back and sign the 10040 event, then I’ll be able to publish scores as Trusted Assertions (which for the prototype takes hours, for the rebuild takes another 5 min) and you’ll see them on amethyst. 😃
By relying on PageRank, your zapstore users are going to get fooled by the first bad actor who decides to employ a link farm attack. If that hasn’t already happened, it’s only bc it’s not yet a big enough honeypot.

2026-02-06 03:14:37 UTC

- reply

Mostly text - NIPs, protocols written and rewritten, articles, blog posts. Some implementations in code of varying completeness. A few video talks. A few keynote presentations.

2026-02-06 02:16:07 UTC

- reply

I’m in the process of digging up everything I’ve ever written about Brainstorm (even before I called it Brainstorm) so I can train an LLM on it, then vibe code the Knowledge Graph half of it, the counterpart to Grapevine which is currently already being built into a product. I’ve never done this before though. Is it worth getting a Mac mini? Splurge for a studio?

2026-02-06 01:52:03 UTC

- reply

If anyone ever starts to use zaps as a real and meaningful signal, like to select content for trending feeds, global zaps will be a piece of cake to game. Imagine an army of bots controlled by a bad actor. They zap a few sats back and forth in a continuous closed loop swarm of activity. It will cost basically nothing for them to top the zap leaderboards on both sending and receiving.

2026-02-06 01:46:21 UTC

- reply

That’s kind of how NIP-85 already works. To change providers, I just replace my 10040 event with a new one.

2026-02-06 01:45:18 UTC

- reply

One-click sign up for the service once and forget about it (unless you want to personalize the parameters, which some users really seem to want to do). From then on, your personalized scores are available throughout nostr.

2026-02-06 01:43:20 UTC

- reply

Remains to be seen how often we’ll update. My prototype takes forever. The rebuild is an order of magnitude or two faster. Most likely, we’ll have a free tier that updates automatically but very infrequently plus on-demand for a limited number of times per day or week or something. Then at some point we will have paid tiers to update more frequently, then probably add-on services — for example, support for red flag events like if a highly trusted pubkey suddenly gets multiple trusted NIP-56 reports of key compromise, that pubkey will get updated immediately. (Note that if this were to happen, for example if Dorsey’s key were to get compromised, any PageRank-based system would be toast, bc his pubkey’s PageRank score will for all practical purposes NEVER drop below yours or mine.)
When nostr starts to get millions of users, there are going to be all sorts of strategies for scaling. None of them perfect, but many that are more than good enough. GrapeRank has a natural scaling solution which is that if you know a few trusted pubkeys in some distant community (like China or Japan, for instance) and they run Brainstorm over their communities, then if they allow you to import their Trusted Assertions, you don’t really need to recalculate over the pubkeys in their community.