Is there an implementation of dependent type theory using Javascript? I've been thinking about building a web-based proof-assistant but there's a lot of learning material to go through. Plus most implementations seem to be in Haskell/Ocaml/C++.

I spoke at Postgres London this morning about using PostgREST to simplify the tech stack for some kind of open-source webapps intended to be self-hosted - both bringing convenience to users and easing the burden of developers.

Slides and code example are here:

Fixed enough issues in my visual graph dataset editor that I can now call this alpha-quality software:

Give this a try (preferably in chromium-based browser):

Identity MUST NOT be confused with location/address. This architecture inherently introduces intermediaries and raises issues of control and ownership (in addition to security and privacy).

Show thread

OTOH, I find Twitter's BlueSky project's approach far more promising. Self-authenticating identities and data will be the way to go:

This is one thing ActivityPub / Mastodon devs should have gotten right but did not. I don't know if a reinvention is possible now.

Show thread

Came across - yet another identity solution - this time from Tim Berners Lee's SOLID project:

At 1:28 in the video, Henry adds a person as his contact because it's using his friend's picture. This is a surprisingly weak trust model in 2022.

Uninformed opinion but Solid seems destined for failure. Overly complex where it doesn't need to be (uses RDF) and too simple where it shouldn't be (cryptography).

I will be speaking at Postgres London on 24th June about a new architecture pattern that will drastically simplify the development and maintenance burden for open-source webapps.

I have been maintaining a list of my software ideas here:

How many of these you think are terrible? There's a reason why I haven't moved this list to GitHub. 😂

I made a twitter thread trying to explain Geometric Algebra:

This is all revolutionary stuff for physics education. I would be VERY disappointed if in 10 years, physics textbooks haven't switched to this. We've been living with a mistake which made things far more complicated than they really are.

I once wrote a 2D physics simulation engine. Took me a month and ~5K lines of code. With this algebra in place, it takes less than 60 lines of code. And works in 2D,3D,4D etc.

I have been looking for a visual graph dataset editor that allows me to add arbitrary structured data fields on nodes and edges as well as manually layout the graph for presentation.

A flash of inspiration and 12 hours of furious coding later, I have this working!


Actually, coding this was quite pleasant. This uses SVG and Svelte so I get most of interactivity, styling, reactivity for free.

Updated the visual style and liking it much better than the previous one:

Can you suggest an awesome domain name (that's available) for this?

Was hoping that DoltHub would be a good solution for collaborating on datasets. But the interface is a bit clumsy and seems to assume that users are technically proficient. doesn't have any UI for modifying individual records.

My search for a "GitHub for datasets" continues.

Now showing convenient links for books and research papers thanks to projects like LibGen, SciHub, OpenLibrary etc.

Quite liking this interaction mode (zoomable treemap), but as always, I'm quite awful at choosing colors. 🙂

What should a definitive taxonomy for all human knowledge look like?

Is there a standard vocabulary I can use or must I reinvent it?

I've looked at Library classification systems like Dewey Decimal / Colon, and university department hierarchies. But still haven't found a satisfactory answer.

Some pointers here:

Show older

Fosstodon is an English speaking Mastodon instance that is open to anyone who is interested in technology; particularly free & open source software.