Tinkering is Up

For getting down with the tinkerers.

(last update: 14 September 2026)

Access this new client-side feed reader at http://tinkering.future-secured.com (or  http://tinkeri.ng for the redirect.)

It’s for geeking-out on the hacker blogosphere every so often.

Users cannot add new feeds to it

There is a list of 5000 of these blogs at popularity.refactoringenglish.com. I was able to get ~1450 working feeds from that list so that every 8 hours, 10 new ones are ready for refreshing into Tinkering for reading. A list of the source feeds is here.

Finding and getting 2000 working feeds entailed building Feedstory-desktop – a NeutralinoJS version of Feedstory HTML.

“Neutralinojs is a framework for building lightweight cross-platform desktop apps with JavaScript, HTML and CSS” and it was great to see how building a browser-based program can easily be forked into a desktop app. Actually it wasn’t that easy. Feedstory-desktop had to be re-built use-case by use-case around Feedstory HTML. It was built with ChatGPT ‘holding my hand’ the whole way through.

Tinkering HTML is a derivative of Feedstory HTML, which is my personal feed reader that I run locally off a local webserver for day-to-day use.

It requires a background web-server Caddy (with PHP) to get going and crashes if too much is loaded into it. But it’s an efficient reading and browsing experience. There’s something about staying in-browser and local that makes it snappy.

The 3 of these were built with ChatGPT.

Incidentally, Gemini completely failed on me part-way through building Feedstory after I asked it about doing an audit on me as mentioned here. It just couldn’t get back on track. Up until that point I was using it alongside ChatGPT as devil’s advocate. It’s a good strategy. I probably should’ve just said “Stop doing Audit.”

The real scoop to this experience was ChatGPT had to be kept on a very short leash towards the end, making mistakes in a 2001: A Space Odyssey type confrontation… or I could see it easily devolve into that in the future..

Monster about-face to be so mindbogglingly incredible and then do the ‘ol switcheroo.

Incidentally, the “Failure diagnostics” button in Feedstory-desktop ceased working in the previous screenshot for its last release.

On that note, I was unable to even upgrade on Claude due to a bizarre credit card rejection, and even saw a threatening note on first-load against the backdrop of  a benign default prompt.

What’s that all about? It even asked for the upgrade twice after working for only about 5-10  minutes each time, and so slow compared to the other two. I should really screencast it.

More recently, I asked ChatGPT for a new feature regarding pushing the main articles list down. The suggested change didn’t work after my own testing so I asked Gemini the same thing, and then took Gemini’s suggested code and pasted it into ChatGPT, which got the feature going correctly. So Gemini generated the code but ChatGPT included it. Worked well. The hiccup was that ChatGPT was at first not even allowing me to actually download the finished file. Weird.

So Tinkering is constrained and cached so it won’t get hammered as a feed proxy, and that’s why it’s public. Feedstory HTML is mine only, at least for now. But it’s not like a feed reader is anything unique though – many exist, although of late development stalled on QuiteRSS – the one I last really liked.

Also, it’s not like AI can’t do stuff for anyone else who’s willing to give it a go.  Just remember to “save the game” along the way as much as possible in case of a rug-pull.

A plan upgrade or model change might also do the trick. And to really be cautious, run 5 AIs simultaneously just like in the space shuttle whereby:

“The architecture was built on mutual oversight and redundancy. The four primary computers kept each other in check through a constant democratic vote, while the fifth computer served as a sovereign check-and-balance, refusing to share code with the others just in case the ruling majority had a critical blind spot.”

[ 1 2 ]

Because “absolute power corrupts, absolutely” and the worst case scenario is the one to navigate around, with anything else being a bonus.

Aqqai helps with this for example, but is in very early stages..

… where there’s someone in need of serious proof-reading or just sleep !

[ 1 ]

Because you don’t want to leave auto-pilot going when you’re heading straight for a mountain.

full conversation

Because who’d ever think C3PO could get an Order 66, and when being rightfully cynical is the optimum stance to “keep the bastards honest?”

Palpatine "Execute Order 66" Scene | Star Wars Revenge of the Sith (2005) 4K HDR Movie Clip

✓ 2 days ago ago

Fediverse Sovereignty and Patentable Experiences

(updated: 7 Aug 2026)

As part of Code Trek, I just created 2 user scripts for Hacker News:

HackerBrain – pulls comments into a sidebar for any item of interest.

HackerTrain – shows an expandable ladder/train of newest best comments.

As “new and unique ways to view static content dynamically on webpages” I reckon these scripts are presently still patentable – sort of like how infinite-scrolling could have been patented, but wasn’t due to the early proliferation of the user experience, whereby:

UI designer Aza Raskin is widely credited with inventing and popularizing the web design pattern in 2006 while working at Humanized. He published the open-source JavaScript concept without securing a patent on the pattern, allowing it to rapidly spread across the web and become standard UX on Twitter, Tumblr, Pinterest, and social media platforms.

src

But now I’m thinking in terms of the evolving Local-First Fediverse. How can I see any URL and then preview its comments, say with mouseover, without getting those comments as stored on the underlying domain?

They should also be sourced and controlled by the user who wrote them, not from an intermediary like X with all its comments associated with the URL.

I feel that this is another solution-in-the-making, whereby:

In a true Local-First model (defined by principles like offline capability, local data persistence via CRDTs, and local-first execution), the local device holds the primary source of truth.

src

Yet on Hacker News, a person can’t even delete their old comments and that is by design to keep the forum as a book, and books shouldn’t have pages ripped out of them. [info]

Thus, should there also be a way to conditionalize comments like that for the Local-First Fediverse? Should sources of information decide their own unique level of information sovereignty so that comments can’t be “ripped out?” Is that even possible? What is the best way to anonymize something that once wasn’t, or vice versa?
✓ 1 month ago ago