UnScroll

Paul Ford timeline project

Unscroll Intro, by Paul Ford

About fifteen years ago I had this idea: Timelines on the World Wide Web! Hardly an original idea. But I got super into it. I thought I could somehow fix the world a little by making a great website that organized things chronologically. I had all kinds of pipe dreams along these lines. Laugh away at me, I deserve it.

As happens with my big ideas I kept aiming too high and never getting anything done. I kept approaching the problem from different angles, rather than taking small steps. For example, I decided that the only way I would finish my book about the web was to write a content management system that let you organize everything...in timelines. Now I had three problems!

I did a ton of thinking about time. I learned a lot about how, for example, the Postgres database handles dates. I learned about different dating systems and calendars, and how various disciplines date things back to the beginning of the universe, and how the Library of Congress dates things. Chronology is hard. Time doesn’t lend itself to becoming data, no matter what the stock market tells you

A few months ago I grabbed all the old code and threw it into Claude. Goodness did it have some things to say about my programming

Work is very busy and I’m suspicious of this project’s ability to ruin my life so I try to have lots of little tasks for the bots over the weekend. Sometimes I work on it while on bike rides. Type, bike, type. It’s in a very, very rough state. It’s sub-alpha. But I thought it would be useful to resurrect it and talk about how things are changing

it was also instrumental in helping me download Kiwix, which gives me an offline Wikipedia, with images, so that I can “read” thousands of Wikipedia pages to look for relevant chronological content, then assemble that, then feed that into further queries for Archive.org and other platforms. This way I don’t hammer Wikipedia. It also helped me convert the entire Wikiquote database, and found 90,000 quotes with dates.

What does it actually do? Well in the background it makes events, basically—little data blobs. Then it assembles them

It’s not a good product yet and may never be. But it sure does keep me off social media. And it’s something to talk about. I’d rather talk about work in particular than vibecoding in general.

I think some things to discuss are:

How hard it is to get a system to be coherent when it has different ways of looking at data. LLMs excel at “view of the database” programming but when you go too far astray you need to think hard.

Refining and targeting a data model, and what that means for skills.

Filtering for accuracy.

What “product” means in this context.

What it means when platforms can be little hobbies.

I’m noting this because I feel obligated to narrate the moment, especially since Simon Willison is putting in so much work, I should talk about this stuff too if it’s helpful