Eighteen days,
three versions,
one site that kept
rewriting itself

The empty folder

It started at 7:42 on a Sunday evening with nothing but a folder name and a sentence: I want a place where people can read what I write, see what I photograph, and talk back to me — and I want to own it.

By 11:16 that night there was a working site on my laptop. Sign in with your email, get a six-digit code, post something, like it, comment on it. Not a mockup. The real thing, running.

That first night set the constraint everything since has been built inside: no frameworks, no build step, and almost no outside dependencies. Three, in total. It sounded like a limitation. It turned out to be the reason the whole project stayed understandable to someone who cannot read the code.

Somewhere to put things

Five days of evenings. Photographs and videos that keep their own shape instead of being cropped into squares. A full-screen viewer with arrow keys. A gallery that merges every photo on the site into one grid without ever duplicating a file. Folders, because a few hundred photos with no albums is a shoebox. Bulk selection, so filing a month of pictures is one action rather than forty.

Then something small that turned out to matter more than any of it: a version number under the logo.

The day I stopped building a feed

One long day, four versions. The home page stopped being a single scroll of everything in order and became topic columns — I write into as many topics as a post belongs to, and the reader picks up to five of them and gets a column each, side by side.

The idea underneath it is the one I am still most pleased with. A photograph of Chongqing at night is Photography and China. Forcing me to file it under one would hide my best work from half the people who came for it. So topics are a library I write into, and columns are one reader's arrangement of that library. Same posts, everybody's own front page.

A first-time visitor never sees a chooser or an empty screen. They get my default three. Customising is something you discover, not a form you fill in before you are allowed to read anything.

That same day I also added something with no visible benefit at all: 25 automated checks that prove I can still undo everything. Not that the undo script exists — that it genuinely works, that it restores a broken project, that it does not delete itself in the process, and that it never touches my posts, my photographs or my members. A way back that has not been tested is a story you tell yourself, not a safety net.

Launch day

888story.com went live. Nine days from an empty folder to a real site on a real domain, with real photographs on it, that anyone could read without signing up for anything.

There is a version of this story where launch is the ending. It was not even halfway.

The Atlas

Columns are a good answer to "what have you written". They are a terrible answer to "how does any of this connect" — and connection is the whole point of what I am doing. China and photography and the businesses and the reason all three are the same project.

So I built a second layout. The Atlas draws the site as a map: my photograph in the middle, every topic a circle orbiting it, and lines between the circles. The lines are the good part. On any post I can say this connects to that topic, and here is why — and the note matters more than the link. "The shop is most of that 41%" tells you something a bare arrow never could.

It answers a question no version of the site could answer before: not just what this topic contains, but who points at it.

Two more things about that day I would defend to anyone. First, phones get a different view rather than the map made small — a constellation shrunk to 375 pixels is a smudge. Second, and this is the one I am proudest of: I gave the Atlas thirty days to prove itself.

Nobody asked me to do that. The site now counts the thirty days itself and will ask me, and only me, when the time is up.

What a real site found in an hour

The Atlas went live. Within the hour I had found two faults, and both were the same kind of fault — things that worked perfectly in a demo full of sample content and misbehaved on a real site that is still filling up.

A photograph's description was being cut off at 90 characters for a tidy one-line bar. Every caption in the demo was short, so nobody ever saw it. In the one place a reader has stopped and is properly looking, my own writing ended mid-sentence.

And a button that hides and shows the lines between topics appeared completely broken — because the live site had no lines yet. Pressing it changed nothing. It now carries the count, dims when there is nothing to toggle, and says so when you press it.

The lesson went straight into the project's own notes, where the next session will read it: a demo is a best case, and best cases hide the states a real site spends its first weeks in.

The deadline I set for myself

Right now 888story.com runs two layouts at once. Classic columns are the default; the Atlas is one click away for anyone curious. Thirty days from the night the Atlas first ran, one of them gets deleted.

I do not know yet which. That is rather the point — I would rather find out by living in both than by guessing in advance, and I would rather have a date than an opinion.

In the meantime there are three separate ways to undo everything, and they are not interchangeable:

A layout switch is not a rollback. If the server is wrong, every reader is still getting it wrong no matter which button they pressed. Knowing which of your safety nets actually catches which fall is worth more than having a lot of them.