The Greg Wide Web
The site, drawn as a web
This is the atlas mapping itself: every page a node, every internal link
a thread, and the centre wherever you put it.
How to walk it
Every page is a node, every concept a page declares is a node, and every
internal link is a thread; the web re-spins around whichever node sits
at its centre, with each ring one link further out.
Rest on a node and it names itself and lists the
concepts it declares. Click a node to light up its
neighbours, click it again to re-centre the web there,
and double click (or click the callout) to go to the
page itself. The concept tags in that callout are buttons: press one and
the web re-spins around the idea rather than around the page.
Three controls sit under the figure. Centre on takes the
name of any page or concept and jumps there, which is faster than walking
a graph of this size by hand. Rings is how far out the
web reaches, one to three links. Show is both the legend
and the filter: each row is a section of the site in the hue the figure
draws it, pressing one draws that section alone, and
Concepts draws the ideas by themselves, joined wherever
a page declares two of them together or a hand-written relation says they
belong together. A filter re-spins the web rather than dimming it, so the
rings keep meaning link distance inside whatever is left.
None of it needs a mouse. Tab to the figure and the arrow keys walk the
threads: left and right along the spiral you are on, up the radial toward
the middle, down the radial away from it. Enter selects, Enter again
re-centres, and Escape lets go.
What the web is made of
The page nodes and the link threads are extracted: a
builder walks the built HTML in site/dist and reads the
anchors it finds there, so a new page joins the web because it exists,
not because anyone remembered to add it. Asking the built artifact
rather than the source is deliberate, and it is a lesson this repo
learned the expensive way on a fragment check that read page source and
declared every deep link broken while the built HTML carried all of
them.
The concept nodes and the relations between them are
curated. No concept is invented by scanning prose for
interesting words, which would produce a graph nobody chose and nothing
could falsify; a concept is here only because some page already declares
it, in its frontmatter keywords, in a glossary, or as a layer of the
atlas. The dashed threads between two concepts are hand written, and
each one carries a note saying why the relation holds.
The web is therefore honest about thin cross-linking: where a corner of
the site connects to nothing, it hangs on one thread and looks like it.
A page with no thread arriving at all is drawn inside a dashed ring,
which is the one thing here that is a verdict rather than a picture: it
says no other page's body links there, so a reader following the writing
will never find it. The masthead and the footer are not threads, so a
marked page is usually still one click from a menu; the legend under the
figure says exactly that. The full data contract, including what the
builder refuses to include and why, is in docs/gww.md.