Pages and nodes

Form follows weight, and weight changes

A node is for something carrying a claim and the argument for it, a list item for something carrying one line, a cell for something carrying a word. A table is for comparing values along its rows, so a set of entities belongs in nodes rather than in its rows. Weight changes as a subject grows and shrinks, so things cross those lines in both directions: a bullet that swelled into paragraphs becomes a node, a list that outgrew the name above it becomes sections, and several nodes each holding one sentence become one list.

A claim hidden inside a body has no address and never reaches the outline. It hides as a bold lead-in, a bullet running for paragraphs, a row in a catalogue table, or a second claim in the same block; a ## heading typed into a body is the same thing. It hides in plain prose too, with no marker at all: a page of correctly named sections and no children under any of them holds every claim it makes inside a body, and a reader scanning its outline learns none of them. The inverse is a heading over a single sentence, spending an outline row and a screenful of whitespace on what a bullet said.

Both failures show from outside. On a page in shape, every claim a body makes has its row in the outline. And a body that, with its title covered, reads as a footnote to its own heading is a list item.

The title carries exactly what the body holds

At h2 the title names a section, a noun phrase of a few words: the top level is read as a table of contents, and a sentence there competes with the page's own title. Deeper, the title states the claim and the body argues it, which is what lets a reader take the claim off the outline and open the body only for the argument. A node that groups children names a section at any depth.

The sections give the claims their shape. Without them a page is a row of h2 statements, the kind a single writing pass leaves behind, and has to be read whole to find anything.

What earns a page

A subject is a page when a reader would arrive at it directly, and a block when it means nothing away from its parent. The same test divides a subject that has grown: past a couple of dozen outline rows, or a few screens of body, a page is scrolled rather than read and its contents rail stops being a map, so a section that can be opened cold is a page of its own, and sections that make sense only in sequence stay short instead.

A page, shown

Titles alone, which is how a reader meets a page:

Visa rules
  Who needs one
    Nationals of the EU do not
    A transit stop counts as entry
  Applying
    The form is filed online only
    Fees are paid at the consulate

The two h2s name the sections. The four claims beneath them are what the page holds, so a reader takes a claim off the outline and opens a body only for the argument. Under the third title, the body argues it and says nothing the title said already:

Immigration counts an airside transfer as an entry, so a traveller changing planes needs the same visa as one leaving the airport. The airline enforces it at check-in, which is where a missing visa is usually discovered.

A page with children sells them rather than lists them

The home page and every page with child pages open with what is worth reading below and link a few pages readers come for, even deep ones: featured navigation, not a copy of the tree, which the reader already has beside the text. A list of every child belongs only where the reader must choose among all of them, as with ways to install. Such a page has sections of its own rather than a lead alone: without them it has no contents rail and reads as unfinished.

A body that restates the engine's record goes stale

Every node carries its own history, its own updated_at, its place in the contents rail, and the list of nodes linking to it. A body repeating any of that goes stale the moment someone edits the page or its structure without updating that sentence, and then it misreports.

So a sound body holds none of it: no block documenting its own obsolescence, because history holds what it replaced; no "checked in April" stamp, because it goes stale on its own and says nothing reliable about whether the content is current; no prose listing the sections below, because the rail lists them; no "back to X" footer, because the backlinks are on the page already. What decides is where a sentence gets its truth from. From the tree of this page, and it goes, because the reader sees that tree and it changes without asking the sentence. From the subject, and it stays, even where its number happens to match the tree today: a founding year is content, a review date is metadata, and three laws of robotics stay three whatever this wiki does.

How a reference reaches a node

A short ref, page:N or block:N, points at the node rather than at the words in its heading. A block ref names a claim, a page ref names a subject. It does not cross a wiki boundary: another wiki is reached by its URL and behaves as an external link, because a ref that crossed would resolve into wikis the reader cannot open.