How the panels are distributed

The first counts panels in each section. The second sums the words in each section, including the section page itself. Both are computed by walking the page registry at request time.

They tell slightly different stories, and the difference between them is the interesting part. Equal panel counts with unequal word counts means some subjects produced longer panels, which is a real property of the material rather than a plan.

Because the board was designed that way. Each section was given a target and the panels were written to fill it. That is worth saying plainly, because an even distribution has a way of looking like a finding: as though the subject naturally divided into equal parts.

It did not. Some of these sections could support many more panels and some had to be worked at to reach the target. The evenness is a shape imposed on the material, and a reader who takes it as evidence of anything about the subject is reading a decision as a discovery.

Where the material was thicker. A section whose panels ran long is a section where there was more to say per subject, and that is at least a weak signal about the material rather than about the plan.

It is still weak. Longer can mean richer and it can mean less disciplined, and nothing in a word count separates those. This is the same caveat as on the panel about how much is here, and it applies with full force.

Money and orders are the sections where this board has no data at all. Every tile on every panel in them measures either the onion strings or the page you are on, and the prose is structural rather than specific.

They were kept anyway, at the same size as the others, because they are the subjects readers actually arrive with. Dropping them would have made the board tidier and less useful, and keeping them meant writing panels that say clearly what they do not know. That is an uncomfortable shape and it is the honest one.

A section is a slug, a label and a page. Any panel that declares that slug belongs to it, and the counts above follow from that with no manual list anywhere.

This is why the numbers cannot drift. There is no place where a section size is recorded, so there is nothing to become wrong. Adding a panel to a section changes the bars on this page by itself, on the next request.

• Whether a section is complete. Completeness is a judgement and no count reaches it.
• Whether the divisions are the right ones. A different board would divide the same material differently and produce equally tidy bars.
• Whether a large section is more important. Size here was decided in advance.
• Anything at all about the market. These bars describe a set of text files.

It is worth imagining the alternative for a moment, because it makes the arbitrariness concrete. This board is divided by subject: strings, reading, route, session, money, orders, and itself. That is one cut.

Another board could divide the same material by when a reader needs it: before you start, while something is loading, while something is waiting, after something went wrong. A third could divide it by how confident the answers are, which would put nearly everything from the money and orders sections in one bucket and the string arithmetic in another.

Each of those would produce a tidy set of sections with a plausible looking distribution, and none of them would be more correct. The bars on this page describe a decision somebody made, and there is no version of this measurement that describes the subject itself.