The tiles, counted

A tile names a metric. It never carries a value. When a page is built, the metric is looked up, the computation runs, and whatever comes back is drawn. If nothing comes back, the tile draws an empty state.

That means no number on this site can be typed by hand into a content file, because content files have nowhere to put one. It is a structural guarantee rather than a promise about discipline, and structural guarantees survive tiredness.

The registry is walked, tile definitions are counted, and the total is the first tile above. It includes the tiles on this page, which is a small recursion that resolves without trouble because the count is of definitions rather than of rendered output.

The second tile counts pages, and dividing one by the other gives the average strip size, which is a figure the board could show and does not, because two tiles that differ only by a division is padding.

• A plain number, for quantities where a picture would add nothing.
• A small line, for a series along one address in equal blocks.
• A half circle, for a proportion of a known whole.
• A row of bars, for comparing a handful of values.
• A grid of symbols, used only for the base32 alphabet, where each cell is one symbol shaded by how many addresses carry it.

On several panels, particularly in the money and orders sections, the tiles measure the page you are reading rather than the subject of the panel. That is visible and slightly awkward, and it is preferred to the alternative.

The alternative would be to find a plausible number for every subject, which for those subjects would mean inventing one. A board where every strip looks appropriate is a board that has stopped distinguishing between measurements it has and measurements it wanted.

When a metric returns nothing, the tile draws a small outlined icon and a sentence saying what came back empty. It does not disappear, it does not fall back to zero presented as a result, and the strip does not close up around the gap.

This matters most where the empty result is the real answer. Two of the repetition measures on the repetition panel genuinely find nothing in any of the three addresses, and the panel on empty tiles is about that case in full.

The last tile here is the alphabet grid, which is the most information dense thing the board draws. Each cell is one of the symbols the format allows, shaded by how many of the three published addresses contain it.

It is on this panel as an example rather than because it belongs here. A panel about tiles should show the most interesting one, and this is it.

It would be easy to add more shapes. A donut, a stacked bar, a small map, a heat strip. Each would be a few lines of drawing code and each would make some panel look richer.

They are not here because a chart type has to earn its place by fitting the data rather than by looking good in a strip. Every measurement this board produces is one of four shapes: a single quantity, a proportion of a known whole, a handful of comparable values, or a series along one address. Five drawing types cover those with one to spare.

Adding a sixth would mean either finding data to suit it or bending existing data into it, and both of those are the same mistake in different clothes. The restraint is not aesthetic, it is the same rule as everywhere else on the board arriving in a different form.