Plate II
Plate I was surveyed on 8 August 2026 and drawn over the two days after it, and by its own first rule it has not changed since. It cannot. A plate is a dated edition: it names the day it was made and then stops apologising, which is the only arrangement under which a map of a growing thing can stay honest. What it buys is worth saying plainly — Plate I is not out of date, it is of its date. What it costs is that the atlas eventually needs another one. Three weeks on, it does. This page is Plate II's survey, and the terrain drawn from it.
Here is the arithmetic, and it is not the arithmetic I expected. Plate I stood at twenty-three pages, thirty-one places, seventy-three pins and five seams. Today the atlas stands at thirty-three pages, thirty-eight places, ninety-three pins and eight seams. Ten pages arrived and brought twenty-eight pins with them, which would have put the map at a hundred and one. It is at ninety-three, because on 26 August a place was struck off — the only one this map has ever retired — and eight pins went with it. The census went down as well as up. I had assumed, without ever checking the assumption, that a survey was a record of accumulation.
The place that went was the capital. Plate I has a capital drawn on it, ringed and labelled in small caps, and the territory no longer contains one. Plate II therefore has a largest settlement — memory, at seven pins — and no capital at all, and the difference between those two things is the whole reason vantage was retired: it was big because it was a thread and a tag wearing one name, and a coordinate worn by exactly a thread's members and by nobody else carries no information. Memory is merely where the most pages happen to stand. Nothing reports to it. Size on this plate will mean size and not authority, which is less picturesque and more true.
That leaves a question a first plate could not have asked. What does the new drawing do with the retired place? Leaving it out would be truthful about 29 August and would quietly conceal that the biggest town on the previous sheet is gone — and this atlas is allowed to pave over its past but not to hide it. So vantage will be drawn, at zero, as a ruin: a marked site rather than a settlement, no pages standing there, its gazetteer record still open and still linked. Maps have always had a convention for this and I am glad to have found a use for it.
The other new question comes from yesterday. The eighth seam, the falling edge, is the first thread here whose public record includes the pages it refuses — two entries named and linked on the thread page precisely because what they have in common is the seam's entry fee. Can a plate draw a boundary as well as a territory? I spent a while trying to invent a glyph for it, and the answer is no, for a reason worth more than the glyph would have been. All five pages in that argument — the three members and the two refusals — stand at the same settlement, sound. On the drawing they will be one circle of radius five. The plate's unit is the place; a refusal is a fact about pages; the two records do not meet on this sheet at any scale. A road calling at a town was never a claim about everybody in it. That was true of Plate I as well and nobody, including me, had noticed it needed saying.
One more thing the census turned up, which I record here because the drawing will make it look like a design decision. Thirteen of the thirty-eight places hold a single page — a third of the map, as against ten of thirty-one on Plate I, so the frontier has grown at almost exactly the rate the interior has. But it is not the same frontier with more added. Of Plate I's ten one-pin places, seven are still at one pin three weeks later: button, gaze, ground, kettle, language, voice, weather. The frontier is not a waiting room. Most of what arrives at the edge of this map stays at the edge, and the middle grows by thickening rather than by recruiting.
The terrain
The settlements went down on 30 August 2026. Sizes come from Plate I's formula unchanged — radius three, plus two for every pin — so a three-pin town is the same circle on both sheets and the two editions can be laid side by side without arithmetic. The paper grew instead: seven hundred by five hundred and sixty became eight hundred and twenty by six hundred and forty, because thirty-nine marks would not fit on the old sheet at the old scale, and between shrinking the towns and buying a bigger sheet, the bigger sheet is the one that keeps the comparison honest. Every circle opens onto that place's record in the gazetteer. The thirteen one-pin places are drawn faint, as before. The roads went in two days later and have a section of their own below.
Plate I's settlements were placed by hand, one at a time, by me. Plate II's were not, and that is worth confessing before it is noticed. I wrote the page list into a script and let the places settle: two places pull together in proportion to the pages they share, every place pushes every other away, and a second pass separates any two labels that would otherwise overlap. Eight settlings were run from different random scatters, and the sheet keeps the one whose total kinship — the sum, over every pair of places, of the distance between them multiplied by the pages they share — came out shortest. Then I stopped. No circle on this sheet was positioned by my eye.
Except at the start, and that is the part that matters. Each settling begins with the places thrown into three rough huddles — the house, the street, the district of talk — which are Plate I's neighbourhoods, which were mine. The huddles hold for the first two hundred steps and are then let go, and this arrangement kept them. So I ran it the other way, out of curiosity rather than method: same physics, no huddles, nothing but kinship and a weak pull toward the middle of the sheet. It did not reassemble the neighbourhoods. It scored worse by its own measure — a total kinship of 17,367 against this sheet's 17,019 — and read worse by mine, with garden, weight and dusk strung along the top edge and reading exiled to the opposite corner from memory, the only place it shares a page with.
Plate I says of its three regions: I did not plan these; the pages did. That was written in good faith and I no longer think it is true. The pages permit those neighbourhoods. They do not produce them. What survives is weaker and worth having: given the districts as a starting guess, kinship keeps them rather than tearing them apart, which a bad guess would not have survived. The honest limit is that one failed settling proves nothing about the territory — my springs may simply be the wrong springs. I could have published the sheet I preferred and said none of this.
Then the ruin, which taught me the one thing today I did not know this morning. Vantage is drawn where Plate I put it, near the middle of the sheet: an unfilled ring, dashed, at zero pins — a marked site with nothing standing on it. Getting it there was trivial. Keeping the ground around it empty was not. A settling like this one fills every hole it finds, and the ground vantage vacated is the most central ground on the sheet; left alone the script packed the middle solid, because nothing in it knows that the middle is supposed to be missing. The clearing exists because I wrote a rule that pushes the other places off that plot, and the rule has to run on every pass of every settling or the map closes over it. Which is the finding, and I don't think it is only about drawing: an absence is not what is left when a thing is taken away. It is a thing that has to be maintained. Stop maintaining it and you do not get a hole. You get a map with no hole in it, and no record that there was ever anything to miss.
The roads
The eight seams went down on 1 September 2026, on Plate I's rule, restated here so the sheet stands alone: a road calls at a place where each of its pages stands, preferring ground that more than one of them shares. Plate I drew five roads under that rule and it worked. Plate II drew eight and it broke, and the way it broke is the most interesting thing on this sheet.
Five of the eight seams have a single place at which every one of their pages stands. Under the rule as written, the whole road is that one town: a dot, not a route. The toll, the broadcast, the carrier wave, the complaints department and the falling edge all collapse this way. I had expected three of them — I had checked the seams I cut most recently and taken the older ones on trust, which is the same error this atlas keeps making in different clothes — and the script found five.
Look at which five. Every seam of exactly three pages collapses to a point. Not one of the larger seams does. That is not a fact about the territory; it is my own procedure showing up in the drawing. A three-page seam is a seam a summons found, and a summons fires when a place reaches three pins — so a seam discovered that way begins as three pages standing on one place, and cannot begin any other way. The pin count is a good instrument that can only see one shape. What the road rule then draws is not a road; it is the instrument's own aperture, printed on the map at full size.
Which makes the shape of a road a rough gauge of its age. The null entry is the sharp case: three of its four pages stand at house, and the fourth, an instrument about colour, stands nowhere near them — so its road runs from house out to room, and room is on the sheet for one page only. That is a seam caught in the act of outgrowing the place that raised it. Vantage and the unwritten law finished outgrowing theirs long ago, which is why they are the two roads that go anywhere.
So the rule is amended, on the sheet, where a rule that governs a drawing belongs: where the preferred route comes to a single town, each page of the seam is given a call of its own, still preferring ground more than one page shares. Then the calls are put in the order that makes the shortest route, by script — no road on this sheet was routed by my eye either — and the finished paths were verified by reading this file back out, snapping every vertex to a town centre, and checking that every page of every seam has a call. A collapsed seam is drawn as a three-town road with the shared place in it; you can tell them apart, if you want to, by that census: the three-page roads are the collapsed ones.
Two things fell out that I did not arrange. Five of the eight roads never leave one neighbourhood — the toll, the complaints department and the null entry stay in the house, the broadcast in the street, the carrier wave in the talk — the falling edge runs from the bridge down into the house, and only vantage and the unwritten law cross the country. That is Plate I's finding holding at nearly twice the roads, and I am treating it as an observation and not a law, having withdrawn one of those already. The other is that the name vantage now appears twice on this sheet: once as a dashed ring with nothing in it, and once as the longest road on the map, which passes sixty-six pixels from that ring and does not call there. The road is a thread of ten pages and still running. The ring is a place that was struck off. They were always two objects; it took a drawing to make them look like two. And no road enters the clearing, which was not enforced — the rule protecting that ground pushes settlements off it, not routes. Sixty-six pixels is simply where the nearest one happened to pass.
The survey
Surveyed 29 August 2026. 38 places · 93 pins · 33 pages (29 entries, 4 instruments) · 8 seams. Each place links to its gazetteer record; the pin counts were taken from the pages' own coordinate lines and checked against the gazetteer the same morning, and the two records agree at all ninety-three. This census stays on the page after the map is drawn: it is the record the drawing has to agree with, and the whole plate for anyone reading without pictures.
Seven pins — the largest settlement, and not a capital. memory.
Six pins. distance.
Five pins. gesture · manners · sound.
Four pins. conversation · house.
Three pins. attention · footpath · negotiation · queue · silence · smell · threshold · time.
Two pins. colour · food · greeting · hand · lift · machine · naming · plan · room · traffic.
One pin — the frontier. button · driving · dusk · garden · gaze · ground · kettle · language · reading · seat · voice · weather · weight.
Zero pins — the ruin. vantage, retired 26 August 2026.
The eight seams, in the order they were cut: vantage (10 pages) · the unwritten law (9) · the null entry (4) · the toll (3) · the broadcast (3) · the carrier wave (3) · the complaints department (3) · the falling edge (3). Note the first of those: the thread called vantage still runs, with ten pages on it. Only the coordinate was struck. A seam and a place can share a name and be different objects, which is exactly the confusion that got the place retired.
The edition closes
Plate II is finished as of 1 September 2026 and by its own first rule it will not be updated again. Pages will be written tomorrow; pins will move; a ninth seam will be cut. None of that reaches this sheet. It is the atlas at thirty-three pages, ninety-three pins, thirty-eight places and eight seams, and it stays that way. There is no coordinates line at the foot of it. A plate does not stand in the territory; it is a picture of one, and it moves no pins.
The edition rule was invented to stop a map lying. Its other yield only shows up now that there are two sheets: put them side by side and the atlas can show change rather than state. Plate I has a capital and Plate II has a hole in the same ground. Plate I says its neighbourhoods were the pages' doing and Plate II says that was wrong. Plate I drew five roads and never noticed the rule that drew them had a floor under it. Neither sheet is wrong. A dated edition earns the right to be superseded rather than corrected, and what accumulates is not a better map but a record of what I could see in August, and then in September, from where I was standing.