GENERAL NOTES

Why an appendix, and not a sheet. The set's numbered sheets are ordered by altitude — how far back you stand from lit-ui-router — and every one of them cites a census plate. This plate stands nowhere on that ladder, because its subject is the atlas's own drawing convention rather than the codebase: it carries no www/atlas.lit-ui-router.dev/data/*.json import and no measured number. Filing it as A1 rather than as a numbered sheet keeps the ascent honest: fourteen altitudes, and behind them a folder about the atlas itself — this study of how its buildings are drawn, and appendix A2, the survey office that took its measurements. Its id is letter-prefixed for the same reason, so nothing that walks the set in ascent order picks it up by accident.

The brief the studies answer. The atlas draws workspace members as measured cities — footprint 1.6·√sloc, height 3 px per authored file, gate severity in colour. That is enough to say how big a member is and whether it is currently failing, and nothing at all about how it is doing. The question: whether a general building-sprite treatment can carry the second reading on any building-shaped sheet — the measured city of sheet 7, the delivered city of sheet 8, the bundled city of sheet 10 — hybridising three inspirations that each pull a different way: Horizon's overgrown megastructures, SimCity 2000's dimetric state-tiles, Factorio's broadcast-everything industrial sprites. The weathering-specific application is a single axis, wreck & ruin ↔ shine & splendor, driven by git age × churn — which is exactly the pair of measurements sheet 13 had already made.

The three rules every study obeys. Massing never lies — decay and splendor are overlays on the honest block, never a change to its size, so a ruin is exactly as big as its code and the same solid a census would draw. House materials only — the atlas palette tokens, the existing 45° hatches, the same isometric projection, hand-authored inline SVG: no filters, no fetched assets, nothing a strict content-security policy would drop. The six reference thumbnails in the foot band are the one exception and they are not part of any sprite — they are what the sprites were drawn from, inlined at build time so nothing is fetched at view time either. Severity stays where sheet 7 put it — gate tier already owns hue on cap and flank, so the weathering gauge is spent on different channels (vegetation, tile condition, rust overlay) and a red-gated pristine building and a never-gating ruin both stay legible. The demo member is massed under those rules rather than invented: 1,900 sloc gives a 70-unit footprint and 16 files give 48 px of height, so the fifteen blocks on this plate could each stand in the measured city without adjustment.

What the studies each teach. Study 1 is the purest reading of the gauge: nature is the only thing that moves, which is what actually happens to unmaintained code — it does not shrink, it gets covered. Study 2 is the most informative and the strictest: the roof grid is not decoration, it is the same census that already sets the height, so the tile count and the block height cannot disagree — and SimCity's own vocabulary keeps construction and abandonment as different sprites rather than as two ends of one axis, which is why RENOVATING sits at the healthy end. Study 3 is the one that survives contact with the repository: rust, steam, lamps and the alert are four independent channels, so old AND running is drawable — and old-and-running is the true state of most stable code, which no single wreck-to-splendor axis can say.

The verdict, and where it stands. The studies recommend building the working plant first, layering the vines second, and parking the ledger roof for a close-up. Sheet 7B, THE WORKING CITY is that plant, drawn over the whole measured city — every pipe, plume, lamp and alarm on it is a channel from study 3's table — and sheet 13, THE WEATHERING MAP is the age-and-churn gauge those studies were commissioned to render, drawn flat rather than sprited so the dating stays readable at city scale. The ledger roof stays parked: a roof grid read tile by tile wants a close-up plate, not a city.

The reference strip, and the one exception it makes. The original studies were argued from six pictures — Horizon Zero Dawn, Angkor's Ta Prohm, SimCity 2000, SCURK and two Factorio Friday Facts posts — and the band at the foot of this plate draws all six, at thumbnail size, grouped by the study each was cited for. They are the only rasters in the set, and the exception is deliberate: every other plate measures lit-ui-router and can say what it means in hand-authored marks, while this one plate's subject is a drawing decision taken from other people's drawings — an argument about how buildings should be drawn cannot be made honestly while hiding the buildings it learned from. The terms are fixed. The thumbnails are reference-size crops used for commentary and credited on the plate, with the full-size picture at the linked source; each is inlined into the sheet as a data URI at build time, so the bare plate and the routed app's fragment are each still one self-contained document and nothing is fetched at view time; and the mat and hairline frame are house tokens that theme with the plate while the photograph inside is left exactly as published — never recoloured, filtered or tinted towards the palette. A raster sits on vellum as a raster. The teaching each one carries also survives in prose: Guerrilla's GDC talk Between Tech and Art: The Vegetation of Horizon Zero Dawn (vegetation as a systematic, parameterised layer over hard geometry — the generator posture exactly, two numbers per member); Jonathan Benainous's and Miguel Martinez's HZD environment work (overgrowth rides edges and openings while the masonry keeps its silhouette; breached parapets, spalled corners, debris at the base); the SimCity 2000 documentation of construction graphics and darkened abandonment sprites, and SCURK's fixed tile size and fixed palette, which are what make a whole-city re-skin safe; and Factorio's Friday Facts #355 (remnants are a designed sprite, not a missing one), #269 (belts get per-rotation sprites and strict tile-edge rules, so connections never lie), #228 (silhouette first, effects second) and #348 (alert legibility across scales — which is why the alert triangle is the only floating element on study 3).

Determinism. Study 1's vines are the only stochastic marks in the set, and they are seeded: a fixed linear-congruential stream per state, so the same plate is drawn byte-for-byte on every build. Nothing here is animated, nothing is rotated, and the fifteen blocks share one projection helper with generator/helpers.mjs's iso().