decisions.md

Blame: the commit that last changed each line · file · history

`866c7f96` 2026-09-07 Pavel Safronov	1	# Tower — Decisions
	2
	3	Standing record of choices made and why, so they can be revisited deliberately
	4	rather than rediscovered. Newest sections append at the end.
	5
	6	---
	7
	8	## D-001 — Cylinder geometry: r = 50 km, L = 540 km
	9
	10	Decided. Radius 50 km, length 540 km, giving 169,646 km² of inner surface.
	11
	12	Considered three shapes, all matching Wisconsin's area at 1 g:
	13
	14	\| \| radius \| length \| spin period \| character \|
	15	\|---\|---\|---\|---\|---\|
	16	\| Long & narrow \| 8 km \| 3,375 km \| 3.0 min \| ceiling, not sky; absurd aspect ratio \|
	17	\| Balanced \| 50 km \| 540 km \| 7.5 min \| far side visible overhead through haze \|
	18	\| Wide & short \| 150 km \| 180 km \| 13.0 min \| a drum; far cap looms as a wall \|
	19
	20	Why balanced. It is the only one where the sky reads as sky. At 8 km the
	21	roof is 16 km up and oppressive; at 150 km the far side is 300 km away and lost
	22	entirely. At 50 km the opposite land is 100 km overhead — far enough to be pale
	23	and dreamlike, near enough to be unmistakably land above you. That image is the
	24	point of the whole project, and this geometry is what produces it.
	25
	26	The 7.5-minute spin also sits in a good window: fast enough that star motion is
	27	obvious in a few seconds of watching, slow enough not to look like a fairground
	28	ride.
	29
	30	---
	31
	32	## D-002 — Slats sized so land area equals Wisconsin's land area
	33
	34	Decided. 8 slats, 6.80 km wide, leaving 140,268 km² of land.
	35
	36	The brief specified a Wisconsin-sized inner surface. Wisconsin has two areas —
	37	total 169,635 km² and land 140,268 km². Sizing the slats to consume exactly the
	38	difference makes both true at once: **Tower's total inner surface is Wisconsin's
	39	total area, and Tower's habitable land is Wisconsin's land area. The slats are its
	40	lakes.**
	41
	42	This is a coincidence discovered while computing, not a constraint imposed. It
	43	costs nothing and it is too good to pass up.
	44
	45	*Superseded by D-017: there are no slats, so Tower's land is its entire inner
	46	surface, 169,646 km² — Wisconsin's total area. The second coincidence recorded
	47	below no longer applies.*
	48
`cf37225c` 2026-09-07 Pavel Safronov	49	Slat count is left as a tunable constant. *The runtime control is disabled and
	50	hidden as of D-033; the derivation below is unchanged and still tested.* The width
	51	follows from the count:
`866c7f96` 2026-09-07 Pavel Safronov	52
	53	\| slats \| each \| land strips between \|
	54	\|---\|---\|---\|
	55	\| 6 \| 9.07 km \| 43.3 km \|
	56	\| 8 \| 6.80 km \| 32.5 km \|
	57	\| 12 \| 4.53 km \| 21.7 km \|
	58	\| 24 \| 2.27 km \| 10.8 km \|
	59
	60	8 chosen because 32.5 km strips read as regions — a day's travel across, room
	61	for real geography. At 24 the world becomes narrow ribbons where every point is
	62	within 5 km of a wall of stars, which is striking but claustrophobic. Revisit
	63	after seeing the map view; this is cheap to change.
	64
	65	*"Revisit after seeing the map view" is now literal — D-014 makes the count a
	66	runtime control from 0 to 16, and 8 is the default rather than a fixed value.*
	67
	68	---
	69
	70	## D-003 — Widen the beam to 10 km; the physics now works unbroken
	71
	72	Decided. The beam is 10 km in circumference, carrying 54.3 TW at 5,000×
	73	concentration. No physical law is bent. *This supersedes an earlier version of
	74	D-003 that accepted an étendue violation in order to keep a 1 km beam.*
	75
	76	The mistake in the first pass. I treated the 1 km beam as fixed and concluded
	77	that Earth-like brightness was impossible without an active, physics-breaking
	78	waveguide. That was wrong, because it conflated two independent parameters:
	79
	80	- Collector aperture sets how much power is gathered. This was never limited
	81	by anything but engineering scale.
	82	- Beam cross-section sets how hard the light must be concentrated to fit
	83	through. This is where étendue binds — passive optics cannot exceed the sun's own
	84	surface radiance, ~46,223× at 1 AU, or 62.9 MW/m².
	85
	86	Holding the beam at 1 km forced 313,000× and broke the second law. Widening the
	87	beam relaxes the concentration requirement quadratically and costs nothing else.
	88
	89	Where the number came from. Targeting 54.3 TW — Earth clear-sky noon at local
	90	midday, see design §1.3 — against candidate concentrations:
	91
	92	\| concentration \| beam circumference needed \| status \|
	93	\|---\|---\|---\|
	94	\| 46,223× (étendue limit) \| 3.29 km \| thermodynamically impossible to exceed \|
	95	\| 10,000× \| 7.08 km \| beyond current practice \|
	96	\| 5,000× \| 10.01 km \| Odeillo solar furnace reaches ~10 MW/m² today \|
	97	\| 1,000× \| 22.39 km \| routine, but a conspicuously fat beam \|
	98
	99	5,000× chosen. It lands the beam on a round 10 km, it is demonstrated
	100	terrestrial technology rather than an extrapolation, and it sits at 11 % of the
	101	étendue limit — 9.2× of headroom in reserve. Diffuser mirrors at 99.9 %
	102	reflectivity reach thermal equilibrium at 316 °C in that beam, an ordinary
	103	radiator problem rather than an exotic one.
	104
	105	Cost to the imagery: none. At 1,592 m radius seen from 50 km below, the beam is
	106	3.65° wide — about seven sun-widths. It still reads as a luminous rod running the
	107	length of the sky, which is the image the 1 km figure was protecting.
	108
	109	The unlocked consequence. Because Tower now has surplus capability rather
	110	than a deficit, brightness becomes a setting rather than a fixed value. The
	111	diffusers are software-controlled; illumination can run anywhere from 30 W/m²
	112	permanent dusk to 320 W/m² clear-sky noon. The "embrace the dimness" option
	113	weighed and rejected in the first pass is no longer an alternative to this
	114	decision — it is a position on its slider. The app should expose that control.
	115
	116	## D-004 — Terminator orbit, actively precessed
	117
	118	Decided. Orbital plane held normal to the Earth–Sun line, precessing 360°/year.
	119
	120	The brief requires that one cap face the sun permanently. Any orbit whose plane
	121	contains the sun line puts Tower through Earth's shadow once per revolution, which
	122	would interrupt the beam. A terminator orbit never enters shadow.
	123
	124	The cost is that the plane must be reoriented ~1°/day against the gyroscopic
	125	stiffness of a spinning 540 km cylinder. This is enormous and continuous. Canon
	126	adopts it as a feature: Tower's existence is a permanent act of maintenance, not a
	127	stable configuration. That is more interesting than a free lunch.
	128
	129	---
	130
	131	## D-005 — Vanilla TypeScript, no UI framework
	132
	133	Decided. Vite + TypeScript + Three.js, no React/Vue/Svelte.
	134
	135	The app is three canvases, a picker and a control panel, sharing one small state
	136	object. A framework would add build weight and a rendering model that fights
	137	Three.js's own loop, in exchange for conveniences this UI does not need. Revisit
	138	only if the control panel grows genuinely complex.
	139
	140	Canvas 2D rather than Three.js for the map: it is a flat unrolled rectangle with no
	141	projection distortion, and 2D canvas draws that natively.
	142
	143	---
	144
	145	## D-006 — Interior fidelity: real geometry, simple ground
	146
	147	Decided. The shell, curvature, haze, beam, diffusers, slats and stars are all
	148	genuinely modelled and correct. The ground is procedural terrain and texture only —
	149	no scenery meshes, no settlements, no vegetation.
	150
	151	The illusion of standing inside Tower comes almost entirely from the sky: land
	152	curving up into blue, the lit shaft overhead, stars wheeling past every 7.5
	153	minutes. Detailed ground scenery adds cost proportional to realism and contributes
	154	little to that illusion. Deferred, not abandoned.
	155
	156	---
	157
	158	## D-007 — All canon derived from four constants
	159
	160	Decided. `canon/constants.ts` exposes radius, length, slat count and target
	161	land area as free parameters. Everything else — spin period, rim speed, Coriolis
	162	parameter, slat width, angular spacing, axis pressure, beam power — is computed.
	163
	164	The alternative, transcribing the figures in `design.md` as literals, would drift
	165	the moment any parameter is revisited (and D-002 and D-003 both flag likely
	166	revisits). Derivation makes the whole app answer coherently to a changed radius.
	167
	168	---
	169
	170	## D-008 — Diffusers are a 1.2 cm swarm; the beam glows from air, not from them
	171
	172	Decided. 78 billion 1.2 cm mirrorlets, and the beam's visible glow is
	173	atmospheric scattering that does not depend on them.
	174
	175	The question. Would a modest number of large diffusers read as a string of
	176	separate suns rather than a continuous glowing rod?
	177
	178	Yes. At 50 km, naked-eye resolution is 14.5 m; anything larger shows a visible
	179	disc. But shrinking below 14.5 m is not sufficient, which was the non-obvious part.
	180	A flat mirror preserves radiance, so an aimed mirrorlet shows the *sun's own
	181	surface brightness* over a small angle — unresolved but blinding, exactly as stars
	182	are unresolved and plainly visible. Magnitudes: 1 m → −13.6 (brighter than a full
	183	moon), 10 cm → −8.6, 1.2 cm → −4.0, which finally sits under the daylight
	184	visibility threshold. The binding constraint is flux, not angular size, and it is
	185	two orders of magnitude tighter than the resolution limit implies.
	186
	187	The better answer, found while checking. The beam glows regardless. Rayleigh
	188	scattering of a 6.8 MW/m² beam through 3.18 km of 5.2 kPa axis air gives 779
	189	W/m²/sr — 26× blue sky, 25,700× below the solar disc. The continuous-glow imagery
	190	is therefore a robust property of beam plus atmosphere, not a consequence of
	191	getting the diffuser design right. Canon leans on this; the swarm sizing is then
	192	only about keeping individual glints out of the sky.
	193
	194	Consequences taken. 11.3 km² of mirror at 1.2 cm is 78 billion units massing
	195	158 tonnes, spaced 3.8 m apart. Optical depth 1.41 leaves 24 % of the light
	196	reaching the far cap, which satisfies the brief's "shines to the far end" without
	197	stipulating it. *The passive drift-and-recirculate mechanism first recorded here is
	198	superseded by D-009 — it does not work.*
	199
	200	Story consequence. 78 billion independently aimable mirrors make illumination a
	201	render target rather than a schedule. Sunsets are compositing, not optics.
	202
	203	---
	204
	205	## D-009 — Tower flies every foil itself, because nothing else works
	206
	207	Decided. All 78 billion foils are under continuous individual control by Tower.
	208	There is no recirculation machinery at either cap. *This supersedes the passive
	209	drift mechanism in D-008.*
	210
	211	Origin. Proposed by the author as characterisation — Tower is an intelligent
	212	being, so it controls each foil directly and moves the swarm partly for its own
	213	enjoyment. Checking the physics turned this from a flourish into a requirement.
	214
	215	A passive swarm is impossible. D-008 assumed the axis was effectively
	216	weightless. It is not. At the beam's edge, 1,592 m out, g = ω²s = 0.312 m/s². A
	217	2 mg foil sinks outward at 0.26 m/s and falls clear of the beam in 1.7 hours. A
	218	drifting swarm dissolves in an afternoon. Continuous active flight is the only
	219	configuration that holds together at all.
	220
	221	They can easily afford it. Each foil intercepts 979 W; harvesting 0.1 % yields
	222	~1 W on a 2 mg body — 489,600 W/kg, ~50× a jet engine. Flying up-beam at 1.5 m/s
	223	costs 15 µW against 14,400 µW harvestable from ambient world-light: **a 1000×
	224	margin, without touching the beam.** So the foils fly home unaided, no structure is
	225	required at the far cap, and the exterior render is simpler than D-008 implied.
	226
	227	The interesting part is the slack. With 1000× of margin, efficiency is not a
	228	constraint on Tower's behaviour. It can move the swarm in spirals, shoals and
	229	patterns that cost energy for no reason except preference, and nothing about the
	230	world degrades. Play is not a luxury Tower affords despite the physics — it is
	231	budget the physics hands over for free. The weather is therefore not a system
	232	Tower operates but something closer to body language.
	233
	234	Falling out of this: aiming is human-scale. A 1.2 cm mirror is
	235	diffraction-limited to 4.6×10⁻⁵ rad, giving a 2.3 m spot at 50 km. Tower can
	236	place a glint on one person that a bystander five metres away cannot see. Offered
	237	to the author as a story mechanic; canon records the optics either way.
	238
	239	And: mirrorfall. A failed foil sinks out of the beam, takes ~2 days to descend,
	240	and lands at 0.48 m/s — snowflake speed. A permanent slow glitter falls out of the
	241	sky.
	242
	243	---
	244
	245	## D-010 — The app is a companion to a novel, not a tech demo
	246
	247	Decided. Tower is the setting of a novel the author is writing. The app serves
	248	that purpose first.
	249
	250	Two consequences, recorded because they change priorities rather than mechanics:
	251
	252	- Coherence outranks spectacle. These documents are series-bible material and
	253	the prose is downstream of them. A number that looks better but contradicts
	254	something is worse than a duller number that holds. This is already how the
	255	physics has been handled; it is now the stated rule.
	256	- Shareable permalinks to a point move into scope (phase 4, previously
	257	deferred). "Stand here and look at this" is the entire value of a companion
	258	piece, and a link is how a reader receives it. Named locations from the novel
	259	should be markable on the map for the same reason.
	260
	261	Not yet decided, and deliberately left open: whether the app should ever surface
	262	narrative content — place names, events, characters — or remain a pure physical
	263	model that the prose refers to from outside. That is the author's call and does not
	264	need answering to proceed.
	265
	266	---
	267
	268	## D-011 — Writer mode: the app is an instrument, and the world file is git
	269
	270	Decided. Two modes over one committed `world.json`. Readers get a finished map;
	271	the author gets editing plus a set of measuring instruments. No backend.
	272
	273	Origin. The author intends to use the app to write — placing things and asking
	274	questions like "can the far face be seen from the midpoint?", "where does a central
	275	embassy go?", "can a lake overhang?". That makes measurement, not rendering, the
	276	core feature. D-010 said the app serves the novel; this is what that costs.
	277
	278	Why those questions justify the build. Two of the three had non-obvious answers
	279	that only fell out of computation, and both became canon (§1.8): the far cap is
	280	invisible from the midpoint at 4 % transmission, and — unexpectedly — the far side
	281	of the world overhead is clearer than the ground at the same distance (78 % vs
	282	31 % at 100 km), because a sightline across the interior rises through thin axis
	283	air. An author cannot guess these. The instrument earns its place immediately.
	284
	285	Static, file-based, no backend. GitLab Pages serves static files, so: edit in
	286	browser → export → commit → publish. The world lives in git beside the prose,
	287	diffable and versioned, with no service to run or pay for. Drafts sit in
	288	`localStorage` between exports.
	289
	290	Coordinates in the file are canon (θ, z), never map pixels, so the world
	291	survives any change to how the map is drawn. Consistent with D-007.
	292
	293	Writer mode is a UI mode, not a security boundary. `?writer=1` is guessable and
	294	that is fine — an editor with no persistence and no backend can only alter the
	295	visitor's own copy. Recorded explicitly so it is never mistaken for access control
	296	and never load-bearing for privacy.
	297
	298	The real privacy risk is the repo, not the flag. Working notes must not ship to
	299	readers, but anything committed to a public repository is public whatever a
	300	`"public": false` field says. Hence two export files: `world.json` (public entries,
	301	committed) and `world.private.json` (everything, gitignored). Whether private notes
	302	should be version-controlled anyway — private repo, or a second one — is left to
	303	the author and does not block anything.
	304
	305	Sequenced early. The sightline probe lands in phase 6 but should be built first
	306	within it: it is useful the day it exists, and §1.8 is evidence the author is
	307	already asking questions only it can answer.
	308
	309	---
	310
	311	## D-012 — Ids follow the name until pinned by hand
	312
	313	Decided. A place's id is re-derived from its name while it still matches it,
	314	and freezes permanently once the author edits the id field.
	315
	316	The tension. design.md §3.2 makes ids permalink targets, which argues for
	317	never changing them. But places are created from a placeholder name ("New city"),
	318	so a never-changing id produces a world file of `new-city`, `new-city-2`,
	319	`new-city-7` — unreadable in git diffs and unusable for hand-editing, both of which
	320	D-011 counts as reasons for the file-based design in the first place.
	321
	322	Resolution. During world-building, readability wins; after publication,
	323	stability wins; and the author is the only one who knows which phase a place is in.
	324	So the id tracks the name automatically until they say otherwise, and the id field
	325	is shown in the editor so saying otherwise is one edit. `isAutoId()` in
	326	`src/world/schema.ts` is the test, and is unit-tested directly.
	327
	328	Found by running it. The first implementation derived the id once at creation
	329	and never updated it, which the browser check caught immediately — every exported
	330	place was called `new-city`. Two further bugs surfaced the same way: renaming
	331	blanked the editor (the store re-rendered before the selection moved to the new id),
	332	and clicking an existing marker while a creation tool was active dragged the old
	333	place instead of creating a new one. None were visible in the unit tests.
	334
	335	---
	336
	337	## D-013 — The map is bounded, and cut at a slat centre
	338
	339	Decided. The map shows the inner surface once, as a bordered rectangle, rather
	340	than tiling vertically. The cut falls at the centre of the last slat.
	341
	342	Why not tile. The circumference wraps, so tiling is the honest rendering and it
	343	makes places near the seam continuous. But it also makes the world look boundless:
	344	with no edge, the author cannot see the extent of what they are building, and
	345	repeated strip labels read as more strips rather than as repeats. For a map — a
	346	thing whose job is to show where things are relative to the whole — legible extent
	347	beats a hidden seam.
	348
	349	Why cut there. Any bounded map needs an edge, and the edge destroys whatever it
	350	falls on. Land strip 0 is centred on θ = 0, so the obvious cut would slice a whole
	351	strip in half across the top and bottom margins. Cutting at a slat centre instead
	352	leaves every land strip whole and splits only a band of stars, which carries no
	353	geography. This is exactly why Earth's maps are cut at the antimeridian. With zero
	354	slats there is nothing to protect, so the cut returns to θ = 0.
	355
	356	Consequences taken. Panning is clamped so the surface cannot be lost off-screen
	357	— without tiling there is nothing to pan back into. A sightline whose short route
	358	crosses the edge is drawn as two segments, entering and leaving opposite edges, so
	359	the drawn path always matches the measured distance rather than contradicting it.
	360
	361	---
	362
	363	## D-014 — Slat count is a runtime control, 0 to 16
	364
	365	Decided. `SLAT_COUNT` stops being a compile-time constant. The count lives in
	366	`src/canon/slats.ts`, adjustable from 0 to 16, persisted per browser, with every
	367	slat-derived figure computed from it.
	368
	369	Why. D-002 already called this "the most tunable number in canon" and said to
	370	revisit it after seeing the map. Making it a slider makes that revisiting real
	371	rather than a code edit, and the author is choosing the shape of a world rather
	372	than configuring software.
	373
	374	The invariant that makes it safe. D-002 fixes the total slat width from the
	375	land-area target, and that total does not depend on how many slats it is divided
	376	into. So **land area stays exactly Wisconsin's 140,268 km² at every count from 1 to
	377	16** — the coincidence survives the whole range, and only the per-slat width moves
	378	(18.1 km at 3 slats, 3.40 km at 16).
	379
	380	Zero is a genuinely different world, and the app says so. With no slats the
	381	land becomes the entire inner surface, 169,646 km², and nobody in Tower can ever
	382	see a star. That breaks the D-002 coincidence, so the control states it plainly
	383	instead of silently reporting a changed number. Rounding that away would have been
	384	the easy option and the wrong one — it is exactly the kind of quiet inconsistency
	385	the documentation gate exists to catch.
	386
	387	Cost. Slat geometry moved from constants to functions, which rippled through
	388	`geometry.ts`, the map, the instruments and the tests. Worth it: D-007's principle
	389	was always that canon is derived rather than transcribed, and this is that
	390	principle extended one parameter further.
	391
	392	---
	393
	394	## D-015 — Hydrology: shaped terrain, no sideways Coriolis, climate from the slats
	395
	396	Decided. Recorded as canon in design.md §1.9, and it amends §1.4.
	397
	398	The provable part. The floor's tangent plane is spanned by θ̂ and ẑ and the spin
	399	axis is ẑ, so surface flow gets `ẑ × ẑ = 0` axially and a purely radial `ẑ × θ̂ = −ŝ`
	400	circumferentially. The Coriolis force has no component inside the surface at any
	401	heading. Tower's rivers do not curve, and Earth's entire rotational repertoire —
	402	hemispheres, Baer's law, gyres, cyclones — is absent. This is not a simplification;
	403	it falls out of the geometry.
	404
	405	Where the force goes instead. Into vertical motion. Waterfalls lean retrograde
	406	as h^1.5 (133 m over a 1 km drop, 377 m over 2 km), and rain hangs at a permanent
	407	1.47° slant that never varies. Rising things lean the opposite way. A world where
	408	every falling thing drifts one way and every rising thing the other is a strong,
	409	cheap, always-visible signature.
	410
	411	The absence that matters more. Potential depends only on radius, so the floor is
	412	one equipotential and a perfect Tower has no rivers at all. Every river exists
	413	because the floor was shaped, there is no tectonic uplift, and erosion is therefore
	414	one-way — ~30 Myr to erase a 3 km range. Tower's hydrology is maintenance, which is
	415	D-004's theme arriving from a new direction.
	416
	417	An inconsistency found and fixed, not smoothed over. §1.4 specified a
	418	glass/diamond composite slat fill. Diamond conducts ~2,000 W/m·K; such a slat
	419	would radiate 385 W/m² to space, exceeding Tower's entire 320 W/m² mean insolation
	420	and freezing the world. Canon now specifies low-conductivity glass, which loses
	421	~28 W/m² through 5 m. Raised explicitly per the documentation gate rather than
	422	quietly retconned.
	423
	424	The gift in that loss. Tower has no latitude — every point is equidistant from
	425	the light, so there is no natural climate gradient whatsoever, which is a real
	426	worldbuilding hole. The leaking slats fill it: eight cool belts running the full
	427	540 km, parallel to the axis. Tower's climate bands run lengthwise, not around,
	428	and the slat count sets the climate as well as the geography — further support for
	429	D-014 making it a control.
	430
	431	---
	432
	433	## D-016 — Slat orientation: axial or rings — OPEN
	434
	435	Not decided. Canon keeps axial strips for now. Rings are a live alternative and
	436	this records what the choice costs, so it can be made deliberately.
	437
	438	Why it matters more than it looks. Slat layout turns out to set three things at
	439	once — the shape of Tower's countries, its entire climate map, and what its people
	440	can see of the universe. It is not a detail of the hull.
	441
	442	Geometry. At equal total area: rings are 11.69 km wide with 55.8 km land bands.
	443	Each land region is then 314 × 55.8 km rather than 32.5 × 540 km — the same area at
	444	aspect 5.6:1 instead of 16.6:1, a country instead of a corridor. Ring bands are also
	445	closed loops: you can walk 314 km around your whole region and come home without
	446	crossing a window. Axial strips are ribbons with two dead ends at the caps.
	447
	448	Climate. Axial strips put the cool belts lengthwise, so a 540 km river runs its
	449	whole length through identical country. Rings put them crosswise, giving Tower a
	450	genuine latitude analogue — a long river crosses all eight bands like the Nile.
	451
	452	The sky, which is the deep one. The spin axis is the celestial pole, so stars
	453	move along the celestial equator. Orientation decides whether a window lies across
	454	that motion or along it:
	455
	456	- Axial: 3.90° across the motion — each star crosses in 4.9 s, but the strip
	457	spans pole to pole so essentially the whole sky passes through, strobing.
	458	- Rings: 6.70° along the motion — a star rides the band horizon to horizon over
	459	3.7 minutes, but only the 5.8 % of sky within 3.35° of the celestial equator
	460	is ever visible. The rest is never seen at all.
	461
	462	Earth follows the same rule: a five-second flicker under axial slats, a slow bright
	463	traveller under rings. The sun is at the celestial pole and invisible through both.
	464
	465	The cost of rings. Hoop stress dominates the structure — ρv² is 785 MPa in a
	466	composite shell plus ~500 MPa of air pressure — and it runs along θ. Axial slats sit
	467	across that path, so ribs bridge the gap and the glass carries only local pressure.
	468	A ring runs along the path and is continuous in θ, so **the window itself must carry
	469	~1.2 GPa**. Glass cannot; it would need hoop cables through every pane, putting bars
	470	across every view forever. Characterful, but a real price axial slats do not pay.
	471
	472	Consequence for the app if rings are chosen. D-013 cuts the map at a slat centre
	473	because there is an empty θ to cut at. With rings there is not — every θ is land —
	474	so the map seam would have to slice through inhabited country, or the map would need
	475	reorienting. Worth knowing before the choice is made rather than after.
	476
	477	---
	478
	479	## D-017 — The hull is sealed; the only views out are at the caps
	480
	481	Decided by the author. No slats, in any orientation. The cylinder wall is
	482	opaque and unbroken for its full 540 km. Supersedes D-002's slat sizing, D-014's
	483	default, and closes D-016.
	484
	485	What it costs. The wheeling star field was canon's answer to "what is most alien
	486	about standing inside Tower" (§1.2) and the interior view's headline feature. It is
	487	gone. Tower also loses its only passive climate driver and its only natural source
	488	of geographic variation.
	489
`da6bcb0e` 2026-09-08 Pavel Safronov	490	*Partly recovered by D-045, and worth reading as a correction to the paragraph
	491	above rather than a reversal of the decision. The wall is still opaque and unbroken
	492	for its full 540 km — that is what D-017 decided and it stands. What came back is
	493	the pinwheel, through the far cap window, where §1.4 had always said it would be:
	494	one sky, at one end of the world, at night. It is no longer everywhere, which is
	495	the whole difference between the slats and a window.*
	496
`866c7f96` 2026-09-07 Pavel Safronov	497	What it buys, which is more.
	498
	499	- One unbroken surface. 169,646 km² of continuous land — Wisconsin's total
	500	area, the coincidence in §1.1 now standing alone and cleaner. Nothing divides the
	501	world into strips; you can walk anywhere, and a river can circle the whole 314 km
	502	circumference and close on itself.
	503	- The spin inverts. With no windows and a co-rotating far side, *nothing in the
	504	sky moves at all*. But rain still hangs at a permanent 1.47° slant and a 1 km
	505	waterfall still lands 133 m off its cliff. **The sky is still and the water is
	506	not** — a stranger signature than wheeling stars, and one that shows up in every
	507	puddle rather than only when you look up.
	508	- The caps become the only view, and it is a good one. A cap window looks along
	509	the spin axis — the celestial pole — so stars do not sweep but *rotate about a
	510	fixed centre*, one turn per 7.5 minutes, one motionless star at the middle. The
	511	caps are also walls rather than floors: gravity there is tangent to their surface,
	512	so each is a 50 km cliff at the end of the world with its windows at the base.
	513	- Nobody can ever see Earth. The terminator orbit (D-004) puts Earth ~90° off
	514	the axis, out the side, which is now solid hull. A habitat in Earth orbit whose
	515	people have never seen the planet they orbit.
	516
	517	The consequence the author accepted explicitly. With no latitude and no windows,
	518	nothing breaks the thermal uniformity, so there is no passive weather whatsoever.
	519	Every cloud, every rainfall and every jet stream is Tower deliberately aiming light.
	520	This is D-009 reaching its end point: the swarm was already Tower's body language,
	521	and is now its lungs. There is no autonomic layer beneath. **If Tower's attention
	522	lapses, the rain stops.**
	523
	524	---
	525
	526	## D-018 — Rivers are a closed, pumped circulation, and Tower can easily afford it
	527
	528	Decided. Tower's rivers follow the architecture the author proposed: reservoir →
	529	shaped descent → lake → lift back to the reservoir. Source and sink are one system.
	530
	531	Why it is the only option. The floor is an equipotential (D-015) so there is no
	532	natural slope, and the sealed hull (D-017) leaves no natural water cycle. Water in
	533	Tower moves only because something moves it.
	534
	535	Why it is not a compromise. Costed against the 54.3 TW budget:
	536
	537	\| \| \| \|
	538	\|---\|---\|---\|
	539	\| Earth-like rainfall, 1,000 mm/yr \| 12.2 TW \| 22.4 % — Earth spends the same share \|
	540	\| Recirculating all runoff (1,792 m³/s, a Rhine) against 500 m \| 8.8 GW \| 0.016 % \|
	541
	542	Pumping the entire hydrology of the world is a rounding error — 6,000× headroom.
	543	And full Earth-intensity weather is affordable at the same fraction of budget Earth
	544	itself spends. Tower does not have to choose.
	545
	546	Canon uses both, because they give different rivers. A pumped river *begins at a
	547	machine* — walk to its source and find a wall of water leaving a structure. A
	548	rain-fed river begins at a cloud Tower made over the highlands, and looks entirely
	549	natural while being entirely deliberate. Made weather runs the world's general
	550	hydrology; pumps run the watercourses Tower wants under precise control.
	551
	552	Two consequences worth keeping. A closed loop has no net descent, so a river
	553	circling the world must be pumped — which is the same architecture, closed the whole
	554	way round. And since nothing evaporates to space, **Tower's water inventory is
	555	conserved and exactly known, forever.**
	556
	557	---
	558
	559	## D-019 — Geography is authored as data; terrain is generated from it
	560
	561	Decided. `content/geography.json` holds the authored geography. `npm run terrain`
	562	(tools/terrain.mjs) reads it and emits a height map, a colour map, a corridor detail
	563	render, and a seed `world.json` of places.
	564
	565	Why data first. The author asked to work the geography out before returning to
	566	app code. A JSON source plus a generator means the map is re-derivable from one
	567	readable file: move the river by editing eighteen waypoints, not by repainting.
	568	Coordinates are canon (θ, z) throughout, consistent with D-007 and D-011.
	569
	570	No dependencies. Node's `zlib` is enough to write PNGs directly (tools/png.mjs,
	571	~60 lines for 16-bit greyscale and 8-bit RGB), so the tool adds nothing to
	572	`package.json`. Terrain noise is a small value-noise/fBm module.
	573
	574	Three bugs the renders caught that no test would have.
	575
	576	- A float hash. The noise hash multiplied large integers in floating point,
	577	silently losing precision and streaking the whole map with axis-aligned lines.
	578	Fixed with `Math.imul` throughout.
	579	- A fractional wrap period. The circumference is not a round multiple of any
	580	useful feature size, so the noise lattice did not close on itself and tore at the
	581	seam. Sample coordinates on the θ axis are now snapped to a whole number of
	582	cells; the resulting few-percent anisotropy is invisible.
	583	- A discontinuous river. Valley geometry switched between reaches abruptly,
	584	stepping the valley from 5.4 km wide to 18.2 km in one pixel column and drawing a
	585	visible wall across the map. Reach parameters now blend along the river. A
	586	`--probe` flag prints elevation jumps along a row, which is how it was found.
	587
	588	Known gap, deliberately not fixed here. The seed `world.json` contains
	589	`polyline` geometry (the river, the canyon) and kinds the app does not yet know.
	590	`parseWorld` accepts only points and polygons, so those two places would be dropped
	591	silently on load. App work is paused by request; this is the first thing to fix when
	592	it resumes. Closed by D-022.
	593
	594	---
	595
	596	## D-020 — The world is belts, and the desert rings it
	597
	598	Decided by the author. The desert and its three ridges ring the world; the
	599	capital's water becomes a sea running along the axis; both end caps get ring oceans;
	600	the range feeds four parallel rivers.
	601
	602	The ring desert is a barrier no Earth desert can be. 22,000 km², 13% of the
	603	surface — modest against Earth's ~33% — but on a cylinder there is no way around a
	604	band. Combined with three ringing ridges it is three concentric walls across Tower's
	605	waist.
	606
	607	The drainage was the hard part, and Earth answered it. Many rivers each cutting
	608	their own canyon would destroy the chokepoint. But on Earth most drainages do not
	609	cut through a rising ridge — they are defeated and diverted, and only an antecedent
	610	river keeps its course. So three of the four rivers are turned by First Ridge, run
	611	along its front gathering each other, and leave with the fourth through the one gap.
	612	The Narrows carries the entire world's runoff, ~392 m³/s. The chokepoint is
	613	strengthened by the extra rivers rather than diluted, and the hydrology is more
	614	realistic than a single river would have been.
	615
	616	The Deep's orientation is the decision, not its size. Long down the axis and
	617	narrow around, so it is crossed the short way — around θ, where the floor curves
	618	up and the far shore stands 2.6 km above eye level. Lengthwise it would be dead
	619	flat. It also gives Tower back a thermal flywheel: 19× the diurnal heat storage of
	620	bare ground, 5.9 days per kelvin at full power. D-017's "if Tower blinks the rain
	621	stops" becomes "the world has about a week of momentum" — a lapse now has time to
	622	become visible before it becomes fatal.
	623
	624	The cap seas were not chosen. Gravity on an end cap is tangent to its surface,
	625	so water cannot rest there and runs to the rim. Both ring oceans are consequences,
	626	not designs.
	627
	628	Skiing came out of §1.9.2 unprompted. A descending skier has radial velocity and
	629	therefore real Coriolis: a steady antispinward push on axial fall lines, and a
	630	weight change of up to 11% on circumferential ones. A retrograde run gets lighter as
	631	it accelerates; a prograde run grips harder. Every slope in Tower has a handedness,
	632	which is a sport that cannot exist on Earth.
	633
	634	Tool work this required: `ring` footprints, a `flood` op for troughs and bands,
	635	and a `tarn` op that fills from the bowl floor rather than the rim — the first
	636	version took its level from the rim mean, so water spilled downhill and every
	637	mountain lake rendered as a half-moon.
	638
	639	---
	640
	641	## D-021 — The terrain tool reports progress, and got 13× faster doing it
	642
	643	Prompted by the author: `npm run terrain` ran for a long time saying nothing.
	644	Adding progress output meant first working out where the time went, which turned a
	645	logging task into a performance one.
	646
	647	86 s → 6.4 s. Three causes, all found by measuring rather than guessing:
	648
	649	- No spatial index on the river carve. `opCarve` tested all 592 smoothed river
	650	segments against all 2.4 million pixels — ~2.9 billion distance tests across both
	651	renders. Segments are now binned into a 4 km grid and each pixel tests only its
	652	own cell. 86 s → 17.8 s.
	653	- Deflate level 9 on the colour maps. Hillshade noise compresses poorly, so
	654	level 9 spent 4.2 s on one map for a few percent of size. RGB now encodes at
	655	level 6: 8× faster for 10% larger files, which is the right trade for a tool run
	656	on every edit. The 16-bit height map stays at level 9 — smooth terrain under the
	657	Up filter is nearly all zero deltas, and it costs 0.1 s. 17.8 s → 8 s.
	658	- Domain warps computed before range checks. Several ops evaluated two noise
	659	functions per pixel then discovered the pixel was nowhere near the footprint.
	660	A conservative unwarped-z rejection now runs first. 8 s → 6.4 s.
	661
	662	The log is a table, not a stream. One row per relief op with its timing, then
	663	relief/colourise totals per view, then each PNG with its size, then the seed. The
	664	point is to see which feature is expensive when the file grows, not just that
	665	something is happening. `--quiet` collapses per-feature rows.
	666
	667	Worth recording as a habit: the request was for logging, and the logging is what
	668	exposed that the colour PNG cost more than the entire mountain range. Instrumenting
	669	first and optimising second found three real problems in one pass.
	670
	671	---
	672
	673	## D-022 — The app catches up with the decisions
	674
	675	Implemented. Five outstanding decisions, plus two bugs found while doing it.
	676
	677	The schema now matches the world (closes D-019's gap). `polyline` geometry
	678	joins point and polygon — a polygon needs three vertices for an interior, a
	679	polyline only needs two to go somewhere — and the kind list grows from six to
	680	thirteen (`mountain`, `river`, `canyon`, `forest`, `desert`, `tower`, `town`), with
	681	`AREA_KINDS` / `LINE_KINDS` deciding whether a traced shape closes. Before this the
	682	seed silently dropped The Long Water and The Narrows. A test now loads
`62458672` 2026-09-07 Pavel Safronov	683	`static/world.json` and asserts nothing is lost, so the two files cannot drift apart
`866c7f96` 2026-09-07 Pavel Safronov	684	again unnoticed.
	685
	686	The rendered terrain is drawn under the map, which is what the whole tool was
	687	for. Area regions become outlines rather than fills when it is showing: the raster
	688	already says where the desert is, and a translucent rectangle over it only fights.
	689
	690	A desync that would have been baffling. The app cut the map at θ = 0 with no
	691	slats; `content/geography.json` renders against θ = π. The raster would have landed
	692	half a world out of register with its own place markers. Both now use
	693	`MAP_CUT_WITHOUT_SLATS = π` — opposite the capital, which is D-013's antimeridian
	694	principle applied to habitation rather than geology — and a test asserts the two
	695	files agree.
	696
	697	Reader and writer modes (D-011). `?writer=1` reveals the tools, editor, canon
	698	controls and import/export; readers get the map, the place list, a read-only detail
	699	card and the sightline probe. A bug worth recording: `renderEditor` un-hid the
	700	editor whenever a place was selected, overriding the mode — so readers were shown
	701	the edit fields including private notes. Mode is now decided inside
	702	`renderEditor`, not applied around it. Hiding a panel elsewhere is not the same as
	703	never rendering it.
	704
	705	Permalinks (D-010). `#place=<id>` selects and centres a place; `#at=θ,z` restores
	706	a sightline endpoint. This is why D-012 made ids follow names only until pinned —
	707	a permalink is worthless if the id churns.
	708
	709	Labels are placed by priority and skipped on collision. Settlements outrank
	710	regions; overlapping labels are dropped rather than drawn on top of each other. With
	711	22 seeded places the mountain stations were an unreadable smear.
`5f38efd0` 2026-09-07 Pavel Safronov	712
	713	---
	714
	715	## D-023 — The exterior view, and what it is honest about
	716
	717	Implemented. Phase 3: Tower from outside, with the hull, both caps, the
	718	collector, the converging beam, the spin, and picking on the hull.
	719
	720	Canon numbers are the code's numbers. Scene units are kilometres and the axis
	721	lies along Z with the sun cap at z = 0, so a canon point is literally
	722	`(R cos θ, R sin θ, z)`. No conversion layer to get wrong.
	723
	724	Picking has to undo the spin. The hull turns, so a ray hit is in world space
	725	while the answer wanted is canon θ, which is fixed to the hull. Resolving the hit
	726	through `worldToLocal` handles it. Verified by clicking the same screen pixel at two
	727	spin phases 34° apart and confirming the reported θ moved by 0.588 rad against a
	728	0.593 rad phase change, with z identical — the pixel really is landing on different
	729	ground as Tower turns beneath it.
	730
`11a13dbf` 2026-09-07 Pavel Safronov	731	Three departures from physics, all stated in the code. Sunlight runs parallel to
`5f38efd0` 2026-09-07 Pavel Safronov	732	the axis, so the flank is edge-on and correctly renders almost black; fill lights
	733	make it readable. And in vacuum the beam between collector and cap has nothing to
	734	scatter off, so it would be invisible; it is drawn faintly because it is the point
	735	of the view. Same licence as the map's hillshade, and worth naming rather than
	736	letting a reader mistake either for a physical claim.
	737
	738	Three.js loads on first use of the view. It is by far the largest dependency and
	739	starts an animation loop; a reader who only opens the map should not pay for it.
	740
	741	Earth is not drawn. It is ~90° off the axis and hundreds of thousands of km
	742	away; at true scale it is invisible, and at any visible scale it is a lie about
	743	distance. Given how much of this world came out of taking distances seriously, a
	744	compressed Earth would be the wrong kind of shortcut. The sunward glow is honest —
	745	the sun really is that way — and D-017's point stands: from inside, Earth is never
	746	visible at all.
`11a13dbf` 2026-09-07 Pavel Safronov	747
	748	---
	749
	750	## D-024 — The terrain is wrapped onto the hull as an x-ray
	751
	752	Implemented at the author's request. The exterior view paints the rendered inner
	753	surface onto the outside of the cylinder, alongside the place markers.
	754
	755	It is a projection, and the code says so. From outside, Tower is structure; the
	756	texture shows what is inside. That makes it the frankest of the view's three
	757	reading aids — nobody could mistake a forest on the outer hull for a physical claim,
	758	where the fill lighting and the visible beam are subtler about it. It earns its
	759	place because the hull is opaque: this is the only view that shows the whole world
	760	at once.
	761
	762	The UVs are rewritten, not derived. Three's cylinder runs u around the
	763	circumference and v along the axis; the raster is the other way round, x = z and
	764	y = theta. The obvious approach — recovering the angle from each vertex position —
	765	is wrong, because the seam vertices are duplicated and `atan2` gives both the same
	766	angle, smearing one column of triangles across the entire image. Remapping the
	767	original uv attribute stays linear and lets RepeatWrapping absorb the offset.
	768
	769	Alignment was verified, not eyeballed. The raster was sampled at each place's
	770	computed position and the colour classified: four of four mountains landed on rock,
	771	both cap seas on water, the desert on desert. Judging this by eye would have been
	772	hopeless — the whole map is bands of green and tan, and a half-world error looks
	773	much like a correct one from most angles.
	774
	775	Two fixes the terrain forced. Small bright dots read fine against a bare hull
	776	and vanish into forest and desert, so markers gained a dark outline shell; and they
	777	are now scaled per frame to a roughly constant screen size, because 4 km is
	778	sub-pixel at the 1,200 km the default view sits at. The terrain also needed an
	779	emissive map: sunlight runs along the axis, so most of the hull faces away from
	780	every light and the map would otherwise be legible only on a sliver.
	781
	782	Also fixed here: selecting a place from the list moved the panel but not the
	783	camera — only permalinks called `focusOn`. Both paths now do.
`3b520afd` 2026-09-07 Pavel Safronov	784
	785	---
	786
	787	## D-025 — An empty draft is not a draft
	788
	789	Fixed after the author hit it. The map showed one place where twenty-two were
	790	expected. The local draft was empty and, because any draft beat the seed, the
	791	published world had been shadowed since the first time the app was opened.
	792
	793	How it happened. `public/world.json` originally shipped as `{"places": []}`.
	794	Opening the app once in that window persisted an empty world to `localStorage` —
	795	`pagehide` flushes on close — and from then on `load()` saw a draft, took it, and
	796	never fetched the seed again. Twenty-two places were sitting in the served file, and
	797	nothing in the UI suggested anything was wrong: an empty world and a deliberately
	798	emptied world look identical.
	799
	800	The fix. A stored world with no places is treated as no draft, so the seed
	801	loads. There is nothing to protect in an empty world, and the opposite failure — a
	802	deliberately emptied world refilling on reload — is visible and one click to undo,
	803	where this one was silent and permanent.
	804
	805	Plus an explicit way back. Reload published world in the Data section discards
	806	the draft and refetches `world.json`, for the case where a draft has drifted rather
	807	than emptied. Verified end to end: an empty draft now recovers 22 places on reload,
	808	and reseeding after a local edit returns 23 to 22.
	809
	810	The general shape is worth remembering. Local-first storage that silently wins
	811	over shipped content will, sooner or later, win with nothing in it. Precedence rules
	812	between a cache and its source need a defined answer for the empty case, and a
	813	manual override for every other case.
`04d6779e` 2026-09-07 Pavel Safronov	814
	815	---
	816
	817	## D-026 — Two grids that did not agree, both removed
	818
	819	Removed at the author's request. The exterior had structural ribs every 20 km;
	820	the map had a graticule every 50 km. Neither was wrong on its own, but side by side
	821	they invited a comparison that does not hold — twenty rings against ten lines,
	822	counting nothing in common.
	823
	824	Both are gone. Neither was load-bearing: the map keeps its scale bar and the
	825	exterior its spin readout, and with the terrain drawn the ground itself carries the
	826	sense of distance far better than either grid did. A reference grid that has to be
	827	mentally reconciled with another reference grid is worse than none.
	828
	829	Worth generalising: two independent scales shown in two views of the same object
	830	must agree or one must go. It is not enough for each to be individually correct.
	831
	832	---
	833
	834	## D-027 — A band is its own geometry, not a polygon
	835
	836	Fixed. Regions that ring the world — the desert, the forest, both cap seas — are
	837	now `{ type: 'band', zFrom, zTo }` rather than four-cornered polygons.
	838
	839	Found while removing the graticule. With the grid lines gone, a stray hairline
	840	was left across the middle of the map. A band written as a polygon runs theta from
	841	0 to 2pi, and both of those normalise to the same map row — so all four corners
	842	land on one line and the region collapses into a one-pixel streak. It had been
	843	hiding among the graticule lines the whole time.
	844
	845	Why a new type rather than a workaround. On a cylinder, "the whole circumference
	846	between two z" is a genuinely different shape from a polygon: it has no theta
	847	extent to give, because it has all of it. Tower's geography is built out of these —
	848	D-020 made the desert ring the world precisely because a band cannot be walked
	849	around — so it deserves to be a first-class shape rather than a polygon that happens
	850	to wrap. The map draws it as a full-height rectangle, hit-testing is a z range, and
	851	the editor reports it as rings the world, z 148.0 km to 218.0 km (70.0 km deep).
	852
	853	Tests assert the four ringing regions survive as bands through parse and public
	854	export, and that a band with unusable bounds is rejected like any other geometry.
`8b6287ba` 2026-09-07 Pavel Safronov	855
	856	---
	857
	858	## D-028 — The beam was inverted, and cones fringe
	859
	860	Found because the author asked what the brown cone and the white circle were.
	861	That question is its own answer: the two most important objects in the exterior view
	862	were unlabelled and one of them was drawn backwards.
	863
	864	The beam pointed the wrong way. `CylinderGeometry(radiusTop, radiusBottom)`
	865	combined with `rotateX(-pi/2)` — which sends +Y to −Z — put the first radius at
	866	the collector. Writing the pair in the order the eye reads them gave a cone wide at
	867	the sun cap and narrow at the mirror: light converging on the collector instead
	868	of into the aperture, with a 122 km mouth swallowing a 50 km hull. It had been that
	869	way since the view was built, and the flat brown wash it rendered as hid it.
	870
	871	The brown was a second bug. A uniform additive layer at 0.085 opacity over a
	872	black sky sums to about RGB(43,39,30) — a dull brown that reads as a solid object,
	873	which is the opposite of a beam.
	874
	875	Cones fringe, so the beam is rays now. Brightening it exposed bright dashes
	876	scattered across the collector's rim. Isolated by bisection: removing the cone
	877	cleared the disc, removing the stars did not, and shortening the cone moved the
	878	dashes with its mouth. At the mouth the cone's shell is edge-on, so its triangles
	879	project to slivers and the additive blend stacks them. Rays from the mirror to the
	880	aperture have no silhouette to alias, need no alpha ramp to fake convergence, and
	881	read more like light than a translucent solid ever did.
	882
	883	Everything is captioned now. Collector, beam, sun cap, far cap, as screen-space
	884	labels. If the person who commissioned the world cannot identify an object in it, a
	885	reader has no chance.
	886
	887	---
	888
	889	## D-029 — The view persists, which surfaced a boot-order bug
	890
	891	Implemented at the author's request. Map or exterior is remembered in
	892	`localStorage` and carried in the permalink hash, so a shared link can say *look at
	893	this from outside*. The hash wins; storage is the fallback for a bare URL.
	894
	895	And it immediately broke the exterior on load. Restoring that view un-hides the
	896	canvas and constructs the renderer in the same task, before the browser has laid the
	897	element out — so `resize()` measured a zero-sized box, bailed out, and the renderer
	898	kept its default 300×150 buffer stretched across a full-size canvas. Clicking into
	899	the view had always worked because layout had long since settled by then; only
	900	restoring it on boot hit the window.
	901
	902	Fixed with a `ResizeObserver` on the canvas rather than another guess about timing,
	903	which also covers the sidebar and window resizing for free.
`15d933c9` 2026-09-07 Pavel Safronov	904
	905	---
	906
	907	## D-030 — The camera persists, and the caps are places
	908
	909	Both at the author's request.
	910
	911	Camera. Position and target are stored per browser and restored on load, so a
	912	reload returns to the vantage you left rather than the default framing. Every field
	913	is validated before use — three finite numbers each, and a distance inside the
	914	controls' own limits — because a corrupt entry here would leave the view pointing at
	915	nothing with no obvious way back, and silently-bad persisted state has already cost
	916	this project once (D-025). Saves are debounced: OrbitControls fires `change` every
	917	frame of a drag.
	918
	919	Caps. The sun cap and far cap are now features in `content/geography.json`, so
	920	they appear in the place list and can be focused like anywhere else. Two things had
	921	to be special-cased, both for the same underlying reason — *a cap is not a patch of
	922	ground*:
	923
	924	- Its marker belongs on the axis. Placed on the hull at theta = 0 like every
	925	other marker, it sat on the cap's rim and rode round the edge as Tower turned.
	926	A cap's location is its centre.
	927	- Its framing is different. The usual from-the-side shot shows an end cap edge-on
	928	as a line. Standing off along the axis instead is worse — it looks down the barrel
	929	and foreshortens all 540 km behind an unlit disc. What works is the default view's
	930	oblique angle slid onto the cap, and a light from deepward so the far cap's outer
	931	face is not a black hole exactly when you focus on it.
	932
	933	A note on the debugging. Verifying camera persistence took several wrong turns
	934	because the test kept invalidating itself: `location.reload()` fires `pagehide`,
	935	which saved the live camera over the value the test had just planted. The
	936	persistence was working the whole time. Worth remembering that a `pagehide` save
	937	makes "set state, reload, check state" an invalid experiment.
`0e0c2dba` 2026-09-07 Pavel Safronov	938
	939	---
	940
	941	## D-031 — Sidebar: an accordion, and a label that stopped lying
	942
	943	**Five UI corrections from the author, all of them about the panel saying the wrong
	944	thing or being in the wrong place.**
	945
	946	"Canon" was a section heading for one slider. It read Canon → Star slats — 0,
	947	which framed an exploration control as if the whole section were canon. The section
	948	is now Star slats, its value is a Count, and canon is a chip on the
	949	summary that appears only when the count is 0 — because that is the one value canon
	950	actually asserts (D-017). A label should describe what it labels; "canon" is a
	951	property of a value here, not a category of setting.
	952
	953	Order follows scope. View moved to the top: it changes the whole app, not one
	954	panel. About moved to the bottom, where the least-used thing belongs.
	955
`5e405272` 2026-09-07 Pavel Safronov	956	Foldable sections. Seven panels in a fixed column pushed the useful ones
	957	off-screen. They are `<details>` elements now — native disclosure, keyboard and
	958	screen-reader behaviour for free. *This first shipped as an accordion, one open at a
	959	time, with the Selected panel opening itself on selection; both were removed — see
	960	D-032.*
`0e0c2dba` 2026-09-07 Pavel Safronov	961
	962	Mode gating moved from a CSS class to the `hidden` property, since `<details>`
	963	already carries open/closed state and stacking a `.hidden` class on top of it was a
	964	second, conflicting notion of visibility.
	965
	966	And a claim removed. The About text said the canon was "worked out in the open",
	967	with links to design.md and decisions.md. Those documents are the author's working
	968	notes on an unpublished novel, not a public invitation, and the links went to a repo
	969	a reader may not have. Now it says what is true and no more: *Tower is the setting
	970	of a novel. This map is its companion piece.*
`5e405272` 2026-09-07 Pavel Safronov	971
	972	---
	973
	974	## D-032 — No automatic opening or closing
	975
	976	Reverted at the author's request: the accordion was annoying, and nothing should
	977	open or close by itself.
	978
	979	The two behaviours were one mistake. Exclusive opening meant that clicking a
	980	place could collapse the panel about to describe it, so the Selected panel had to
	981	open itself to compensate. Removing the accordion removed the need for the
	982	compensation: sections the author opens stay open, so a selection simply appears in
	983	a panel that is already there. A workaround that only exists to undo another
	984	behaviour is a sign the first behaviour is wrong.
	985
	986	What remains is state, not behaviour. View and World start open, and Selected
	987	starts open but stays hidden until something is selected — so the first selection
	988	shows without anything having moved. After that the panels are wherever the author
	989	put them: closing the editor keeps it closed through every later selection.
	990
	991	The general rule this is worth remembering by: an interface that rearranges
	992	itself in response to a click costs the user their place. Prefer a sensible initial
	993	state and then hold still.
`cf37225c` 2026-09-07 Pavel Safronov	994
	995	---
	996
	997	## D-033 — The slat control is disabled and its UI removed
	998
	999	Removed at the author's request: changing the count did not fully work. It did
	1000	not, and for two independent reasons, both worth writing down because this is
	1001	deferred rather than abandoned.
	1002
	1003	The raster has no slats in it. `terrain-color.png` is generated from
	1004	`content/geography.json` at canon's 0 slats. Raising the count drew slat bands as a
	1005	vector overlay on top of terrain that shows unbroken land underneath them — two
	1006	layers asserting different worlds. The corridor is authored the same way: Snowcrown,
	1007	the rivers and The Narrows all assume no slats, so bands above 0 cut straight
	1008	through them.
	1009
	1010	And the map's cut angle moves with the count. `mapThetaOrigin()` returns
	1011	`slatCentre(n-1)` when slats exist and π when they do not (D-013, D-024). The raster
	1012	is drawn against π, so every non-zero count rotates the terrain out of register with
	1013	its own place markers — by 120° at 3 slats, 157° at 8, up to 137 km of
	1014	displacement. This is the same desync D-024 fixed for the zero case, and the test
	1015	added there only guards the zero case.
	1016
	1017	What was done. The section is gone from the sidebar and `loadSlatCount()` is no
	1018	longer called, so a stored non-zero count from before cannot leak back in. What was
	1019	kept is everything that makes the count correct: `src/canon/slats.ts`, the
	1020	derivation, and the tests covering counts 1 to 16 — including that land stays
	1021	exactly Wisconsin's 140,268 km² across the whole range, and that the map cut lands
	1022	on a slat centre at every count.
	1023
	1024	What re-enabling would need. The terrain tool would have to bake slats into the
	1025	raster from the same count the app uses, which means the count becomes an input to
	1026	`npm run terrain` rather than a runtime toggle — and the geography would need
	1027	re-authoring around wherever the bands fall. That is a content decision as much as a
	1028	code one, which is the real reason it is parked.
`28f9a6bc` 2026-09-07 Pavel Safronov	1029
	1030	---
	1031
	1032	## D-034 — The interior view: real geometry, and haze doing the work
	1033
	1034	Phase 5, and the piece design.md §2.4 said would carry the illusion. The scope held:
	1035	real geometry, simple ground. The shell is the actual cylinder, displaced by the
	1036	height raster and painted with the colour raster; there is no scenery, no buildings
	1037	and no vegetation.
	1038
	1039	Nothing in the scene rotates, and that is the point. The observer, the ground and
	1040	the air all co-rotate, so from inside a sealed hull the spin is genuinely invisible.
	1041	No stars wheel overhead because there are no stars (§1.4). What is left is
	1042	`src/canon/motion.ts`, and it says something the design did not anticipate:
	1043
	1044	\| \| lean, antispinward \|
	1045	\|---\|---\|
	1046	\| rain at terminal velocity \| 1.5° — fixed, whatever the cloud height \|
	1047	\| a 1 km waterfall, at the bottom \| 11.3°, and 133 m of drift \|
	1048
	1049	The waterfall figure is the velocity at the bottom, which is the angle a streak of
	1050	falling water is drawn at. §1.9.2 tabulates 7.6° for the same drop, which is the
	1051	tilt of the path — where it lands against how far it fell. Both are right and they
	1052	differ by exactly half again in tangent (`freeFallPathLeanRad` alongside
	1053	`freeFallLeanRad`, pinned in test/motion.test.ts), and noticing they were not the
	1054	same number is the only reason this is written down rather than silently
	1055	contradicting the canon.
	1056
	1057	Rain barely leans, and it leans by a constant angle because drag holds the fall
	1058	speed constant. Only free fall keeps accelerating, so only free fall keeps leaning.
	1059	Design.md §2.4 offered "slanted rain and a leaning waterfall" as a pair; they are not
	1060	a pair. The waterfall is the compass. The rain is drawn, and the readout prints
	1061	1.5° beside it so the smallness reads as the finding it is rather than as a bug.
	1062
	1063	Haze is integrated per fragment, 24 midpoint samples through the rotational
	1064	density profile, and it reproduces the §1.8 table from the other direction: the
	1065	readout prints 78.1% straight up against 31.3% for the same 100 km along the ground.
	1066	The extinction is carried per channel (Rayleigh goes as λ⁻⁴, so red survives twice
	1067	as far as green and blue half as far), which is what makes distance blue rather than
	1068	merely grey.
	1069
	1070	The airlight is neutral, deliberately. Tinting it sky blue is the obvious move
	1071	and it is wrong: blue already scatters 2.4× harder in the per-channel extinction, so
	1072	a blue source colours the haze twice and every distant thing turns mauve regardless
	1073	of what it is made of — desert included, which is how the bug was spotted. With a
	1074	neutral source the blue falls out of the physics and the far side stays recognisably
	1075	forest and desert at the 78% canon promises.
	1076
	1077	## D-035 — Standing on the ground requires the ground on the CPU
	1078
	1079	The shell is displaced in the vertex shader, which is fine until somebody has to
	1080	stand on it. The first build put the camera 1.7 m above the undisplaced cylinder —
	1081	underground everywhere the land is higher than 1.7 m, which is everywhere. From
	1082	inside the rock the ground is invisible: its underside is a front face and the shell
	1083	is drawn back-side only. The view showed sky below the horizon and nothing else, and
	1084	at 366 m of eye height it looked perfect, which is what made it puzzling.
	1085
	1086	There is no way to ask the GPU where the ground is, so `terrain-height.png` is now
	1087	read back through a canvas and sampled here. Two mismatches had to be handled:
	1088
	1089	- The mesh is coarser than the raster. Cells are 0.7 km, texels are 0.26 km, so
	1090	the surface actually drawn is the raster sampled at the cell corners and stretched
	1091	flat between them. The lookup interpolates the same four corners rather than
	1092	sampling the raster directly, which would put the observer's feet tens of metres
	1093	off in rough country.
	1094	- Underground is much worse than hovering. Half the spread across the cell is
	1095	added as clearance: nil on the plains, and on a ridge it lifts you clear of the
	1096	facet instead of dropping you inside it.
	1097
	1098	The readback is 8-bit — browsers decode 16-bit PNGs down to 8 — so elevations
	1099	quantise to 19.5 m steps. Both the shader and this lookup read the same quantised
	1100	data, so they agree with each other, which is what matters for not being buried.
	1101
	1102	## D-036 — A custom ShaderMaterial writes linear colour into an sRGB canvas
	1103
	1104	Three inserts its colour-space handling into its own materials and never into a
	1105	custom one, and the first build silently got both halves wrong: the colour raster
	1106	was sampled as if it were linear, and linear output was written straight to the
	1107	canvas. Everything read dark and oversaturated, and the far side of the world came
	1108	out navy where it should have been hazy green.
	1109
	1110	The fix is both halves explicitly — decode the raster with a `pow(c, 2.2)` and end
	1111	each fragment shader with `#include <colorspace_fragment>`. The trap on the way:
	1112	adding `#include <colorspace_pars_fragment>` too fails to link, because Three
	1113	already puts that chunk in the prefix it builds for every material in order to
	1114	define `linearToOutputTexel`. The duplicate definitions produced no console error at
	1115	all — only `INVALID_OPERATION: no valid shader program in use` once per frame, and a
	1116	window filled with clear colour.
	1117
	1118	## D-037 — A shared `#view=` link was erased before it was read
	1119
	1120	Found while opening the interior view by URL. `App`'s constructor rendered before it
	1121	decided which view to open, and `render()` rewrites the hash from current state — so
	1122	`#view=interior` was replaced by an empty hash a few lines before `initialView()`
	1123	went looking for it. Every shared link that named a view had been landing on the map
	1124	since D-029, and it went unnoticed because the last-used view is also persisted in
	1125	`localStorage`, which restored the right view for whoever had visited before. The
	1126	view is now decided before the first render.
	1127
	1128	## D-038 — Selecting a place walks you there
	1129
	1130	Turning to face a place was the wrong default, and standing in the view for a minute
	1131	makes it obvious: at ground level the haze horizon closes at 100–150 km (§1.8), so
	1132	"face the Deepward Sea" from the capital points at 270 km of nothing. Selecting a
	1133	place now moves the observer to it, which is what "focus" already meant in the
	1134	other two views — the map centres, the exterior swings round, the interior walks.
	1135
	1136	Places with extent also get a heading: stand at the head of a river and look down
	1137	it, taking the centroid of the traced points as the aim. A point has no direction of
	1138	its own, so arriving is the whole answer and the heading is left alone. The separate
	1139	"Stand at the selected place" button is gone, since selection now does exactly that.
	1140
	1141	## D-039 — The height raster gets a declared working range
	1142
	1143	`terrain-height.png` was normalised 0..5,000 m against a fixed `maxElevationM`, and
	1144	both halves of that were wrong.
	1145
	1146	The floor mattered more than the ceiling. Clamping at zero flattened everything
	1147	below the datum, so every sea and lake floor was a plane — the Deepward Sea read as
	1148	0 m where it is actually −320 m. `content/geography.json` now declares
	1149	`minHeightM` / `maxHeightM` and the raster is stretched across that range.
	1150
	1151	The range is declared, not fitted. Fitting it to the terrain would silently
	1152	change what every grey level means whenever the geography is edited, which is the
	1153	kind of drift D-024 came from. Instead the generator prints the range it found
	1154	against the range declared, warns and clamps if the terrain has outgrown it, and
	1155	test/raster.test.ts asserts the declared range still contains the terrain.
	1156
	1157	The precision win is smaller than it looks, and worth stating plainly: the
	1158	terrain spans −508..4,392 m, so tightening the range to −520..4,400 m moves the step
	1159	from 19.6 m to 19.3 m. The quantisation is not set by the range at all — it is set
	1160	by browsers decoding a 16-bit PNG down to 8 bits, which leaves 256 levels however
	1161	wide the span. The fix for that is a second byte in a second channel, which is a
	1162	change to the encoder and the shader, and is not made here.
	1163
	1164	## D-040 — The axial sightline names the distance it measured
	1165
	1166	The readout compared "100 km up" against "100 km on" — except the along-axis probe
	1167	was clamped to the far cap, so standing 22 km from the end it measured 22 km and
	1168	still called it 100. It reported 77.5% where the honest figure for 100 km is 31%,
	1169	which is exactly backwards from the point the line exists to make. It now measures
	1170	whichever axial direction has the most room and prints the distance it used.
	1171
	1172	## D-041 — The caps get a pattern, and it is polar
	1173
`da6bcb0e` 2026-09-08 Pavel Safronov	1174	*Half superseded by D-045: the far cap is now a window, and this pattern survives
	1175	on it as the window's mullions rather than as a filled checker. Everything below
	1176	still holds — the tessellation, the pitch, the mipmapping and the reason for all
	1177	three are unchanged, and the sun cap is still exactly this.*
	1178
`28f9a6bc` 2026-09-07 Pavel Safronov	1179	The end caps were flat tinted discs. They are the only surfaces in Tower that are
	1180	not landscape — 100 km of structure closing each end — and blank they read as fog
	1181	with an edge. A checker fixes that, and does something more useful besides: it is
	1182	the one thing in the world with a known cell size, so a glance at the far cap
	1183	tells you roughly how far away it is. That is worth having in a world where §1.8
	1184	says distance is otherwise unreadable.
	1185
	1186	Polar, not square. A cap is a disc on a spinning cylinder, so its natural
	1187	coordinates are radius and angle; a square grid meets the rim at every possible
	1188	angle and reads as something laid over the geometry rather than part of it. Cells
	1189	stay roughly square by doubling the sector count each time the radius doubles, which
	1190	leaves a seam at each doubling — which is what a segmented diaphragm would have
	1191	anyway. The alternative is slivers at the hub or slabs at the rim.
	1192
	1193	One texture, both views. Drawn once into a canvas and shared, so the exterior
	1194	and the interior cannot drift apart. Three maps UVs across the bounding box for both
	1195	`CircleGeometry` and `RingGeometry`, so the same image lands correctly on the far
	1196	cap's full disc and on the sun cap's annulus.
	1197
	1198	Two things had to be walked back after looking at it:
	1199
	1200	- It must be mipmapped. A high-contrast geometric pattern is the worst case for
	1201	point sampling. My first pass disabled mipmaps on the theory that resolving the
	1202	checker at distance was the whole point; minified, the far cap moiréd into a dark
	1203	cross-hatch that read as damage. Averaging to grey once the cells fall below a
	1204	pixel is both correct and honest — you cannot resolve them from 540 km either.
	1205	- Six rings, not ten. 5 km cells looked better face-on from inside and collapsed
	1206	from outside, where D-030 deliberately frames the caps near edge-on and
	1207	foreshortens them ten to one. A pattern has to survive its worst viewing angle.
	1208
	1209	The cap framing also moved to ~40° off the axis rather than ~17°, since there is now
	1210	something on the face worth turning to see, and the caps carry an emissive copy of
	1211	the pattern for the same reason the hull carries one of the terrain: sunlight runs
	1212	along the axis, so a cap is edge-on to every real light and renders as a dark disc
	1213	exactly when it is being looked at.
`fc11b064` 2026-09-08 Pavel Safronov	1214
`6364748a` 2026-09-08 Pavel Safronov	1215	## D-042 — Screenshots are taken through the UI, and diffed per shot
	1216
	1217	`npm run screenshots` drives the real application in a headless browser and
	1218	photographs it: nine views across two viewports, plus a contact sheet of all nine.
	1219
	1220	Through the UI, on purpose. Every shot clicks the view buttons, picks places out
	1221	of the list, and drags to orbit and to look up, rather than posing a renderer
	1222	directly. Posing is easier and it defeats the point. A screenshot taken by using the
	1223	application fails when the application fails, so a black canvas or a sidebar that
	1224	has swallowed the window ends up in the picture instead of hiding behind a passing
	1225	unit test. The tests already pin the canon numbers; nothing was checking that the
	1226	thing still draws.
	1227
	1228	The contact sheet is for a human. `overview.png` exists to be looked at once,
	1229	before committing — one image that answers "is it still there?". It is built by
	1230	laying the shots out as an HTML page and screenshotting that, because the browser is
	1231	already running and already knows how to lay images out; `tools/png.mjs` only
	1232	encodes, and this project has no reason to gain a PNG decoder.
	1233
	1234	Diffed per shot, not per sheet. The first version measured divergence on the
	1235	contact sheet, and that was the wrong unit: rewriting two of the nine panels
	1236	outright moved the whole-sheet mean by 0.6%, because a mean over nine panels on a
	1237	dark page dilutes everything. Per shot the same change reads in tens of percent.
	1238
	1239	The 3% threshold is measured, at both ends. Seven of the ten shots come out
	1240	byte-identical run to run — the rain is the only randomised thing in Tower and is
	1241	seeded on the way in, the exterior spin is switched off, and each shot gets a fresh
	1242	browser context so no camera leaks into the next. The two exterior shots are the
	1243	exception, drifting up to 0.98% because OrbitControls damps toward its target
	1244	asymptotically and lands a hair short of the same place each time. Against that,
	1245	hiding both 3D canvases scored 7–31%; a view going entirely black scores 12% to 39%.
	1246	3% is three times the worst noise and less than half the mildest real failure.
	1247
	1248	It is a smoke alarm, not a verdict: when it fires the answer is to look at the
	1249	sheet. And it only catches regressions — a view that never worked scores 0% when
	1250	it breaks further, which is exactly how mobile-exterior behaved during the sabotage
	1251	test that calibrated this, because at the time it was already a black slot.
	1252
	1253	Two browsers, one tour. The contact sheet gets its own browser instance. In the
	1254	one that has just taken the tour — nine WebGL contexts and two Three.js scenes —
	1255	`Page.captureScreenshot` on the sheet fails outright with "Unable to capture
	1256	screenshot", and the sheet is plain HTML and images, so the only thing wrong with it
	1257	is its neighbours.
	1258
	1259	What it found immediately. The mobile row: the sidebar was a fixed 340 px
	1260	against a 390 px viewport, so the world got about fifty pixels and both 3D views
	1261	rendered into a black slot. Fixed in D-043, which the same tool then verified —
	1262	the six desktop shots moved 0.00% and the three mobile ones moved 10–33%, which is
	1263	exactly the shape a layout fix should have.
	1264
`fc11b064` 2026-09-08 Pavel Safronov	1265	## D-043 — On a phone the sidebar is a drawer
	1266
	1267	The layout was one grid, `1fr 340px`, with no media query anywhere. On a 390 px
	1268	phone that is a 340 px panel and fifty pixels of world, and nothing to hide the
	1269	panel with — the application was, on a phone, a control surface for something you
	1270	could not see.
	1271
	1272	A drawer, not a stack. The other obvious move is to put the panel below the
	1273	world in a single column and let the page scroll. That is wrong here: the world is a
	1274	`position: absolute` canvas filling `main`, all three views are drag-to-look, and
	1275	`touch-action: none` means every gesture over the world belongs to the world. A
	1276	scrolling page and a canvas that eats touches are in permanent conflict. A drawer
	1277	keeps the canvas full-bleed and the gesture unambiguous.
	1278
	1279	Below 720 px, where the panel would otherwise be taking more than half the
`6364748a` 2026-09-08 Pavel Safronov	1280	window. Above it nothing changes at all — the desktop shots come out of the
	1281	screenshot tour byte-identical.
`fc11b064` 2026-09-08 Pavel Safronov	1282
	1283	Never the full width. `min(340px, 86vw)` leaves a strip of world showing, which
	1284	is what tells you the panel is over the map rather than a page you navigated to.
	1285	That strip is live canvas, though, and in writer mode a stray tap on it drops a
	1286	marker, so a scrim covers it and closes the drawer instead.
	1287
	1288	Two buttons, not one. The opener is fixed at the top right, opposite the view
	1289	switcher. The closer is in the panel header, in normal flow. A single fixed
	1290	toggle would have to float over the panel's own scrolling content once open, and
	1291	would occlude whatever scrolled under it; this way each button is only ever over the
	1292	thing it belongs to. The spin readout, which also lives at the top right, drops to
	1293	`top: 54px` to clear the opener.
	1294
	1295	Picking a place closes it. In every view selecting a place moves you: the map
	1296	centres, the exterior swings round, the interior walks there (D-038). On a phone the
	1297	list is the drawer, so staying open means watching none of it happen.
	1298
	1299	The closed drawer is `inert`. It is only translated off-screen, so without that
	1300	its fields stay in the tab order — typing into a panel nobody can see, scrolling it
	1301	under the world. It is set from the same media query the CSS uses, and crossing that
	1302	breakpoint (a phone turned on its side) resets to closed, which is also what clears
	1303	`inert` on the way back to the desktop column.
	1304
	1305	One thing came along with it. The standing readout is ~65 monospace characters
	1306	of aligned columns and overran 390 px at 11 px type. It shrinks to 9 px on the same
	1307	breakpoint rather than wrapping — wrapping those columns destroys the thing they are
	1308	for.
`de65ab10` 2026-09-08 Pavel Safronov	1309
	1310	## D-044 — Tower's wall is a kilometre thick
	1311
	1312	Canon had no figure for it. Both 3D views drew the hull, the caps and the collector
	1313	as zero-thickness surfaces, which is invisible until something is seen edge-on and
	1314	then reads as a paper cutout rather than a structure.
	1315
	1316	1 km, floor to outer skin — 2% of the radius. The alternatives were a 200 m
	1317	technical minimum, which is defensible on shielding and services and is sub-pixel at
	1318	every exterior zoom, and 5 km, which is unmistakable and a great deal of mass to
	1319	justify later. A kilometre reads as built without making Tower a pipe.
	1320
	1321	It is only visible at the openings, which is worth stating because it took a
	1322	false start to notice. A closed can has no visible thickness from outside — the
	1323	silhouette of a 50 km cylinder does not change when you give its wall a kilometre.
	1324	Depth shows only where there is a hole: the beam aperture, and now the far cap
	1325	window. So the caps sit at the outer face of the wall with a 1 km collar
	1326	standing between them and the inner surface, and it is the collar you see down.
	1327
	1328	The exterior camera had to be let closer. A kilometre is 1% of the cap's
	1329	diameter, so at the old 120 km minimum distance the collar was two pixels and the
	1330	near plane sat at 8 km — the depth was being drawn correctly and could not be got
	1331	close enough to to see. 60 km and a 2 km near plane make the choice legible; without
	1332	that change the constant would have been a number in a file and nothing else.
`392bb8a7` 2026-09-08 Pavel Safronov	1333
`769726ea` 2026-09-08 Pavel Safronov	1334	## D-045 — The far cap is a window, and the checker becomes its frame
	1335
	1336	§1.4 has said since the first draft that the caps are the only views out of Tower
	1337	and that the sky through them turns rather than sweeps — a pinwheel about the axis
	1338	with one motionless star at the middle, one revolution every 7 min 29 s. It was
	1339	written and never built: §2.4 said flatly "no stars", and D-041 had meanwhile made
	1340	the far cap a checkered structural plate.
	1341
	1342	Glazed whole, not windows at its foot. §1.4 put the windows at the base of a
	1343	50 km cliff. A strip at the bottom of something 540 km away is nearly nothing to
	1344	look at, and the canon fact worth having in the view is the pinwheel, which needs
	1345	sky. So the whole far cap is a window. The sun cap is not — it faces the sun and
	1346	needs glare management rather than a view, and that asymmetry is already canon.
	1347
	1348	The checker survives as the mullions. Deleting it would have retired the only
	1349	object in Tower with a known cell size, which is the only rule anyone has for
	1350	judging distance (D-041, §1.8). Instead the same polar tessellation at the same
	1351	pitch is drawn as the frame around transparent panes: the cells are exactly where
	1352	they were, and you look through them rather than at them. The rule now reads against
	1353	the sky, which is a better place for it than against a plate.
	1354
	1355	The physics fell out of the shader that was already there. A pane is black with
	1356	stars in it, so the existing haze integration does the rest by itself: by day the
	1357	airlight term swamps everything and the cap is a disc of blue, which is what §1.5
	1358	already said would happen to anything 540 km off; at night the airlight collapses
	1359	with the beam and what is left is the sky. Nothing needed special-casing.
	1360
	1361	The stars are not scaled by the exposure, and that is the whole point. Nothing
	1362	Tower does to its own beam changes how bright Vega is. What does change is the eye,
	1363	so a separate gain carries that half: full at night and below, 2% at temperate. The
	1364	stars are there at noon and unseeable, which is a fact about eyes and not glass.
	1365
	1366	Two projections were wrong before the third. Gnomonic (xy/z) about the axis puts
	1367	the pole star at the origin and blows up as the ray goes perpendicular to the axis —
	1368	which is the ordinary grazing view from the foot of the cap, not an edge case.
	1369	Painting the field on the cap plane instead is stable everywhere and has no fixed
	1370	angular scale: from a kilometre away the entire 60° field lands inside one star
	1371	cell and the sky comes out empty, which is exactly what the first screenshot showed.
	1372	Azimuthal-equidistant has neither failure. It is defined for every direction but
	1373	exactly sunward, stars keep the same angular size however far off the window is —
	1374	which is what being at infinity means — and the pinwheel is not animated but falls
	1375	out of the projection, because azimuth plus spin phase is the rotation.
	1376
	1377	The pole star is placed rather than left to the hash. It is the fixed point the
	1378	whole phenomenon is read against, and a sky that happened not to roll one there
	1379	would be missing the only part of it anybody in Tower would have named.
	1380
`da6bcb0e` 2026-09-08 Pavel Safronov	1381	Settled by D-048: the surviving quarter of the beam does not go through the
	1382	window at all. It arrives collimated in a 3.2 km spot — a thousandth of the cap —
	1383	and that spot is a mirror, not glass.
`769726ea` 2026-09-08 Pavel Safronov	1384
`392bb8a7` 2026-09-08 Pavel Safronov	1385	## D-046 — Night is the beam banked, not the swarm aimed away
	1386
	1387	The obvious way to make night in a habitat lit by a steerable mirror swarm is to
	1388	aim the swarm off. It does not work here, and §1.3's own numbers say so.
	1389
	1390	The beam glows by itself: §1.3 gives the glow to the beam and the air rather
	1391	than to the diffusers, and says outright it "would be there if the beam were empty".
	1392	779 W/m²/sr over a strip 3.65° wide running horizon to horizon is about 0.2 sr, and
	1393	with the mean cosine of a strip through the zenith that is **~100 W/m² on the
	1394	floor** — half of temperate. A lit beam is a light source no matter where the
	1395	mirrorlets are pointed, so a dark world requires a dark beam. There is a test for
	1396	this figure in `test/canon.test.ts`, because the conclusion rests entirely on it.
	1397
	1398	So night is the beam banked to a thousandth of full, 0.32 W/m²: a rod you can
	1399	still find overhead and a world you can just make out under it. And **dark is the
	1400	beam shut down** — no airlight, no lit ground, no rod, and the only thing left in
	1401	the world is the sky through the far cap (D-045). The two settings are genuinely
	1402	different states rather than two points on a dimmer, which is what made both worth
	1403	having.
	1404
	1405	The beam's brightness scales with the exposure already, so it banks and goes out on
	1406	its own; neither setting needed the beam shader touched.
`a3f01bd7` 2026-09-08 Pavel Safronov	1407
	1408	## D-047 — The exterior hull reads two ways, and they are different claims
	1409
	1410	The exterior had a "Show terrain" checkbox that wrapped the raster onto the hull.
	1411	It was doing something stronger than a checkbox admits: from outside Tower is
	1412	structure, so terrain on the outside is not what Tower looks like — it is a diagram
	1413	of what is inside, drawn on the outside because the wall is opaque. Useful, and not
	1414	a view. The two readings are different claims about the same object, so the control
	1415	is now a two-way switch that names both:
	1416
	1417	- True — the outer skin. Bare hull, checkered sun cap, glazed far cap.
	1418	- X-ray — the inner surface projected onto the outside.
	1419
	1420	The wall is drawn as two meshes now, not one double-sided one. Its faces show
	1421	different things: the outer skin changes with the mode and the inner surface is
	1422	always the land, and a single `DoubleSide` material has no way to say so. The inner
	1423	mesh is emissive for the same reason the x-ray always was — it is lit by a beam this
	1424	scene does not model, and unlit it reads as a hole rather than a world.
	1425
	1426	This is what made the far cap window worth having from outside. D-045 glazed it
	1427	for the sake of the interior, where the payoff is the pinwheel. From outside the
	1428	payoff is different and better: the panes are clear, so in true mode you look in
	1429	through the glass and the world is really there, mountains and desert and lakes
	1430	receding 540 km down the tube behind a grid of mullions. It is the only line of
	1431	sight into Tower that is not a projection, and it exists only because the wall got a
	1432	window and the hull got two faces in the same week.
	1433
	1434	The dark backing disc the first version of D-045 put behind the glazing is gone. It
	1435	was there to stop the panes showing the inside of the tube — which turns out to be
	1436	exactly the thing worth showing.
`da6bcb0e` 2026-09-08 Pavel Safronov	1437
	1438	## D-048 — The far cap's bullseye is a mirror
	1439
	1440	Decided by the author, from four options: mirror it, let it out, absorb it
	1441	behind fins, or grade the swarm so it never arrives.
	1442
	1443	The shape of the load is the argument. §1.3.1 leaves 24% of the beam
	1444	unscattered after 540 km, and it survives collimated — so this was never 13 TW
	1445	spread over a 7,854 km² cap. It is 1.7 MW/m² in a 7.96 km² spot, about 1,200
	1446	suns, and an absorber there sits at 2,330 K: survivable for tungsten, and for
	1447	no structure. But the spot is a thousandth of the cap's area, which means the
	1448	whole problem lives on a 3.2 km bullseye and 99.9% of the far cap needs nothing
	1449	done to it at all. That is what makes D-045's window and this coexist.
	1450
	1451	A mirror is the cheap thermal answer, not the expensive one. At 99.9% it
	1452	absorbs 1,665 W/m² — about 1.2 Earth solar constants, ordinary sunlight. The thing
	1453	that would have glowed white is merely warm. The absorber-with-fins option needed
	1454	the entire cap radiating at 400 K and 13 TW moved 50 km sideways from the spot to
	1455	get there; it was strictly worse than every alternative.
	1456
	1457	It also recovers the energy, and fixes a number that was already wrong. The
	1458	return trip makes the optical column 1,080 km rather than 540, so utilisation is
	1459	1 - e^(-2τ) = 94% against 76%. That matters more than it looks, because canon
	1460	had a contradiction sitting in it: §1.3 sets full brightness at 54.3 TW ÷ 169,646
	1461	km² = 320 W/m², which assumes the whole beam reaches the inner surface, while
	1462	§1.3.1's optical depth says only 76% does — implying 243 W/m². Both were
	1463	printed, and a test pinned 320. With the cavity closed at both ends (the mirror at
	1464	one, the collector at the other, returning what still escapes) essentially all
	1465	54.3 TW is eventually absorbed, and 320 is right as it stands. No test moved.
	1466
	1467	What it does not do is make weather. Aiming light at a region to warm it stays a
	1468	separate lever, and §1.9.4 already says that is how every climate in Tower is made.
	1469	The mirror is about not wasting the beam; the Dry Reach is about where the heat is
	1470	wanted. Keeping them separate means neither constrains the other.
	1471
	1472	A geometric trap, recorded because I fell into it. The mirror sits on the axis,
	1473	so it looks like it must occult the pinwheel's motionless centre — I wrote that into
	1474	canon before checking. It does not. The pinwheel turns about the direction
	1475	parallel to the axis, and that direction lands on the window at the observer's
	1476	own radius, not at the cap's middle. From the floor the spin axis is horizontal,
	1477	so the centre of Tower's only moving sky is a point on the horizon, dead ahead
	1478	at eye level — while the mirror is ~10° above it from 270 km. The two coincide only
	1479	for an observer standing on the axis, and nobody can stand there.
`e27aaf20` 2026-09-08 Pavel Safronov	1480
	1481	## D-049 — Rain is not drawn any more
	1482
	1483	Rain existed in the interior view for one reason: with the hull sealed it was the
	1484	only moving thing in the sky, and therefore the only evidence Tower turns. §1.9.2
	1485	then undercut it — drag pins a drop's speed, so rain leans a fixed 1.5° whatever
	1486	the cloud, against a waterfall's 11°. The readout printed the angle to stop the
	1487	smallness reading as a bug, which is an admission that the picture was not carrying
	1488	the claim.
	1489
	1490	D-045 gave the view a whole sky going round once every 7 min 29 s. Against that, a
	1491	1.5° lean is not evidence of anything, and 6,000 line segments were being stepped
	1492	every frame to fail to show it.
	1493
	1494	The physics stays. `canon/motion.ts` still computes the lean and
	1495	`test/motion.test.ts` still checks it — it is a real fact about how anything falls
	1496	in here, and the waterfall is still the compass. It is only the rendering that is
	1497	gone, along with the toggle, the seeded raindrop field, and the six Three imports
	1498	that existed for it. The readout line it occupied now prints the pinwheel's period,
	1499	which is the interval you can actually time against the sky.
`9c891d2e` 2026-09-08 Pavel Safronov	1500
	1501	## D-050 — The sun cap is a window too, and Tower never sees the sun
	1502
	1503	§1.4 was wrong about this cap, and the number that shows it was already in canon.
	1504	It demanded "heavy glare management" on the grounds that the sun cap faces the sun.
	1505	It does not face the sun. It faces the collector, which is 244 km across at 320
	1506	km — 41.8° wide against the sun's 0.53°, seventy-nine times its diameter. The
	1507	collector sits between Tower and the sun by construction, so it occults it entirely,
	1508	and the umbra is 120 km across where Tower's 51 km sits in it. There is no direct
	1509	sunlight on the sun cap and there never has been. Nothing to manage.
	1510
	1511	So this cap is glazed on the same terms as the far one, and what fills it is the
	1512	back of the collector: a black disc, because its front is a mirror and its back
	1513	radiates 9.6 TW of waste heat at about 245 K, which is infrared. Around the edge of
	1514	it, stars. It does not turn with them — the collector flies in formation rather than
	1515	spinning — so the sun cap window is a still black disc against a wheeling sky.
	1516
	1517	Nobody in Tower can see the sun. They already could not see Earth (§1.4, the
	1518	terminator orbit puts it out the side, which is solid hull). A habitat in orbit
	1519	around a planet it has never seen, lit by a star it has never seen.
	1520
	1521	The occultation is computed in the shader rather than modelled: the fragment casts
	1522	its own ray on to the collector's plane and asks whether it lands on the disc. That
	1523	is the same trick the far cap uses for its stars, and it means the interior does not
	1524	have to carry a 244 km mesh 320 km outside the world to draw one black circle.
	1525
	1526	The interior's checkered cap material is gone with this. Both caps are windows now,
	1527	and nothing inside was still using it — the checker survives as the exterior's sun
	1528	cap and as both windows' mullions.
`03607270` 2026-09-08 Pavel Safronov	1529
	1530	## D-051 — The range rings the world
	1531
	1532	The mountains were "a 140 km arc, not a ring" — four named domes covering about a
	1533	third of the circumference, with the other two thirds simply unauthored. Asked to
	1534	close it, and closing it is better than the arc was: a ring of peaks is a watershed
	1535	that divides the whole world rather than a wall you can walk around the end of.
	1536
	1537	The ridge op grows optional ridged noise, which is what turns a ridge into a
	1538	range. Without it a ring band is a smooth swell all the way round — right for the
	1539	three low walls across the waist, wrong for peaks.
	1540
	1541	It blends with `max` rather than adding, and that is not a detail. Added, The
	1542	Girdle's 2,900 m would stack onto Snowcrown's 4,300 and push the massif to 6,900 —
	1543	past the raster's 4,400 m ceiling, where it would clip flat. Taking the greater of
	1544	the two leaves the named summits as the high points of their own range, which is
	1545	what they are: Snowcrown is not a separate mountain from The Girdle, it is the tall
	1546	part of it.
	1547
	1548	Peaks reach about 3,900 m, so they cross the 3,000 m snowline and the ring reads as
	1549	snow the whole way round. The rivers still cut it, because carve runs after relief.
`9a621732` 2026-09-08 Pavel Safronov	1550
	1551	## D-052 — A fourth noise octave, and colour that mottles instead of dimming
	1552
	1553	Height. The base had three noise layers — 78 km, 12 km, 3.4 km — and a pixel is
	1554	264 m. Between the finest layer and the raster's ~530 m Nyquist limit there was an
	1555	octave and a half of nothing, so the ground read as smooth swells with no grain on
	1556	them at exactly the scale you look at it from inside. A fourth layer at 1.5 km and
	1557	±26 m fills that band. It goes in the base, deliberately: carve and flood run
	1558	afterwards, so rivers still cut through it and lakes keep flat surfaces rather than
	1559	inheriting a rippled floor.
	1560
	1561	Colour. There was already per-pixel variation, and it was the wrong kind — one
	1562	noise field scaling all three channels together, which dims and brightens without
	1563	ever changing the colour. A second, finer field now mottles: it pushes the
	1564	channels apart, green moving most and blue least, so ground reads as patchy
	1565	vegetation against bare soil rather than as one paint turned up and down. Keeping
	1566	blue nearly fixed is what stops the desert drifting toward the same green as the
	1567	forest.
	1568
	1569	It exposed a live bug, and the fix matters more than the noise does. The extra
	1570	relief pushed the terrain to -521 m against a declared floor of -520 m, so
	1571	every pixel down there was being silently flattened. The generator printed a warning
	1572	and carried on; nothing else noticed. That is D-024's failure mode exactly — a
	1573	clipped raster displaces terrain rather than breaking it, so nothing downstream
	1574	complains.
	1575
	1576	Two changes: the floor moves to -560 m, and `npm run terrain` now **exits
	1577	non-zero** when any pixel is clamped instead of printing a line into a log nobody
	1578	reads. The unit test could never have caught this — it compares the declaration
	1579	against extremes written down when they were last true, not against the terrain that
	1580	just came out — and it now says so where it sits.
`2bb3e0de` 2026-09-08 Pavel Safronov	1581
	1582	## D-053 — The day is a front, not a dimmer
	1583
	1584	The obvious day-night cycle is a global fade between the light settings. It would
	1585	have worked and it would have thrown away the most interesting thing about how this
	1586	world is lit.
	1587
	1588	What the physics allows. D-046 already established that the swarm cannot make
	1589	night, because §1.3 gives the beam's glow to the beam and the air rather than to the
	1590	diffusers — a lit beam lights the world whatever the mirrorlets are doing. So beam
	1591	power has to ramp, and when it does, the sky brightens everywhere at once. There
	1592	is no way to have one end of the axis get its morning before the other.
	1593
	1594	What the swarm can still do is choose where the light lands. The diffusers hand
	1595	beam light to the ground, and they are individually flown with absurd control
	1596	authority (§1.3.1). So the ground can be lit progressively even though the sky
	1597	cannot, and the natural direction is the one the beam already runs: sun cap to far
	1598	cap.
	1599
	1600	A Tower dawn is therefore a fully lit sky over dark ground, with a line of
	1601	daylight travelling toward you at 225 m/s. That is not a compromise between the two
	1602	options, it is what falls out of taking both constraints seriously — and it is a
	1603	better image than either.
	1604
	1605	It gives Tower longitude. §1.9.4 says the world has no latitude: every point is
	1606	equidistant from the light, which is why nothing about it varies. The cycle adds the
	1607	axis back as 40 minutes of time difference between the caps. Dusk runs the same
	1608	direction, so the sun-cap end has the shorter day by a hair.
	1609
	1610	The panel gets a time slider before it gets a clock. Running the cycle is the
	1611	feature, but a 24-hour day at any watchable rate means waiting for dawn, and dawn
	1612	lasts 40 world-minutes out of 1,440. Being able to scrub straight to 06:17 is what
	1613	makes the front usable at all — and the front is a pure function of the phase, so
	1614	scrubbing and ticking are the same operation.
	1615
	1616	The readout names the distance to the line rather than just the hour, because the
	1617	distance is the part you can act on: 27 km sunward is about two minutes' wait.
`49b0d64d` 2026-09-08 Pavel Safronov	1618
	1619	## D-054 — The collector gets a lit rim
	1620
	1621	Drawing the collector as pure black was right about its face and wrong about the
	1622	object. From inside the sun cap window at night you could not see it at all — only
	1623	that a piece of the sky was missing, which reads as a rendering hole rather than as
	1624	a 244 km structure.
	1625
	1626	The physics gives the rim for free. The collector is backlit by definition: its
	1627	mirror face is pointed at the sun and away from Tower. Anything holding 46,938 km²
	1628	flat has structure standing proud of the mirror plane at the edge, and at the edge
	1629	the sun strikes that structure side-on. Add the scatter every real mirror throws at
	1630	grazing incidence and the boundary is a thin sun-coloured ring. That is simply how
	1631	backlit things look, and it is the one part of the collector that can be lit.
	1632
	1633	What it is not is sunlight past the edge, and the distinction matters because the
	1634	first is a property of the object and the second would break canon. Tower sits wholly
	1635	inside the umbra — §1.3.0 — so the sun's disc is occulted from every point in the
	1636	world. Nobody sees a sliver of star; they see lit metal.
	1637
	1638	Rendered as two exponentials: a tight core at the declared 4 km of edge structure,
	1639	and a wide dim skirt for the scatter around it. One alone reads as a drawn outline
	1640	rather than as light. The first attempt was three times too bright and its skirt
	1641	seven times too wide, which washed the whole disc gold instead of tracing it.
	1642
	1643	It does not scale with the beam. The sun does not dim when Tower banks its own
	1644	lamp, so the rim is outside `uExposure`, like the stars. It does fade as the eye
	1645	adapts, but from a much higher floor than a star's — this is sunlit structure, and
	1646	it stays faintly visible at noon where a star cannot.
`1affced2` 2026-09-08 Pavel Safronov	1647
	1648	## D-055 — Cascadia at 1:1, as a ruler
	1649
	1650	Tower's inner surface is 169,646 km², which design.md §1 has always given as
	1651	Wisconsin's total area. That is a fact about a number, and a number is a poor way
	1652	to feel a size. The question this decision answers is the reader's: **how big is
	1653	this actually?**
	1654
	1655	The obvious answer — put Wisconsin inside it — does not work, and finding out why
	1656	is what produced the answer that does. Wisconsin has Tower's area but not its
	1657	shape: it spans about 525 km east–west and 510 km north–south, while Tower unrolls
	1658	to a 540 x 314 km rectangle. At 1:1 it cannot be laid down at all, because 525 km
	1659	of state will not go round a 314 km circumference; it overlaps itself by 67%.
	1660	Scaled to fit undistorted it covers a third of the surface, and squashed to fill it
	1661	it is stretched 1.77:1 and no longer Wisconsin. **The areas match and the shapes do
	1662	not**, which is worth recording as the honest form of the coincidence.
	1663
	1664	Cascadia works where Wisconsin cannot, because it is a corridor and Tower is a
	1665	corridor. The coast runs north–south, the Sound runs north–south, the Cascade crest
	1666	runs north–south. And the width is a second coincidence, better than the first:
	1667	measured in UTM 10N, **the ground from the Pacific surf at Kalaloch east to the
	1668	Columbia River at Entiat is 312.8 km, against a circumference of 314.2** — 1.4 km
	1669	of slack. Western Washington closes the ring on real terrain. Walk east from the
	1670	beach, over the Olympics, across Puget Sound, over the Cascades, down the east
	1671	flank to the Columbia, and you arrive back at the beach.
	1672
	1673	Nothing is invented and nothing is rescaled. The consequence, which is the
	1674	reason this reads as a world rather than as a diagram: the point 100 km straight
	1675	over Seattle is both the surf north of Kalaloch and the Columbia at Entiat — one
	1676	place, reached by walking either way round. Seattle sits 150 km deepward of the sun
	1677	cap, Olympia 88, Bellingham 277, Vancouver 336.
	1678
	1679	Tower has no north, which is what makes this legitimate rather than a cheat.
	1680	The compass is sunward/deepward/spinward/antispinward (content/geography.json), so
	1681	orienting a real region to the cylinder distorts nothing — there is no north to
	1682	distort against.
	1683
	1684	### The seam, and what was tried against it
	1685
	1686	Theta wraps and real geography does not, so column 0 and column 1191 are neighbours
	1687	on the cylinder while being 314 km apart on Earth. The step between them is 852 m
	1688	median, 1,537 m at p90 — small through the settled half (369–844 m from the
	1689	Columbia bar to Bellingham) and largest past 49°N, in fjord country where kilometre
	1690	walls are already normal.
	1691
	1692	Two levers were measured rather than argued about:
	1693
	1694	- Rotating the sampling frame to follow the Cascadia trend. No help at any
	1695	angle from 0° to 20°: the median stays 850–960 m and p90 and max get worse. The
	1696	mismatch is vertical, not alignment.
	1697	- Sliding the cut east. This works — 36 km east more than halves the step
	1698	through the settled half, to 195–378 m. Rejected anyway, because it selects a
	1699	different 314 km of ground: the outer Pacific coast leaves the world, and with it
	1700	the identity above. Overhead from Seattle becomes the Columbia Basin instead of
	1701	surf-and-river at once. The world also goes drier and more monotone.
	1702
	1703	So the seam stays where it is, and is treated as a landform rather than an error: a
	1704	scarp running the length of the world exactly where the wet side meets the dry one,
	1705	which is the truest thing about this coast. Real analogues are not hard to find —
	1706	the Blue Ridge escarpment, the Drakensberg.
	1707
	1708	### What it costs the codebase
	1709
	1710	Almost nothing, deliberately. **The Cascadia rasters borrow canon's grid and height
	1711	window rather than declaring their own** (-560..4400 m, 2048 x 1192). That is not a
	1712	convenience, it is what keeps the mode cheap: src/views/interior.ts bakes the height
	1713	window into its GLSL as literals, so a surface with its own range would need a
	1714	second shader. It costs nothing to share because Cascadia happens to fit — the
	1715	highest land here is Rainier at 4,352 m, under the 4,400 m ceiling. Deep ocean is
	1716	clamped instead, which is invisible under water; carrying the abyssal plain's 2.6 km
	1717	would coarsen every hillside in the world to pay for it. test/raster.test.ts reads
	1718	the constants back out of tools/cascadia.mjs and holds it to them, because this is
	1719	D-024's failure mode exactly: a surface that quietly disagreed would displace its
	1720	ground rather than fail.
	1721
	1722	So a surface is a swap of two PNG filenames (src/views/raster.ts), and the three
	1723	views each gained a `setSurface`.
	1724
	1725	### What it is not
	1726
	1727	This is not canon and contains no Tower geography. It contradicts §1.9.3
	1728	outright — all relief in Tower is engineered, because a bare cylinder is an
	1729	equipotential with no slope anywhere — and the rivers are wrong in a way worth
	1730	naming: they were carved under Earth gravity with no Coriolis, so they do not lean
	1731	antispinward the way §1.9.2 says everything falling in Tower does. Snoqualmie Falls
	1732	is a compass pointing the wrong way.
	1733
	1734	It is therefore labelled in the panel, defaults off, and draws no places. The
	1735	world list still shows them, because that is the author's data and hiding it would
	1736	look like loss, but nothing is drawn on the terrain: a marker reading "The Capital"
	1737	pinned to Puget Sound would claim a correspondence that does not exist. Probes and
	1738	sightlines stay, because those measure the cylinder, and the cylinder is the same
	1739	either way. The choice is not carried in the URL hash either — a permalink names a
	1740	place in Tower, and Cascadia has none.
	1741
	1742	Elevation comes from the AWS open terrain tiles (SRTM/3DEP over land, GEBCO below
	1743	water, Canadian ground north of 49°N in the same grid), cached under `.cache/` and
	1744	re-fetchable; they are not source and are not committed.