@volter/editor-blender 0.1.149 → 0.1.151

package/contributions/blender-icons.source.mjs

@@ -1,11 +1,10 @@
 /**
  * THE SOURCE of `blender.icons.json`, which is PART OURS AND PART BLENDER'S.
- * 197 of the 346 glyphs it emits are machine-TRACED from Blender's own vector
- * sources and are GPL-2.0-or-later artwork (see "GPL PATHS, DELIBERATELY"
- * below, `blender-icon-trace.mjs`, and the provenance file
- * `blender.icons.traced.json`); the other 149 are drawn HERE in Blender's
+ * Traced glyphs come from Blender's own vector sources and are GPL-2.0-or-later
+ * artwork (see `blender-icon-trace.mjs` and the provenance file
+ * `blender.icons.traced.json`); the remaining glyphs are drawn HERE in Blender's
  * IDIOM. That mixture is why `@volter/editor-blender` is
- * `AGPL-3.0-only AND GPL-3.0-or-later` (`packages/blender/LICENSE`).
+ * `AGPL-3.0-only AND GPL-3.0-or-later` (`packages/editor-blender/LICENSE`).
  * The idiom is what a screenshot conveys: a 16-unit grid, monochrome filled
  * silhouettes, ~1.3-unit strokes rendered as filled capsules with round ends,
  * no outline around a fill, optical size ~12 units inside the 16 box.
@@ -14,7 +13,7 @@
  * the JSON. The JSON is the artifact the style bundle imports; THIS is where a
  * glyph is edited, because a path `d` is not a thing a person edits by hand.
  *
- * Keys are Font Awesome icon NAMES (`icon.iconName`) — the editor's one icon
+ * Keys are icon NAMES (`icon.iconName`) — the editor's one icon
  * primitive swaps a glyph in per name and falls through to its own for any
  * name absent here, so this set is partial by design.
  */
@@ -72,27 +71,7 @@ const TAB_ICON_SOURCES = {
   'properties-data-empty': 'empty_data', // no data     → ICON_EMPTY_DATA
 };
 
-/**
- * THE OUTLINER'S MARKS — every icon `tree_element_get_icon`
- * (`outliner_draw.cc:2619-2945`) and `tree_element_get_icon_from_id`
- * (`:2479-2610`) can pick for a row of Blender's View Layer tree, by FAMILY
- * rather than one line per mark (WORK.md §Blender in the tab is Blender,
- * "Inspection parity", I3).
- *
- * A FAMILY AND NOT A LIST, deliberately. The door reports whatever `ICON_*`
- * Blender picked, and the two places that pick are a `switch` over ID codes and
- * `RNA`'s own enum items (`rna_enum_object_type_items`,
- * `rna_enum_object_modifier_type_items`, `rna_enum_constraint_type_items`) —
- * sixty-three modifier marks and twenty-eight constraint marks among them. A
- * table of names here would be a transcription of those enums maintained in a
- * second place, and the first mark it fell behind on would draw a Font Awesome
- * circle in the middle of Blender's tree. The prefix is what the enums
- * themselves are named by, so tracing the prefix traces the enum.
- *
- * Each traced glyph is named `blender-<file>` with dashes — which is exactly
- * what `blender-outliner-authoring.ts`'s `blenderOutlinerKind` spells from the
- * icon name the door sends, so the two meet with no table at all.
- */
+/** Trace Outliner icon families from tree_element_get_icon, tree_element_get_icon_from_id and RNA enum prefixes. blenderOutlinerKind derives the same blender-<file> names without a second per-icon lookup table. */
 const OUTLINER_ICON_FAMILIES = [
   // `icon_from_object_type` and `tree_element_get_icon_from_id`'s data marks.
   'outliner_',
@@ -159,14 +138,9 @@ const OUTLINER_ICON_NAMES = [
 const TRACED = new URL('./blender.icons.traced.json', import.meta.url);
 
 /**
- * `--trace` — RE-DERIVE the traced paths from a Blender checkout. Separate
- * from the ordinary run on purpose: the checkout is a developer's local clone
- * of Blender at the engine's pin (the recipe is in WORK.md §Blender in the
- * tab is Blender, "Inspection parity"), and a generator that silently emits
- * different glyphs depending on whether a sibling directory happens to exist
- * is worse than one that needs a flag. The JSON it writes is checked in and
- * records each glyph's source path and sha256, so a later trace that disagrees
- * is a visible diff rather than a mystery.
+ * `--trace` rebuilds the traced paths from a Blender checkout at the engine's
+ * pin, selected by BLENDER_SRC or the default checkout path below. Ordinary
+ * generation consumes the saved JSON, which records each source path and sha256.
  */
 if (process.argv.includes('--trace')) {
   const checkout = process.env['BLENDER_SRC'] ?? `${homedir()}/volter/blender-src`;
@@ -230,8 +204,7 @@ if (process.argv.includes('--trace')) {
     } catch {
       throw new Error(
         `${relative} is not in the Blender checkout at ${checkout}. Set BLENDER_SRC, or add ` +
-          `the directory to the sparse checkout: \`git sparse-checkout add release/datafiles/icons_svg\` ` +
-          'followed by the LFS-off checkout line in WORK.md.',
+          `the directory to the sparse checkout: \`git sparse-checkout add release/datafiles/icons_svg\`.`,
       );
     }
     if (svg.startsWith('version https://git-lfs')) {
@@ -267,17 +240,7 @@ const f = (n) => {
   return Object.is(r, -0) ? 0 : r;
 };
 
-/**
- * TWO GLOBAL DIALS, set against the running chrome rather than guessed. The
- * editor paints a glyph at the ACTIVE font size — an 11px hierarchy row, a
- * 10px status bar — where a 1.3-unit stroke in a 16-unit box lands at 0.9
- * device pixels and greys out. `S` grows the optical size to ~14.5 units
- * inside the 16 box — Blender's glyphs FILL their box where Font Awesome's
- * carry padding, and measured against the reference frames that ratio (ink
- * over row height) is what makes a row read as Blender's; `W` fattens every
- * stroke so it still covers a pixel in an 11px row. Geometry below is
- * authored on the plain 16 grid and both dials are applied here, once.
- */
+/** S sets optical extent within the 16-unit grid; W gives strokes enough coverage at small chrome sizes. Apply both once to authored geometry. */
 const S = 1.22;
 const W = 1.12;
 const p = (x, y) => `${f(8 + (x - 8) * S)} ${f(8 + (y - 8) * S)}`;
@@ -287,28 +250,10 @@ const rr = (r) => f(r * S);
 /**
  * A straight stroke with round ends: the capsule its outline encloses.
  *
- * THE SWEEP FLAG IS 0, AND IT WAS 1 UNTIL 2026-09-19 — a one-character defect
- * under every open stroke end in this set. With sweep 1 each end arc bulges
- * INWARD instead of outward, so the path is the rectangle MINUS two
- * half-discs: every open end came out with a crescent bitten out of it, and
- * at a polyline's joint the two bites ate a round notch out of the corner.
- * Nothing reported it because at the 14 px chrome size the bite is one device
- * pixel; it was found at 6x in the editor-type well, whose chevron drew as
- * two arrowheads with a hole where they meet (rendered side by side, sweep 1
- * against sweep 0, before the flag was touched).
- *
- * Closed shapes change too, and toward this docblock's own words: `line(…,
- * closed)` and `rframe` corners were each losing a notch to the same bites,
- * half-covered by the neighbouring capsule.
- *
- * THE BLAST RADIUS, MEASURED rather than reasoned (every glyph rasterized at
- * 64 px both ways and the masks differenced): 97 of the 146 paths move. 96 of
- * them ONLY GAIN ink, at an end or a corner — a knockout that is not there
- * can take nothing. The 97th, `layer-group`, gains 199 px and loses one, on an
- * antialiased boundary. Nothing here was tuned against the bitten rendering:
- * the two glyphs this set fits pixel-by-pixel to a reference frame
- * (`properties-modifiers`, and `outliner`/`properties` below) are built from
- * `poly` and `rrect` and do not move at all.
+ * THE SWEEP FLAG IS 0. With sweep 1 each end arc would bulge INWARD instead of
+ * outward, so the path would be the rectangle MINUS two half-discs: every open
+ * end with a crescent bitten out of it, and at a polyline's joint a round notch
+ * out of the corner.
  */
 function cap(x1, y1, x2, y2, width = 1.3) {
   const w = width * W;
@@ -342,20 +287,7 @@ function ring(cx, cy, r, width = 1.3) {
   return dot(cx, cy, r + w / 2) + dot(cx, cy, Math.max(r - w / 2, 0.01), true);
 }
 
-/**
- * A polyline stroked with round ends and round joins.
- *
- * THE JOINS NEED NO EXTRA DISC, and stamping one was actively destructive.
- * Each `cap` ends in a full semicircle of radius w/2 centred on the join, so
- * the union of two capsules already contains the whole join disc — the joins
- * are round for free. The disc this used to add at every interior vertex had
- * the opposite winding to the capsules it landed in, so under
- * `fill-rule: nonzero` it PUNCHED A HOLE: render `line([[3,3],[3,13],[13,13]])`
- * at 24x and the corner is a white ring around a black core, where the same
- * two `cap` calls alone give a solid corner. At 14px that hole is a sub-pixel
- * dark pip on every corner of every stroked glyph in the set — the
- * "greys out at a small size" symptom, dropped here for all 130 at once.
- */
+/** Round-ended capsules already cover the join disc. An extra disc reverses winding under nonzero fill and cuts a hole in the join. */
 function line(pts, w = 1.3, closed = false) {
   let d = '';
   const n = pts.length;
@@ -386,20 +318,7 @@ function poly(pts, hole = false) {
   return `M${wound.map(([x, y]) => p(x, y)).join('L')}Z`;
 }
 
-/**
- * A filled rectangle with equal corner radii.
- *
- * IT WINDS THE WAY `poly` AND `cap` WIND, and that is load-bearing, not a
- * detail: for most of this file's life the rounded branch ran the other way
- * round the corners, so under `fill-rule: nonzero` every `holeBar`/`dot(ccw)`
- * aimed at a rounded rect UNIONED instead of punching, and every ordinary
- * `poly`/`dot` drawn over one punched a hole instead of adding ink. Nothing
- * reported it — a knockout that does nothing just looks like a solid glyph.
- * It silently ate the collection box's pull slot, the wrench's fork and the
- * bulb's screw thread, and it is why the pan hand carried a bite out of its
- * palm where its thumb should be. Measured, not reasoned: render
- * `rrect(r) + holeBar` and read the centre pixel.
- */
+/** A rounded rectangle winds in the same direction as poly and cap so ordinary geometry unions and reversed geometry cuts holes under nonzero fill. */
 function rrect(x, y, w, h, r = 0) {
   const rad = Math.min(r, w / 2, h / 2);
   if (rad <= 0.01)
@@ -417,19 +336,8 @@ function rrect(x, y, w, h, r = 0) {
   );
 }
 
-/**
- * A rectangular outline of thickness `w`, corners rounded by the joins.
- *
- * A corner radius SMALLER than the stroke's own round end is not a corner
- * radius, it is a defect: the bevel becomes a segment shorter than the capsule
- * is wide, its capsule overlaps both neighbours almost entirely, and the
- * overlap cancels under `fill-rule: nonzero` — at 12x, `border-all` (r 0.4,
- * t 1) and `object-group` (r 0.3, t 2.1) both showed a hollow pip at every
- * corner. Below `t/2` the caps already round the corner by more than `r`
- * asked for, so the plain four-point path is both cleaner and closer to the
- * request.
- */
-function rframe(x, y, w, h, t = 1.3, r = 0) {
+/** A rectangular outline uses rounded joins. For r at or below half the stroke width, use a plain four-point path; the caps already provide the requested corner rounding. */
+function rframe(x, y, w, h, t, r = 0) {
   const inset = t / 2;
   if (r <= inset) {
     return line(
@@ -453,22 +361,18 @@ function rframe(x, y, w, h, t = 1.3, r = 0) {
     [x + inset, y + h - inset - r],
     [x + inset, y + inset + r],
   ];
-  return line(r > 0 ? pts : [pts[0], pts[1], pts[3], pts[5]], t, true);
+  return line(pts, t, true);
 }
 
 /**
  * A stroked arc, angles in degrees, 0 = +x, growing clockwise on screen.
  *
- * ONE annulus band, not a chain of capsules. Built as `line()` over sampled
- * points, each segment's capsule is shorter than it is wide, consecutive
- * capsules overlap almost entirely, and the overlaps cancel under
- * `fill-rule: nonzero` — rendered at 24x, `rotate`'s arc and `lock`'s shackle
- * came out as a row of hollow circles rather than a stroke. A band has no
- * overlaps to cancel: out along the outer radius, back along the inner.
+ * One annulus band follows the outer radius and returns along the inner
+ * radius, preserving a continuous fill without overlapping capsule windings.
  */
-function arc(cx, cy, r, a0, a1, w = 1.3, steps = 0) {
+function arc(cx, cy, r, a0, a1, w) {
   const t = (w * W) / 2;
-  const n = steps || Math.max(6, Math.ceil(Math.abs(a1 - a0) / 9));
+  const n = Math.max(6, Math.ceil(Math.abs(a1 - a0) / 9));
   const at = (i, rad) => {
     const a = ((a0 + ((a1 - a0) * i) / n) * Math.PI) / 180;
     return [cx + Math.cos(a) * rad, cy + Math.sin(a) * rad];
@@ -481,7 +385,7 @@ function arc(cx, cy, r, a0, a1, w = 1.3, steps = 0) {
 }
 
 /** A stroked elliptical arc — cylinder lips, cloud shoulders, orbits. */
-function ellipseArc(cx, cy, rx, ry, a0, a1, w = 1.15, rot = 0, closed = false) {
+function ellipseArc(cx, cy, rx, ry, a0, a1, w, rot = 0, closed = false) {
   const t = (rot * Math.PI) / 180;
   const n = Math.max(4, Math.ceil(Math.abs(a1 - a0) / 15));
   const pts = [];
@@ -539,7 +443,7 @@ function holePoly(pts) {
 
 /** Blender's chevron: a thin V, the outliner's disclosure and every menu arrow. */
 function chevron(cx, cy, s, dir) {
-  const v = { right: [1, 0], left: [-1, 0], down: [0, 1], up: [0, -1] }[dir];
+  const v = { right: [1, 0], left: [-1, 0], down: [0, 1] }[dir];
   const perp = [-v[1], v[0]];
   const tip = [cx + v[0] * s * 0.55, cy + v[1] * s * 0.55];
   const back = -s * 0.55;
@@ -553,14 +457,7 @@ function chevron(cx, cy, s, dir) {
   );
 }
 
-/**
- * Blender's isometric cube — a SOLID hexagonal silhouette with its three
- * interior seams knocked out, never a wire box. `outliner.png` at matched
- * scale settles it: every object mark in a row is a filled silhouette with its
- * detail punched through, and the wire version of this glyph collapsed into an
- * undifferentiated grey circle at the 14px row size (its three 1.15-unit
- * spokes merged and the hexagon's corners rounded away).
- */
+/** Blender’s isometric cube is a filled hexagonal silhouette with three interior seams knocked out. */
 function isoCube(cx = 8, cy = 8, s = 5.4, seam = 1.05) {
   const top = [cx, cy - s];
   const tr = [cx + s * 0.87, cy - s * 0.5];
@@ -577,16 +474,10 @@ function isoCube(cx = 8, cy = 8, s = 5.4, seam = 1.05) {
 }
 
 // ------------------------------------------------------------------- glyphs
-// Keyed by Font Awesome icon name. Grouped the way the chrome reads them.
+// Keyed by icon name. Grouped the way the chrome reads them.
 
 const glyphs = {};
-/**
- * `meta` is the glyph's COLOUR channel (`IconSetContribution`): `tone` names
- * a category the palette inks, and `tonedPath` splits the drawing so only
- * that second path takes it. Whole-glyph tones are assigned from one table
- * further down, next to the frames they were measured in; `tonedPath` has to
- * be authored HERE, because it is a division of the geometry.
- */
+/** meta carries category tone and optional tonedPath. A whole-glyph tone belongs to the table below; a tonedPath is part of the geometry. */
 const g = (name, d, meta) => {
   glyphs[name] = { path: d, ...(meta ?? {}) };
 };
@@ -595,18 +486,7 @@ const g = (name, d, meta) => {
 g('chevron-right', chevron(7.4, 8, 3.1, 'right'));
 g('chevron-left', chevron(8.6, 8, 3.1, 'left'));
 g('chevron-down', chevron(8, 7.4, 3.1, 'down'));
-// BLENDER HAS NO SOLID CARET. Both places the editor spends one — a dropdown
-// trigger's disclosure and a tree row's expander — Blender draws an open
-// CHEVRON, measured in two frames: `modeling.png` at its native 2x, the
-// orientation well's mark at x 1304..1316, y 76..83 (13x8 device px = 6.5x4
-// CSS, peak 216); and `outliner.png` (also native 2x), the collapsed Camera
-// row's mark at x 96..105, y 146..163 (10x18 = 5x9 CSS) and the expanded
-// Collection row's at x 52..69, y 110..119 (18x10 = 9x5 CSS), peak 194. Both
-// are a 1 CSS px stroke and both run 1.8:1, which is exactly what `chevron()`
-// draws. The names stay `caret-*` — a set draws the editor's names its own
-// way, and Font Awesome's filled triangle is still what every other skin gets.
-// `s` is 4.4 rather than `chevron-*`'s 3.1 so the mark keeps the EXTENT the
-// filled caret had here (8.8 of 16 units, ~7 CSS px against Blender's 6.5).
+// Caret names draw open chevrons. Their 4.4-unit extent preserves the dropdown caret’s size; chevron-* uses 3.1.
 g('caret-right', chevron(7.4, 8, 4.4, 'right'));
 g('caret-down', chevron(8, 7.4, 4.4, 'down'));
 g(
@@ -686,24 +566,11 @@ g(
     cap(9.3, 7.4, 9.1, 11.5, 1.1),
 );
 g('magnifying-glass', ring(7.1, 7.1, 3.7, 1.3) + cap(9.9, 9.9, 13.2, 13.2, 1.4));
-/**
- * ZOOM is not SEARCH, and Blender draws the two differently — both frames
- * agree: the Search field's magnifier (`outliner.png`, `properties-object.png`)
- * is a thin RING with a handle, while the viewport navigation cluster's zoom
- * (`modeling-object-none.png`) is a SOLID lens disc with the sign knocked
- * through it and a short fat handle. So `magnifying-glass` keeps its ring and
- * the two signed ones become lenses; drawn as rings, their 1.1-unit sign
- * strokes were the faintest ink on the whole navigation cluster.
- */
-// The handle STARTS INSIDE the lens (9.3,9.3 is 3.1 from the centre of a
-// 4.25 disc): begun at 10.2 it cleared the disc's edge by a tenth of a unit
-// and the round cap floated off as a separate lozenge at 16px — measured in
-// round 1's frame, where the navigation cluster showed a disc and a diamond.
+/** Search uses a ring magnifier; viewport zoom uses a solid lens with a cut-out sign. */
+// Start the handle inside the lens so its rounded cap joins the disc.
 const zoomLens = (sign) => dot(7.1, 7.1, 4.25) + sign + cap(9.3, 9.3, 13.4, 13.4, 2);
-// THE PLUS IS ONE HOLE, a cross outline: two crossing hole bars wind the crossing twice, and
-// under nonzero fill that square paints lens-coloured again (the frame showed a pale centre).
-// Its arms are Blender's on `gizmo-select-box.png`'s cluster: 14 device px across a 26 px
-// lens, 2.2 px thick.
+// Draw the plus as one cross-shaped hole; crossing two hole bars fills their shared center under nonzero winding.
+// Its arms span 14 of the 26-device-pixel lens and are 2.2 pixels thick.
 const plusHole = (cx, cy, a, t) =>
   poly(
     [
@@ -730,17 +597,7 @@ g('grip-lines-vertical', cap(6.3, 3, 6.3, 13, 1.2) + cap(9.7, 3, 9.7, 13, 1.2));
 g('copy', rframe(2.4, 2.4, 8.4, 8.4, 1.2, 1) + rframe(5.9, 5.9, 8.4, 8.4, 1.2, 1));
 
 // -- visibility, the outliner's right column ------------------------------
-/**
- * The eye is the KNOCKOUT construction the header describes, and it is what
- * `outliner.png` shows at matched scale: one solid almond, a dark iris punched
- * out of it, a bright pupil inside the iris — white / dark / white, three
- * bands that survive a 14px row. Drawn instead as two thin lid strokes with a
- * dot between them (what this was through round 0), every band is a
- * partial-coverage stroke: measured against its own row, the glyph peaked at
- * p95 145 where the label beside it peaked at 185, while Blender's peaks
- * ABOVE its label (151 vs 129). A silhouette reaches full ink; a hairline
- * cannot.
- */
+/** The eye is a solid almond with a dark iris cut out and a bright pupil inside, preserving three visible bands in small rows. */
 const eyeLid = [
   [1.8, 8],
   [3.6, 5.3],
@@ -767,12 +624,7 @@ g('cubes-stacked', isoCube(8, 4.6, 3.3, 0.8) + isoCube(8, 11.4, 3.3, 0.8));
 g('square', rframe(3, 3, 10, 10, 1.3));
 g('circle', ring(8, 8, 4.4, 1.3));
 g('circle-dot', dot(8, 8, 6) + dot(8, 8, 3.9, true) + dot(8, 8, 2));
-// Blender's collection: a lidded box with a pull slot.
-// Filled body, filled lid band, and the pull slot KNOCKED OUT of the body —
-// exactly what `outliner.png` shows at matched scale, where Blender's
-// Collection is the brightest mark in the row. Drawn as a 1.25-unit frame
-// under a filled lid, it read as a hairline rectangle no heavier than the
-// label beside it.
+// Blender’s collection mark is a filled, lidded box with a pull slot knocked out of its body.
 const box =
   rrect(2.6, 5.4, 10.8, 8, 0.6) + rrect(2, 2.4, 12, 2.6, 0.5) + holeBar(6.2, 8.6, 9.8, 8.6, 1.6);
 g('box-archive', box);
@@ -969,15 +821,7 @@ g(
   'table-cells-large',
   rframe(2, 3, 12, 10, 1.2, 0.5) + cap(2.6, 8, 13.4, 8, 1.15) + cap(8, 3.4, 8, 12.6, 1.15),
 );
-/**
- * A FLAT 3x3 grid, and it STAYS FLAT. Four sites ask for `border-all` and
- * three of them mean a flat grid — the canvas viewport's grid toggle, the
- * 3D viewport overlay's, and the CSS border editor's "all borders" — so the
- * trapezoid Blender draws on the navigation cluster's projection toggle is
- * NOT this glyph's to become. It gets its own name below, which is the same
- * ruling `editorIcons.modeling`'s docblock already records for the mesh
- * operators: a glyph that IS an operation owns its own name.
- */
+/** border-all draws a flat 3×3 grid for grid and border controls. The viewport’s perspective projection toggle has its own glyph name. */
 g(
   'border-all',
   rframe(3.2, 3.2, 9.6, 9.6, 1, 0.4) +
@@ -986,32 +830,7 @@ g(
     cap(3.7, 6.4, 12.3, 6.4, 0.85) +
     cap(3.7, 9.6, 12.3, 9.6, 0.85),
 );
-/**
- * THE PROJECTION TOGGLE at the foot of the navigation cluster — Blender's
- * fourth cell, and it is a PERSPECTIVE grid, not a square one. Read off
- * `modeling-edit-none.png` (device x 2774..2805, y 541..572): a symmetric
- * trapezoid 32 device wide by 26 tall, four horizontal rules whose spacing
- * OPENS downward (6, 8, 10 device) and four verticals converging upward, the
- * outer pair running x 6.8 -> 1.5 on the left and 24.2 -> 29.5 on the right.
- * Ours was a flat square 27x26 — the height was already Blender's and the
- * width was 5 device short, because a square cannot be as wide at its foot as
- * a trapezoid.
- *
- * The prior cut said so itself and stopped there, for the right reason: it
- * shared `border-all` with three flat-grid sites. The name is what was in the
- * way, so the name is what moved (`editorIcons.viewport.projection` is now
- * `viewport-projection`, carrying Font Awesome's `border-all` drawing as its
- * fallback, so a set without this key paints exactly what it painted before).
- * THE PRIOR CUT ALSO MIS-COUNTED IT as "four cells across": it is three, in
- * both axes — four rules bound three columns.
- *
- * Every constant below was seeded off that character map and then settled
- * against the cell's own half-coverage mask, the same instrument the wrench
- * and the hand use. Seeded 150, settled 7 of 374 reference ink px (the old
- * square: 316). No constant moved as much as one device pixel from its
- * reading, and the four rows moved by ONE SHARED half-pixel, which is the
- * cell's own centring rather than a fit.
- */
+/** The viewport projection glyph is a perspective grid with four rules bounding three columns. Horizontal spacing increases downward and vertical lines converge upward. */
 const perspectiveGrid = () => {
   const cx = 8.01; // the trapezoid's axis
   const yTop = 3.71; // its near and far rules
@@ -1037,11 +856,7 @@ const perspectiveGrid = () => {
   );
 };
 g('viewport-projection', perspectiveGrid());
-/*
- * VIEW_ORTHO, the same toggle in an orthographic view: a flat square grid of three by three
- * cells, as wide as the perspective trapezoid (32 device px on `gizmo-select-box.png`'s
- * navigation capsule at 2x) and centred with it, its rules the trapezoid's weight.
- */
+/* VIEW_ORTHO draws a centered 3×3 square grid at the perspective trapezoid’s width and rule weight. */
 const orthographicGrid = () => {
   const lo = 2.61;
   const hi = 13.41;
@@ -1089,9 +904,7 @@ g(
       [8.8, 11.4],
     ]),
 );
-// The Light row in `outliner.png`: the bulb FILLS its 14px slot — a big head,
-// a wide screw base, a foot. At the old 3.5 head it rendered ~7px of ink in a
-// 14px row and read as a dim pebble beside its own label.
+// The Light glyph fills its slot with a bulb head, screw base and foot.
 g(
   'lightbulb',
   dot(8, 5.7, 4.3) +
@@ -1273,59 +1086,13 @@ g(
     [12.6, 8.6],
   ]),
 );
-// Blender's pan hand (`modeling-object-none.png`) is COMPACT: ~13 px of ink in
-// the cluster's 16 px slot. Ours spanned 1.8..14.2, which the scale dial pushes
-// to ~15 px rendered — a hand a full size larger than the zoom beside it.
+// Keep the pan hand compact at about 13 pixels of ink within its 16-pixel slot.
 g(
   'hand-pointer',
-  // Blender's pan hand measures 15x15 in the cluster at matched scale — SQUARE.
-  // Round 2's shrink left ours 18x20: right width, three pixels too tall,
-  // because the middle finger still reached the top of the box.
+  // Keep the pan hand square within the navigation glyph box.
   rrect(6.7, 3.9, 2.8, 5.8, 1.4) + rrect(4.9, 8, 8.2, 5.1, 2) + dot(5.2, 9.3, 1.3),
 );
-/**
- * THE PAN TOOL'S OWN GLYPH. `editorIcons.viewport.pan` is `faHand`, whose
- * iconName is `hand` — a name this set did not carry, so the navigation
- * cluster drew Font Awesome's hand and nothing else on the Model workspace
- * did (checked by resolving every `fa*` identifier in `packages/editor/src`
- * and `packages/mesh/contributions` to its `iconName` and differencing
- * against these keys: `hand` was the only genuine miss; `faExclamationTriangle`
- * and `faTimesCircle` are FA-5 aliases that resolve to `triangle-exclamation`
- * and `circle-xmark`, both present).
- *
- * RE-MEASURED 2026-09-19 against the capsule's SECOND CELL on
- * `modeling-edit-none.png` (device x 2774..2805, y 421..452 — the cluster's
- * cells are 32x32 device and this one's glyph FILLS it, 32x32 of ink). Ours
- * inked 28x27: "15x15 in a 16 px slot" was a reading of an earlier frame and
- * of the wrong pixels; Blender's pan hand is 16.0 x 16.0 CSS, the same box its
- * magnifier fills, which is the whole point — the cluster's glyphs are one
- * size. Three things were wrong besides the size and each is visible at 8x:
- * the fingers were PARALLEL and touching where Blender SPLAYS them with a
- * clear gap (the index leans right going down, the ring leans left, only the
- * middle is vertical); the thumb was a horizontal block at the palm's waist
- * where Blender's points LEFT and DOWN out of the palm's lower third; and the
- * palm was a flat-bottomed box where Blender's bottom is a wide dome. Together
- * they read as a mitten.
- *
- * FITTED THE WAY `modifierWrench` WAS: the structure read off the cell as a
- * character map first (four splayed finger capsules, a domed palm, a thumb),
- * then every constant settled by rendering at the reference's own scale and
- * counting the pixels where the two HALF-COVERAGE masks disagree. Seeded from
- * the map alone it scored 106 of 534 reference ink px against the old glyph's
- * 277; settled, 39, of which seven are the dome's last row. Every finger constant stayed within a device pixel of its
- * own reading. THE PALM'S DID NOT, and that is honest rather than hidden: an
- * axis-aligned rounded rect cannot be Blender's asymmetric dome (the thumb's
- * mass makes its bottom-left tighter than its bottom-right), so the fit moved
- * the palm 2 device px left to trade that error against the thumb's. The one
- * run of reference ink we do not cover is the dome's last row.
- *
- * THE INSTRUMENT, stated because it decides every number here: our render box
- * is the `2xl` rung, 18 CSS = 36 device, while Blender's cell is 32 — so a
- * comparison crops our 36 px raster to its central 32, and a glyph that FILLS
- * Blender's cell occupies only 32/36 of our own viewBox. Reading ours at a
- * loose coverage cut instead of half makes it measure a pixel wider than it
- * is, which is the trap the mesh-data glyph's note already records.
- */
+
 g(
   'hand',
   // Four fingers, each a capsule from its own cap centre down into the palm.
@@ -1337,24 +1104,7 @@ g(
     cap(3.3, 8.99, 5.38, 10.82, 2.0), // the thumb, out of the palm's lower third
 );
 g(
-  /**
-   * The MOVE tool. Blender's (`modeling-edit-none.png`, tool shelf) is four
-   * SOLID arrowheads standing apart on short stubs around an empty centre —
-   * not a crosshair. Ours was one continuous 1.2-unit cross with 1.15-unit
-   * chevrons on its ends, which at matched scale is a thin plus sign with
-   * ticks: a different mark, and the lightest thing on the rail.
-   *
-   * RE-MEASURED 2026-09-19 off the same frame, this time as an ASCII map in
-   * AUTHORED units (0.5/char) rather than by eye: the previous fit was RIGHT
-   * IN FORM AND WRONG IN SIZE AND PARTS. Ours inked 35x34 device px against
-   * Blender's 46x46 — 74% — because the heads stopped at radius 6.4 where
-   * Blender's tips reach the box edge (radius 8, base at 5.5, half-width
-   * 2.4); the STUBS were missing entirely (a 1.0-unit bar from radius 2.5 to
-   * 4.6 under each head); and a 1.35 centre dot was drawn where Blender's
-   * centre is empty. Size is not a detail on this rail — the same frame's
-   * inset and bevel marks ink 40x40 and ours match them to a pixel, so the
-   * transform group was the odd one out in its own column.
-   */
+  /** Move: four solid arrowheads on separate stubs around an empty centre. */
   'up-down-left-right',
   [
     [0, -1],
@@ -1374,20 +1124,7 @@ g(
 );
 g('arrows-up-down-left-right', glyphs['up-down-left-right'].path);
 g(
-  /**
-   * The ROTATE tool. Blender draws a heavy arc with a SOLID arrowhead at each
-   * end and a solid diamond at the pivot; ours was a 1.3-unit open circle with
-   * one small tick, which at matched scale is an empty ring — the pivot, the
-   * direction and the weight all missing.
-   *
-   * RE-MEASURED 2026-09-19 (ASCII map in authored units, and the cell read at
-   * 8x): it is TWO arcs, not one — the top arc ends at the LEFT in a
-   * DOWN-pointing triangle and the bottom arc ends at the RIGHT in an
-   * UP-pointing one, the ordinary two-headed circular arrow, where the
-   * previous fit drew one dome with both heads under its ends and no bottom
-   * arc at all. And it was 72% of Blender's size (33x29 device against
-   * 46x38): the ring's radius is 6.5 of the 16 box, not 4.3.
-   */
+  /** Rotate: two heavy arcs with opposing solid arrowheads and a diamond pivot. */
   'rotate',
   arc(8, 7.8, 7.0, 200, 336, 0.8) +
     arc(8, 7.8, 7.0, 20, 156, 0.8) +
@@ -1497,25 +1234,12 @@ g(
 /**
  * THE OTHER TWO TOOLS OF BLENDER'S TRANSFORM GROUP, under names of ours.
  *
- * Move and Rotate above are shared Font Awesome names (`up-down-left-right`,
- * `rotate`) and their Blender drawings say the same thing everywhere they
- * appear, so those stay keyed to the shared name. Scale and Transform do not:
- * the host's Scale tool asked for `maximize` (Font Awesome's window-expand
- * cross) and its Transform tool for `arrows-to-dot` (a crosshair), and
- * Blender's marks for the two — a small square growing into a bigger one, and
- * a square inside a broken ring of arrows — are meaningless under those
- * generic meanings. Same reasoning as `editorIcons.modeling`'s `mesh-*`:
- * a glyph that IS an operation needs its own name or it recuts every other
- * site that asked for the generic one.
+ * Move and Rotate share Font Awesome names (`up-down-left-right`, `rotate`).
+ * Scale and Transform use operation-specific names so their Blender glyphs
+ * do not replace generic window-expand or crosshair icons elsewhere.
  */
 g(
-  /**
-   * SCALE. Measured on `modeling-edit-none.png` (button 5) as an ASCII map in
-   * authored units: a 1-unit square OUTLINE from (2.0, 1.5) to (14.0, 13.5),
-   * a SOLID square from (1.5, 7.5) to (8.0, 14.0) overlapping its lower-left
-   * corner, and a diagonal arrow inside the frame running up-and-right from
-   * (8.6, 6.9) to a head whose apex is (12.4, 3.1).
-   */
+  /** Scale: an outlined square overlapped at lower left by a solid square, with a diagonal arrow inside. */
   'tool-scale',
   rframe(2, 1.5, 12, 12, 0.8) +
     rrect(1.5, 7.6, 6.3, 6.3, 0.2) +
@@ -1526,13 +1250,7 @@ g(
       [9.3, 4],
     ]),
 );
-/*
- * SELECT BOX. Measured on `gizmo-select-box.png` (the lit first button, native 2x) and mapped onto
- * the 16-unit grid with the marquee spanning 1..15: a dashed square, 0.7 units thick, dashes about
- * 2.3 units with 1.1-unit gaps and an L at each corner, in the selection tools' orange (the
- * `select` tone, `tonedPath`); inside it a pointer from its tip (6.1, 4.5) down to (6.2, 11.5),
- * notched at (8.2, 9.4), to (11.2, 9.4).
- */
+/* Select Box: an orange dashed marquee with corner marks and a pointer, authored on the 16-unit grid. */
 const MARQUEE_T = 0.7;
 const marquee =
   // corners: an L each
@@ -1754,20 +1472,13 @@ g(
     ]),
 );
 g(
-  /**
-   * The Properties TOOL tab. Blender's (`properties-object.png`, top of the
-   * tab rail, 14x14 at matched scale) is two tools standing UPRIGHT side by
-   * side — a screwdriver and an open-end wrench, both solid. Ours was one
-   * diagonal shaft with a ring at one end and a wedge at the other: at 14px it
-   * reads as a single key, and the ring is an outline that greys out.
-   */
+  /** The Tool tab draws an upright screwdriver and open-end wrench, both solid. */
   'screwdriver-wrench',
   // screwdriver: blade, collar, handle
   rrect(3.5, 2.6, 1.4, 4.6, 0.3) +
     rrect(2.9, 7.2, 2.6, 1.2, 0.3) +
     rrect(3.1, 8.4, 2.2, 5, 0.8) +
-    // wrench: a forked head over a shaft. The fork must be WIDE enough to
-    // survive 14px — at 1.7 units it closed up and the head read as a block.
+    // Wrench: a forked head over a shaft, with a 2.2-unit jaw opening for 14px use.
     rrect(9.6, 6.6, 2.8, 6.8, 0.9) +
     rrect(8.4, 2.6, 5.2, 4.6, 0.5) +
     holeBar(11, 2, 11, 4.9, 2.2),
@@ -1799,19 +1510,9 @@ g(
     cap(11.6, 2, 14, 4.4, 1.6),
 );
 g(
-  /**
-   * The MATERIAL section's mark (`inspection/compose.ts` maps /material/ here).
-   * Blender's Material tab is a SOLID disc with quadrants knocked out of it;
-   * ours was a 1.3-unit outline with three dots inside, and at the 14px rail
-   * the outline greyed and the dots merged — it read as a cookie. Same
-   * palette silhouette, drawn Blender's way: one filled blob, wells and thumb
-   * hole punched THROUGH it.
-   */
+  /** The Material section draws a solid palette with wells and a thumb hole cut through it. */
   'palette',
-  // One blob with the wells and the thumb hole punched THROUGH it. Round 3
-  // grew a handle off the bottom to sell the palette's notch and the glyph came
-  // back a mushroom; Blender's Material tab is a compact rounded mass and this
-  // matches that silhouette.
+  // A compact palette silhouette with wells and thumb hole cut through it.
   poly(
     Array.from({ length: 28 }, (_, i) => {
       const a = (i / 28) * Math.PI * 2;
@@ -1824,10 +1525,7 @@ g(
     dot(10.4, 10.7, 2, true),
 );
 g(
-  // A stack of plates: the top one SOLID (the silhouette that carries the
-  // glyph at a 14px row), the two beneath it strokes that read as its shadow.
-  // Drawn as three outlines it was three grey lozenges with nothing solid in
-  // any of them — the same failure the Collection box had.
+  // A stack of plates: a solid top silhouette with two stroked plates beneath.
   'layer-group',
   poly([
     [8, 2],
@@ -1853,10 +1551,7 @@ g(
     ),
 );
 g(
-  // The GROUP row's mark. Beside Blender's Collection box in `outliner.png` at
-  // matched scale, a 1.1-unit frame with four nubs was indistinguishable from
-  // the letterforms of its own label — same weight, same grey. Blender's is a
-  // solid box: a filled band for the frame, and solid corner handles.
+  // The group mark uses a solid box frame and solid corner handles.
   'object-group',
   rframe(2.4, 3.8, 11.2, 8.4, 2.1, 0.3) +
     rrect(0.6, 2, 3.1, 3.1, 0.3) +
@@ -1866,63 +1561,35 @@ g(
 );
 
 // -- the mesh operators: THE OPERATION, ON A CUBE ---------------------------
-/**
- * `editorIcons.modeling.*` asks for `mesh-*` (see that file's header), so these
- * six are free to be what Blender's toolbar draws: not an abstraction of the
- * verb (an up-arrow for Extrude, scissors for Knife) but THE OPERATION PERFORMED
- * ON A CUBE — a cube with one face pulled out, a cube with an inset face, a cube
- * with a chamfered edge, a cube with a cut running round it. Measured in
- * `modeling-edit-none.png` (2x, halved): each glyph inks ~20x20 CSS px in a
- * 38x34 button (Extrude 20x23 — its lifted slab is taller), the cube is a
- * ~2 px wire with the operated element FILLED, and the cube sits in three-quarter
- * view: a front square with a top and a right face receding up-and-right.
- *
- * MONOCHROME, which is this set's contract (`blender.style.ts`) and a real
- * difference from the reference: Blender tints the operated element (green for
- * these six, purple for Smooth) and reads the operation by COLOUR as much as by
- * shape. Here the same job is done by FILL — the operated face is solid against
- * the wire — and whether the set gains an accent channel is its own decision,
- * not one to smuggle in through six glyphs.
- */
-
-/**
- * THE OPERATOR CUBE, once. `P(u, v, k)` is a point in the cube's own axes: `u`
- * across the front face, `v` up it, `k` back along the depth the top and right
- * faces recede along. Every one of the six is authored in those three, so the
- * cube can be re-measured in one place and all six follow.
- *
- * It fills the 16 grid corner to corner (~14 units of extent), because the
- * toolbar's glyph is the biggest in the chrome: that is the ~20 CSS px of ink
- * Blender's 38x34 button carries, once the file's `S` dial is applied.
- */
+/** Shared three-quarter cube geometry for mesh operators. Fill marks the operated element;
+ *  the rest stays wire geometry. tonedPath tints only the operated element. */
 const CUBE = { x0: 1.1, x1: 10.9, yb: 14.9, yt: 5.1, dx: 3.9, dy: -3.9 };
+/** P(u, v, k) maps the operator cube’s front axes and receding depth. All six operators share this geometry and its optical scale. */
 const P = (u, v, k = 0) => [
   CUBE.x0 + (CUBE.x1 - CUBE.x0) * u + CUBE.dx * k,
   CUBE.yb + (CUBE.yt - CUBE.yb) * v + CUBE.dy * k,
 ];
 
 /**
- * The cube as WIRE — the nine edges a cube shows from here, named so an
- * operator can drop the one it replaces (Bevel replaces the top-front edge,
- * Extrude takes the whole top away with the face it pulled out). `top` is the
- * v the box stops at, so Extrude's body can be shorter than the others' and
- * still leave the glyph the same total ink.
+ * The cube's nine visible edges are named so each operator can omit the edges
+ * it replaces. `top` sets the body's height while the operator's added geometry
+ * completes the glyph.
  */
-function cubeWire({ top = 1, w = 1.15, omit = [], u0 = 0, u1 = 1 } = {}) {
+function cubeWire({ top = 1, omit = [] } = {}) {
   const edges = {
-    frontBottom: [P(u0, 0), P(u1, 0)],
-    frontRight: [P(u1, 0), P(u1, top)],
-    frontTop: [P(u0, top), P(u1, top)],
-    frontLeft: [P(u0, 0), P(u0, top)],
-    topLeft: [P(u0, top), P(u0, top, 1)],
-    topBack: [P(u0, top, 1), P(u1, top, 1)],
-    rightBack: [P(u1, top, 1), P(u1, 0, 1)],
-    rightBottom: [P(u1, 0), P(u1, 0, 1)],
-    seam: [P(u1, top), P(u1, top, 1)],
+    frontBottom: [P(0, 0), P(1, 0)],
+    frontRight: [P(1, 0), P(1, top)],
+    frontTop: [P(0, top), P(1, top)],
+    frontLeft: [P(0, 0), P(0, top)],
+    topLeft: [P(0, top), P(0, top, 1)],
+    topBack: [P(0, top, 1), P(1, top, 1)],
+    rightBack: [P(1, top, 1), P(1, 0, 1)],
+    rightBottom: [P(1, 0), P(1, 0, 1)],
+    seam: [P(1, top), P(1, top, 1)],
   };
   return Object.entries(edges)
     .filter(([name]) => !omit.includes(name))
-    .map(([, [a, b]]) => cap(a[0], a[1], b[0], b[1], w))
+    .map(([, [a, b]]) => cap(a[0], a[1], b[0], b[1], 1.15))
     .join('');
 }
 
@@ -1940,20 +1607,12 @@ function slab(vLow, vHigh) {
  *  face floating above it as a solid slab with its own three faces. */
 g(
   'mesh-extrude',
-  // The body is OPEN at the top — it lost that face, and Blender's glyph says
-  // so by letting the verticals run up and vanish behind the slab rather than
-  // closing a rim under it (round 2 closed the rim and the pair read as a
-  // crate with a lid; round 1 stopped the verticals short and read as a
-  // bracket). So: full-height verticals to the slab's underside, no top rim.
+  // Full-height cube verticals end behind the lifted slab; omit the top rim of the extruded body.
   cubeWire({ top: 0.74, omit: ['frontTop', 'topLeft', 'topBack', 'seam'] }),
   { tone: 'operator', tonedPath: slab(0.74, 1) },
 );
 
-/** INSET — the front face INSET: the rim it leaves is solid, the smaller face
- *  inside it is knocked back out, and the four WALLS between them are knocked
- *  apart by their own seams. Blender's inset glyph is that funnel — four
- *  trapezoids converging on a small dark square — and round 4's flat frame
- *  (no seams) read as a picture frame instead of a recess. */
+/** Inset: four solid rim trapezoids separated by seams around the recessed face. */
 g('mesh-inset', cubeWire(), {
   tone: 'operator',
   tonedPath:
@@ -1965,12 +1624,8 @@ g('mesh-inset', cubeWire(), {
     holeBar(...P(0, 1), ...P(0.3, 0.7), 0.7),
 });
 
-/** BEVEL — one edge CHAMFERED. RE-READ at 8x against the reference: the edge
- *  Blender cuts is the TOP-RIGHT one (the seam between the top face and the
- *  right face), and the glyph shows the top face solid with a second facet
- *  flapping off it along that seam, a hairline of background between the two.
- *  Rounds 2-4 chamfered the top-FRONT edge and the result read as a solid top,
- *  which is Extrude's answer, not Bevel's. */
+/** Bevel chamfers the top-right seam between the top and right faces. A solid
+ * top face and chamfer facet meet along a narrow background separation. */
 g('mesh-bevel', cubeWire({ omit: ['seam'] }), {
   tone: 'operator',
   tonedPath:
@@ -1979,10 +1634,7 @@ g('mesh-bevel', cubeWire({ omit: ['seam'] }), {
     holeBar(...P(0.56, 1), ...P(0.56, 1, 1), 0.7),
 });
 
-/** LOOP CUT — the cube CUT IN TWO. Blender's glyph is not a stripe on a cube:
- *  it is two half-boxes standing apart with the cut itself, a solid bar, in the
- *  gap between them, so the new edge loop is the subject. Round 4 drew a line
- *  across an intact cube and read as decoration on the face. */
+/** Loop Cut: two separated half-boxes with a solid cut bar in the gap. */
 g('mesh-loop-cut', cubeWire(), {
   tone: 'operator',
   tonedPath:
@@ -1994,8 +1646,7 @@ g('mesh-loop-cut', cubeWire(), {
  *  turns at a vertex, its ends pinned by the points a knife sets. */
 g('mesh-knife', cubeWire(), {
   tone: 'operator',
-  // A RISING cut with its turn on the front face and its far end up on the
-  // top face. Round 1 fell to the right and read as a droop.
+  // The cut rises across the front and ends on the top face.
   tonedPath:
     line([P(0.16, 0.06), P(0.52, 0.72), P(0.78, 1, 0.5)], 1.5) +
     dot(...P(0.16, 0.06), 1.45) +
@@ -2020,19 +1671,11 @@ g('mesh-subdivide', cubeWire(), {
     cap(...P(1, 0, 0.5), ...P(1, 1, 0.5), 1.05),
 });
 
-/** SMOOTH — not an operation on the operator CUBE: Blender draws this one (and
- *  only this one of the shelf's marks) on a BALL, because smoothing is what
- *  turns the cage into one. Measured on `modeling-edit-none.png` (button 17)
- *  as an ASCII map in authored units: a faceted sphere ~6.3 units of radius
- *  centred at (8.2, 6.6), its quads separated by seams KNOCKED OUT of the
- *  fill — an equator across the full width, two meridians, and a latitude
- *  seam near the top. The seams are holes, not strokes: drawn as strokes over
- *  the ball they would be the same ink as the ball and vanish. */
+/** Smooth draws a faceted ball with seam knockouts for the equator, meridians and upper latitude. */
 g(
   'mesh-smooth',
   poly(
-    // ELEVEN vertices, not a smooth circle: Blender's silhouette is faceted
-    // and that is half of what says "quads" before the seams are read.
+    // Nine vertices give the ball a faceted silhouette.
     Array.from({ length: 9 }, (_, i) => {
       const a = (i / 9) * Math.PI * 2 + 0.34;
       return [7.8 + Math.cos(a) * 7.1, 6.4 + Math.sin(a) * 7.1];
@@ -2040,9 +1683,7 @@ g(
   ) +
     // equator, meridians, and the top latitude — each a bar-shaped hole.
     holeBar(0.8, 7.6, 15, 7.6, 0.62) +
-    // The meridians and the top latitude BOW, two segments each — a sphere's
-    // seams are not straight, and drawn straight the ball read as a beach
-    // ball with a grid printed on it rather than as faceted geometry.
+    // The meridians and top latitude bow through two segments each.
     holeBar(6.9, -0.7, 5.9, 6.4, 0.52) +
     holeBar(5.9, 6.4, 6.7, 13.5, 0.52) +
     holeBar(12.7, 1.1, 12.2, 6.4, 0.48) +
@@ -2052,13 +1693,7 @@ g(
   { tone: 'operator' },
 );
 
-/** SHRINK/FATTEN — the selection moved ALONG ITS NORMALS, in either direction,
- *  which is why Blender's mark (button 19) carries eight arrowheads: four
- *  outside the cube pointing out and four inside pointing in. The cube here is
- *  a pair of offset squares rather than the operator CUBE's three-quarter box
- *  — measured, the front square is (3.5, 3.5)-(12, 11) and the back one is
- *  offset up-and-right by 1.2 — because the arrows need the box read from the
- *  front to be legible through it. */
+/** Shrink/Fatten draws offset front-facing squares with four inward and four outward arrowheads for motion along normals. */
 g(
   'mesh-shrink-fatten',
   // The BOX, front square plus the two BACK edges the front does not hide
@@ -2219,13 +1854,6 @@ g(
   'mobile-screen-button',
   rframe(4.2, 1.6, 7.6, 12.8, 1.2, 1) + dot(8, 12.4, 0.8) + cap(6.4, 3.6, 9.6, 3.6, 1),
 );
-g(
-  'users',
-  dot(5.4, 5, 2.5) +
-    arc(5.4, 12.6, 4.4, 180, 360, 1.25) +
-    arc(11.4, 5.4, 2.1, 250, 470, 1.15) +
-    arc(11.6, 12, 3.6, 200, 340, 1.2),
-);
 g(
   'circle-user',
   dot(8, 8, 6) +
@@ -2394,8 +2022,6 @@ g(
     true,
   ) + cap(8, 2.6, 8, 13.4, 1.15),
 );
-g('boxes-placeholder', '');
-delete glyphs['boxes-placeholder'];
 g(
   'download',
   cap(8, 2.2, 8, 9.8, 1.3) +
@@ -2457,72 +2083,17 @@ g(
 );
 
 // ------------------------------------------------------- category tones
-/**
- * COLOUR AS A GLYPH'S OWN CHANNEL. Blender's Properties-tab rail groups by
- * HUE — in `properties-object.png` at 14 px the tab shapes are near
- * indistinguishable and the ink is what says which group a tab belongs to —
- * and the outliner tints a row's type glyph the same way. Each name below is
- * a category the palette inks (`EditorTheme.color.category`, where every
- * value's source coordinates are recorded); a glyph paints
- * `var(--volter-category-<tone>, currentColor)`, so under a palette that names
- * no category group it paints exactly as it did monochrome.
- *
- * A NAME, NOT A SITE. A tone here tints EVERY site that asks for this icon
- * name, so a name is toned only when its meaning is the category at every
- * one of them. Measured live, that is not a hypothetical: the first cut
- * toned `cube`, `camera`, `lightbulb` and `globe`, and the running editor
- * painted the Perspective/Orthographic dropdown orange and the world-space
- * toggle red — both are viewport CONTROLS that happen to reuse the drawing.
- * The fix is `editorIcons.modeling`'s: a meaning that needs its own ink gets
- * its own name. `hierarchy-kind-icon.ts` now asks for `outliner-object`,
- * `outliner-light` and `outliner-camera`, which no other site asks for, and
- * they are aliased to the same drawings below.
- *
- *  - `outliner-object`, `outliner-light`, `outliner-camera` → `object`.
- *    `outliner.png`: the Camera, Cube and Light rows' type glyphs all
- *    measure #bb7f4d over the row's #272727 — ONE orange for all three,
- *    because Blender's outliner discriminates OBJECT from DATA, not camera
- *    from light. The same ink is the Object tab's in `properties-object.png`.
- *  - `atom` → `modifier`. Its only site is the inspector's physics/rigid-
- *    body/collider group (`inspection/compose.ts`'s `GROUP_ICON_RULES`);
- *    `properties-object.png` paints Physics — with Modifier, Particles and
- *    Constraints — one blue.
- *  - `palette` → `material`. Its product sites are all material sections
- *    (the same `GROUP_ICON_RULES`, the model-asset and react inspectors);
- *    the Material tab of the same rail.
- *  - `screwdriver-wrench` → `tool`. Its only site is a project tool's own
- *    glyph (`tool-loader.ts`); the Tool tab of the same rail.
- *
- * The six `mesh-*` operator marks carry their tone at the glyph instead,
- * with a `tonedPath`, because an operator mark tints only the element it
- * operates on and leaves the cube neutral.
- *
- * NOT toned, each for a measured reason rather than an oversight:
- *  - `globe` — Blender's World tab is red, but no site in this editor draws
- *    a globe to mean a World datablock: the two are the transform-SPACE
- *    toggle and a build profile. A red globe there reads as an error state.
- *  - `table` — its one product site is the mesh inspector's Vertex Groups
- *    panel, which is Blender's green `DATA_PT_vertex_groups`, but it renders
- *    through `SectionHeader`, which passes `tone="dim"` to every section
- *    glyph — and an explicit site tone wins by contract. A tone there would
- *    never paint.
- */
 /** The union `IconCategoryTone` declares (`@volter/sdk/looks`), and the
  *  reason `blender.style.ts` may narrow this JSON with an assertion: a tone
- *  outside it never reaches the artifact. */
+ *  outside it never reaches the artifact. Assign tones only to icon names whose
+ *  meaning has that category at every site. Outliner aliases own their tint;
+ *  generic viewport controls stay neutral. Mesh operators tint only tonedPath. */
 const TONES = ['object', 'modifier', 'material', 'tool', 'operator', 'data', 'scene', 'collection', 'select'];
 
 // The outliner's three names, aliased onto the drawings they share with the
 // generic glyphs. Same picture, its own name, so only the outliner is tinted.
 g('outliner-object', glyphs['cube'].path);
-// A MESH OBJECT's row mark is not a cube. Measured on the native 2x
-// `outliner.png` (the Cube row, src x 127-154, y 179-206): Blender draws an
-// APEX-DOWN triangle, heavily stroked, with a triangular knockout in its upper
-// half — the object-type sibling of the datablock's `properties-data` mark, in
-// the object orange rather than the data green. On the set's 16-unit grid the
-// outer triangle is (1.25,1.25)-(14.75,1.25)-(8,14.75) and the knockout
-// (4.75,4.25)-(11.25,4.25)-(8,9.75). `outliner-object` keeps the cube for the
-// generic three-object row, which has no mesh to name.
+// The mesh-object mark is an apex-down triangle with an upper triangular knockout. outliner-object remains the generic cube mark.
 g(
   'outliner-mesh',
   poly([
@@ -2539,174 +2110,10 @@ g(
 g('outliner-light', glyphs.lightbulb.path);
 g('outliner-camera', glyphs.camera.path);
 
-// The Properties editor's two package tabs, each its own name so only the
-// rail is tinted. `properties-modifiers` is the wrench Blender draws for the
-// Modifiers tab, in the modifier blue. `properties-data` is Blender's
-// MESH_DATA mark — a triangle with a vertex dot at each corner — drawn on
-// the set's own grid at its own stroke, measured from `properties-object.png`'s
-// Object Data tab (src x 10-56, y 802-830).
-/**
- * BLENDER'S MODIFIER TAB IS ONE OPEN-END WRENCH, and this glyph used to be
- * `screwdriver-wrench` — the TOOL tab's two-tool silhouette — reused under a
- * second name. At the 14px rail the two read as different marks, and the rail
- * paints them side by side, so the reuse was the loudest thing left in it.
- *
- * MEASURED off `modeling-edit-none.png` at its native 2x (the rail's OPEN
- * Modifier tab, ink x 2861..2888, y 1004..1031 — 28x28 device px, exactly the
- * box the Data tab's mark fills two cells below, which is what calibrates the
- * scale: `properties-data` spans 16.6 viewBox units, so this does too).
- * Row-by-row ink runs, thresholded at half coverage and mapped onto the set's
- * plain 16 grid, give a shape with four parts:
- *
- *  - a HEAD. Not a stroke of constant width: the outer boundary is a circle
- *    (5.08 left, 14.81 right at the head's own centre line) and the bore is a
- *    SMALLER, concentric circle, so the ring is ~2.9 thick where the handle
- *    leaves it and the jaw's flats eat into it from the other side.
- *  - a JAW cut as a SLOT WITH PARALLEL FLATS, not a wedge and not a wider
- *    bore. Both inner faces measure dx/dy = -1.00 over their whole length
- *    (left face x+y = 12.84, right face x+y = 19.41) - parallel, ~4.3 apart,
- *    on an axis at -45 degrees, opposite the handle. That parallelism is the
- *    one fact that makes the mark read as a WRENCH rather than a C: a
- *    radial-ended annulus puts the jaw's inner corner 3.8 device px off.
- *  - the slot STOPS about a unit short of the head's centre, which is what
- *    keeps the ring closed across the bottom (the frame's row y 1018 is one
- *    unbroken run, 5.08..14.57).
- *  - a HANDLE at exactly 45 degrees - its right boundary moves one device
- *    pixel per row for nine rows - ~3.7 wide, ending in a round cap whose
- *    extremes ARE the glyph's left and bottom edges.
- *
- * THE CONSTANTS BELOW ARE THE FIT, NOT THE RAW READING. Every feature above
- * is a boundary read at a 50% threshold on a 28px glyph, so each carries about
- * half a device pixel of slack, and the rim's apparent radius swings 4.5..5.0
- * depending on which row it is read from (near the top the circle is almost
- * horizontal, so half a pixel of radius moves x by two). The reading fixed the
- * STRUCTURE; the numbers were then settled by rendering the path at the
- * reference's own scale and counting the pixels where the two half-coverage
- * masks disagree. Every constant stayed inside its own reading's error bar and
- * the count came down 18 -> 5 of 298 ref ink pixels, which is the floor this
- * instrument has: the five are single pixels on antialiased diagonals.
- *
- * DRAWN AS ONE CLOSED BOUNDARY, NO KNOCKOUT AT ALL — and that is the whole
- * reason this glyph is a traced outline rather than the file's usual pile of
- * primitives. Two knockout constructions were tried and both are impossible
- * here, which is worth stating so the third attempt does not repeat them:
- *  - disc + handle capsule, then bore and slot punched. The handle has to
- *    reach PAST the bore to cover the head (its cap would need to sit at
- *    radius >4.04 to clear the bore and <3.11 to stay inside the outer
- *    circle — no such radius), so the head is DOUBLY covered exactly where
- *    the punches land: winding 2 - 1 = 1, and every knockout fills back in.
- *  - one outer polygon with the bore and the slot as holes. The slot has to
- *    run PAST the rim to cut the jaw open, and a hole outside its fill is not
- *    a hole — it is ink (winding +1), which is what it drew: a blue bar
- *    across the top-right corner. The bore and the slot also overlap, and two
- *    overlapping holes cancel to ink in their intersection.
- * The material is SIMPLY CONNECTED — the bore opens into the jaw, which opens
- * to the outside — so its boundary is one walk: the rim from the upper arm's
- * tip round to the handle, the handle's two sides and its cap, the rim again
- * to the lower jaw's tip, in along one flat, across the jaw's stop, round the
- * BACK of the bore, and out along the other flat.
- */
-const modifierWrench = () => {
-  const cx = 10.28; // the head's centre
-  const cy = 5.84;
-  const rOut = 4.78; // its outer rim
-  const rIn = 1.85; // its bore
-  const hw = 2.15; // half the jaw's opening
-  const off = -0.12; // the jaw's midline, off the head's centre
-  const stop = -1.1; // where the jaw stops, along its own axis
-  const hx = 3.2; // the handle's end cap
-  const hy = 12.75;
-  const hr = 1.87;
-  const A = [Math.SQRT1_2, -Math.SQRT1_2]; // jaw/handle axis, up-right
-  const N = [Math.SQRT1_2, Math.SQRT1_2]; // its perpendicular, down-right
-  /** A point in the jaw's own frame: `s` along the axis, `f` across it. */
-  const at = (s, f) => [cx + A[0] * s + N[0] * f, cy + A[1] * s + N[1] * f];
-  const ang = ([x, y]) => (Math.atan2(y - cy, x - cx) * 180) / Math.PI;
-  const sweep = (r, a0, a1, ox = cx, oy = cy) => {
-    const n = Math.max(2, Math.ceil(Math.abs(a1 - a0) / 6));
-    return Array.from({ length: n + 1 }, (_, i) => {
-      const a = ((a0 + ((a1 - a0) * i) / n) * Math.PI) / 180;
-      return [ox + Math.cos(a) * r, oy + Math.sin(a) * r];
-    });
-  };
-  const rim = (f) => Math.sqrt(Math.max(rOut * rOut - f * f, 0));
-  const fUp = off - hw; // the jaw's two flats, across its axis
-  const fLo = off + hw;
-  const t1 = at(rim(fUp), fUp); // the upper arm's tip
-  const t2 = at(rim(fLo), fLo); // the lower jaw's tip
-  // Where each side of the handle runs into the rim, in the same frame.
-  const offAxis = (cx - hx) * N[0] + (cy - hy) * N[1];
-  const pUp = at(-rim(-hr - offAxis), -hr - offAxis);
-  const pLo = at(-rim(hr - offAxis), hr - offAxis);
-  // BOTH FLATS CLEAR THE BORE (2.32 out against a 2.07 radius) and the jaw's
-  // stop crosses it, so the bore meets the jaw at the STOP, never at a flat.
-  const b = Math.sqrt(Math.max(rIn * rIn - stop * stop, 0));
-  const v1 = at(stop, b);
-  const v2 = at(stop, -b);
-  return poly([
-    ...sweep(rOut, ang(t1), ang(pUp) - 360), // the rim, over the top, to the handle
-    ...sweep(hr, 225, 45, hx, hy), // the handle: its two sides and its cap
-    ...sweep(rOut, ang(pLo) - 360, ang(t2) - 360), // the rim again, to the jaw
-    at(stop, fLo), // in along the jaw's lower flat, then across its stop
-    ...sweep(rIn, ang(v1), ang(v2) + 360), // round the BACK of the bore
-    at(stop, fUp), // across the stop again, and out along the upper flat
-  ]);
-};
-g('properties-modifiers', modifierWrench());
-/**
- * RE-MEASURED 2026-09-19 EDGE BY EDGE, and the mark is not a triangle drawn
- * THROUGH three vertex marks. The 2026-09-18 cut got the orientation and the
- * square vertices right and then drew the triangle as one closed `line`
- * running vertex-centre to vertex-centre, which fills each square's hollow
- * from the inside: at the 14 px the Outliner paints it the three squares
- * closed to blobs and the whole glyph read as a fat stem with two legs — a
- * Y — where Blender's reads as a clean triangle with three open boxes on it.
- *
- * The frames are NOT ANTIALIASED here (every ink pixel is the full
- * `#07b189`/`#05ae87` and every background pixel the full row), so these are
- * EXACT EDGES, not fits. Read down `modeling-object-none.png`'s Cube row (src
- * x 269..296, y 179..206 — the Outliner crop `[x2843..3452, y53..419]`'s
- * x 269..296, y 179..206), confirmed identical in `modeling-object-selected.png`
- * and `outliner.png`, and the same mark at the same size in
- * `properties-object.png`'s Data tab:
- *
- *   three squares, OUTER 8 device px, WALL 2, HOLLOW 4 — at [0,8), [20,28)
- *     and [10,18) of the mark's own 28 px box
- *   a top BAR x [6,22)  y [2,4)   — it runs wall to wall between the two top
- *     squares and stops dead at each hollow; it does NOT cross them, and it
- *     sits in the UPPER half of the hollow band, not on the squares' centre
- *   two DIAGONALS, 2 px wide, living only in y [8,20) — the gap between the
- *     top squares' bottom edge and the bottom square's top edge. The left
- *     one's centre runs x 273.04 → 279.44 down that span (a subpixel fit over
- *     twelve rows, residual under 0.1 px), which is the top square's own
- *     centre-x at the top and the bottom square's LEFT edge at the bottom;
- *     the right one is its mirror about the box's centre.
- *
- * So the three edges are drawn BETWEEN the squares and the squares are
- * FILLED-WITH-A-KNOCKOUT rather than stroked outlines — which is also why
- * their corners are square where `rframe`'s round joins gave ours a dot.
- * `poly`/`rrect` take no `W` fattening, which is what this mark wants: it is
- * fitted to pixels, like `properties-modifiers` and the editor-type wells.
- *
- * PROVED, the way the wrench was: the path rasterized at the reference's own
- * scale (28x28, 4x4 supersampled, inked at half coverage) and differenced
- * against the frame's mask — 1 of 222 reference ink pixels disagrees, a
- * single antialiased pixel on the right diagonal's edge.
- *
- * A NOTE ON THE THRESHOLD, because it nearly cost this glyph its geometry:
- * the mark is crisp everywhere EXCEPT the diagonals' edges, and read at a
- * 0.44-coverage cut those two bars mask one pixel wider than they are. Fitted
- * to that mask the diagonals want 2.35 device px; fitted to the frame's own
- * subpixel MASS (2.26 px per row across twelve rows, times cos 28.06 deg)
- * they measure 1.99, and at the honest half-coverage cut the measured width
- * is the one that scores 1/222 where the fitted width scores 3. A shape
- * tuned to a rasterizer's threshold is tuned to the rasterizer.
- */
-/** One device px of the reference, in this file's plain 16 units: the mark
- *  inks exactly 28x28 device px, which is 13.5 units once `S` has scaled it
- *  (1.25..14.75) — the same calibration the wells below use. */
+// Properties tabs use distinct names so their category tones affect only the rail.
+/** The reference's 28-device-pixel mark spans 13.5 glyph units (1.25..14.75). */
 const DATA_PX = 13.5 / 28;
-/** A measured edge, counted from the mark's own first ink pixel. */
+/** An edge offset from the glyph’s first ink pixel. */
 const dpx = (px) => 1.25 + px * DATA_PX;
 /** A bar with SQUARE ends. `cap`'s round ends would push ink past the square
  *  each diagonal dies against, into its hollow, where Blender draws none. */
@@ -2740,6 +2147,7 @@ const dataVertex = (px, py) => {
     ])
   );
 };
+/** The mesh-data mark has three hollow square vertices and three edges drawn only between their walls. Filled shapes with knockouts preserve square corners; diagonal widths use half-coverage edges without optical W expansion. */
 g(
   'properties-data',
   // the bar between the two top squares
@@ -2752,185 +2160,15 @@ g(
     dataVertex(10, 20),
 );
 
-// The Outliner's DATABLOCK row (`@volter/editor-blender`'s datablock hierarchy) —
-// the same MESH_DATA mark the Properties rail's Data tab draws, under its own
-// name because the two sites composite differently: `outliner.png`'s data
-// glyph is #07b189 over the row's #272727 and `properties-object.png`'s tab is
-// #05ae87 over the rail's #1c1c1c, one ink at α 0.791 in both — while our rail
-// paints its tab at full alpha, so the two will not stay one value.
+// The Outliner datablock uses the mesh-data shape under its own name because its row and the Properties rail apply different alpha and category colours.
 g('outliner-data', glyphs['properties-data'].path);
 
-/**
- * BLENDER'S PROPERTIES RAIL, TRACED FROM BLENDER'S OWN ICON SOURCES.
- *
- * `ED_buttons_tabs_list` (`space_buttons.cc:201-255`) draws up to sixteen tabs
- * and each one's mark is named in `buttons_context_items`
- * (`makesrna/intern/rna_space.cc:579-611`) — `ICON_SCENE` for Render,
- * `ICON_SCENE_DATA` for Scene, `ICON_OBJECT_DATA` for Object, and so on. Each
- * of those icons IS A FILE in the Blender checkout at the engine's pin:
- * `release/datafiles/icons_svg/<lowercase name>.svg`, authored on a 16-unit
- * grid inside a 1600 viewBox. The compiled `.dat` sheets under
- * `release/datafiles/icons/` are that file's BUILD PRODUCT, not its source,
- * and Blender 5.x has no combined `blender_icons.svg` master.
- *
- * WHAT I1 LEFT AND THIS REPLACES: thirteen of the sixteen were a distinct NAME
- * over a neighbouring silhouette this set already drew — a globe for World, a
- * magnet for Constraints, a palette for Material — and the remaining two
- * (`properties-modifiers`, `properties-data`) were measured pixel-by-pixel off
- * a screenshot. All sixteen are now traced from the vector source, which is
- * both more exact than a screenshot reading and far less code: the two
- * hand-measured marks above this line took two paragraphs of prose each to
- * justify one shape.
- *
- * GPL PATHS, DELIBERATELY — AND THE PACKAGE'S SPDX SAYS SO. Blender's icon
- * sources are GPL-2.0-or-later, and `@volter/editor-blender` is
- * `AGPL-3.0-only AND GPL-3.0-or-later` (`packages/blender/LICENSE`, whose
- * notice block names this set, its provenance file and the §13 combination):
- * our code is AGPL, these 197 paths are Blender's artwork conveyed as GPL-3.0,
- * and both licences' §13 permit the combination. Nothing traced may be copied
- * into an Apache/MIT part of this repo; `packages/editor`'s own icon set in
- * particular stays free of it.
- *
- * NO `S` AND NO `W`. Those two dials exist to bring OUR drawings up to
- * Blender's ink ratio (see their note at the top of this file); a traced mark
- * already has it. Measured over the traced set, Blender's marks ink 14 of the
- * 16 units (x and y both 1..15 in almost every file), against the ~14.5 the
- * dialled geometry above produces — within 3.5%, so the rail reads as one set
- * while every traced glyph keeps Blender's own proportions exactly.
- *
- * Re-derive `blender.icons.traced.json` with
- * `node blender-icons.source.mjs --trace`, which reads the checkout at
- * `$BLENDER_SRC` (default `~/volter/blender-src`; the recipe is in WORK.md
- * §Blender in the tab is Blender, "Inspection parity") and records each
- * glyph's source file and sha256 beside its path. The ordinary run consumes
- * that JSON and needs no checkout.
- */
-// The TONE rides with the traced glyph, read from that icon's own
-// `DEF_ICON_<GROUP>` macro at trace time (see the `--trace` block). The
-// Properties rail's sixteen keep their assignments in `CATEGORY_TONE` below,
-// which is where they were measured; a traced entry that carries `tone` is
-// answering for itself and that table names none of them.
+/** Trace Properties-rail icons from release/datafiles/icons_svg at the engine pin. Blender artwork is conveyed under GPL-3.0-or-later with the package’s AGPL-3.0-only code; preserve LICENSE and per-glyph source hashes in blender.icons.traced.json. Apply neither S nor W to traced paths. Rebuild the trace with node blender-icons.source.mjs --trace using BLENDER_SRC; ordinary generation consumes the saved trace JSON. */
+// Traced entries carry their source DEF_ICON_<GROUP> tone. CATEGORY_TONE supplies the Properties-rail assignments.
 for (const [name, entry] of Object.entries(traced))
   g(name, entry.path, entry.tone ? { tone: entry.tone } : undefined);
 
-// -- the EDITOR-TYPE WELLS -------------------------------------------------
-/**
- * THE TWO MARKS BLENDER PUTS IN AN AREA'S EDITOR-TYPE WELL, in the two areas
- * the reference frames photograph. Every Blender area begins with that well —
- * a 32x20 CSS rounded box, 1 px `#3c3c3c` over a constant `#272727` inner —
- * and the glyph inside it is what says which editor the area is
- * (`workspace-regions.ts`'s `groupTabs`).
- *
- * BOTH ARE READ AT NATIVE 2x AND BOTH INK EXACTLY 28x28 DEVICE PX, which is
- * the same box `properties-data` and the modifier wrench fill two paragraphs
- * up — so the same calibration carries over unchanged: 28 device px of ink is
- * 13.5 units on this file's plain 16 grid (1 unit = 2.074 device px), and the
- * ink lands at 1.25..14.75 in both axes once `S` has scaled it. Every constant
- * below is a measured pixel edge divided by that one number; none is a taste
- * call.
- *
- * Neither is antialiased in the frame — the runs are the same at a 60 and a
- * 120 threshold — so every edge quoted is exact rather than a fit.
- *
- * THE OTHER FIVE PANELS OF THIS EDITOR GET NO GLYPH HERE, and the reason is
- * that there is nothing to measure: the reference set photographs the Layout,
- * Modeling, Sculpting and Texture Paint workspaces, whose areas are the 3D
- * View, the Outliner, the Properties editor, the Timeline, the Image Editor
- * and the asset shelf. No frame contains a File Browser or an Asset Browser,
- * and Blender's editor-type dropdown is closed in all of them, so the Content
- * browser's and the Asset Library's wells have NO Blender reading at all.
- * They reuse names this set already draws (`workspace-static-panels.ts` says
- * which, and says that the choice is ours).
- */
-
-// `outliner.png`, the Outliner area's own well: ink x 27..54, y 11..38 device.
-// Six parts, each a pixel-exact rectangle with SQUARE corners (no corner in
-// this glyph loses a pixel at either threshold):
-//   the collection box   x [27,37)  y [11,17)
-//   the tree's stem      x [31,33)  y [19,37)
-//   its upper arm        x [31,39)  y [25,27)
-//   its lower arm        x [31,39)  y [35,37)
-//   the upper child box  x [41,55)  y [23,29)
-//   the lower child box  x [41,55)  y [33,39)
-// The two GAPS are the mark's own: 2 device px between the collection box and
-// the stem's top, and 2 more between each arm's end and its child box. Closing
-// either — which is what a tree icon drawn from memory does — turns three
-// separated rows into one connected bracket.
-g(
-  'outliner',
-  rrect(1.25, 1.25, 4.82, 2.89) +
-    rrect(3.18, 5.11, 0.96, 8.68) +
-    rrect(3.18, 8.0, 3.86, 0.96) +
-    rrect(3.18, 12.82, 3.86, 0.96) +
-    rrect(8.0, 7.04, 6.75, 2.89) +
-    rrect(8.0, 11.86, 6.75, 2.89),
-);
-
-/**
- * `modeling-edit-none.png`, the Properties area's own well: ink x 2870..2897,
- * y 437..464 device. TWO SOLID STADIUMS, each 28x12 device (radius = half the
- * height), stacked with 4 device px between them — and each carries a
- * KNOCKOUT, which is what makes the pair read as two sliders at different
- * settings rather than as two blank pills:
- *   stadium rows     y [437,449) and y [453,465), both x [2870,2898)
- *   upper knockout   x [2882,2896)  y [439,447)
- *   lower knockout   x [2888,2896)  y [455,463)
- * Each knockout is FLAT ON ITS LEFT and semicircular on its right: the right
- * boundary reads 2893/2894/2895/2895 down its four half-rows, which fits a
- * radius of exactly 4 device px (= half the knockout's own height) and nothing
- * else; the left boundary is one x for all eight rows. The two differ only in
- * where that flat left edge sits, and that difference is the whole picture.
- *
- * The knockouts are strictly INSIDE their stadiums (2 device px of ink on
- * every side), so the nonzero winding is well posed — unlike the wrench's two
- * failed constructions recorded above, nothing here is punched outside its own
- * fill or punched twice.
- */
-const sliderSlot = (xL, y, xR, h) => {
-  const r = h / 2;
-  const cy = y + r;
-  const cx = xR - r;
-  const cap = Array.from({ length: 13 }, (_, i) => {
-    const a = ((-90 + 15 * i) * Math.PI) / 180;
-    return [cx + Math.cos(a) * r, cy + Math.sin(a) * r];
-  });
-  return holePoly([[xL, y], ...cap, [xL, y + h]]);
-};
-g(
-  'properties',
-  rrect(1.25, 1.25, 13.5, 5.79, 2.41) +
-    sliderSlot(7.04, 2.21, 13.79, 3.86) +
-    rrect(1.25, 8.96, 13.5, 5.79, 2.41) +
-    sliderSlot(9.93, 9.92, 13.79, 3.86),
-);
-
-/**
- * BLENDER HUE-GROUPS ITS RAIL, AND THE SOURCE SAYS WHICH GROUP EACH MARK IS
- * IN. Every icon in `UI_icons.hh` is declared through a `DEF_ICON_<GROUP>`
- * macro, and `interface_icons.cc:126-132` maps those macros onto the theme
- * members `interface/resources.cc:1059-1078` reads:
- *
- *   DEF_ICON_SCENE       → TH_ICON_SCENE       `.tui.icon_scene`       #cccccc
- *   DEF_ICON_COLLECTION  → TH_ICON_COLLECTION  `.tui.icon_collection`  #ffffff
- *   DEF_ICON_OBJECT      → TH_ICON_OBJECT      `.tui.icon_object`      #e19658
- *   DEF_ICON_OBJECT_DATA → TH_ICON_OBJECT_DATA `.tui.icon_object_data` #00d4a3
- *   DEF_ICON_MODIFIER    → TH_ICON_MODIFIER    `.tui.icon_modifier`    #74a2ff
- *   DEF_ICON_SHADING     → TH_ICON_SHADING     `.tui.icon_shading`     #cc6670
- *
- * (`userdef_default_theme.c:272-277`.) So the assignments below are READ, not
- * matched by eye — and reading them corrected two guesses I1 recorded as open:
- * WORLD is `DEF_ICON_SHADING(WORLD)` (`UI_icons.hh:193`), the same red as
- * Material and Texture, NOT the scene grey; and GROUP — the Collection tab —
- * is `DEF_ICON_COLLECTION(GROUP)` (`:248`), a white of its own rather than
- * that grey either. The scene group is exactly four tabs: Render
- * (`DEF_ICON_SCENE(SCENE)`, `:188`), Output (`:187`), View Layer
- * (`DEF_ICON_SCENE(RENDER_RESULT)`, `:265`) and Scene
- * (`DEF_ICON_SCENE(SCENE_DATA)`, `:267`).
- *
- * Every object-data mark — MESH_DATA, ARMATURE_DATA, BONE_DATA and the rest of
- * the per-type set the Data tab draws — is `DEF_ICON_OBJECT_DATA`
- * (`:251-263`), which is why they share one tone rather than list one each.
- */
+/** Map UI_icons.hh DEF_ICON groups to theme categories: scene, collection, object, object-data, modifier and shading. World shares shading with Material; Render, Output, View Layer and Scene share scene; per-type Data icons share object-data. */
 const CATEGORY_TONE = {
   'properties-render': 'scene',
   'properties-output': 'scene',
@@ -2965,9 +2203,7 @@ const CATEGORY_TONE = {
   ),
 };
 for (const [name, tone] of Object.entries(CATEGORY_TONE)) {
-  // A table that names a glyph this set no longer draws is a silent no-op,
-  // and a tone assignment that silently stops applying is exactly the kind
-  // of thing nobody notices until the rail goes grey.
+  // Every tone assignment must name a glyph in this set.
   if (!glyphs[name]) throw new Error(`CATEGORY_TONE names a glyph this set does not draw: ${name}`);
   glyphs[name] = { ...glyphs[name], tone };
 }