fix(scene): hatch boundary + pattern precision for large-WCS drawings

Two precision problems were producing visibly wrong hatches on UTM-scale
(and similarly-offset) drawings even after world_offset was correct:

1. **Boundary tessellation** — `hatch_model_from_dxf`'s polyline-bulge
   arc tessellation ran the whole arc-center computation in f32. With
   vertices at WCS magnitude ~1e5 the f32 ULP is ~1 cm, which propagated
   into the sampled arc points and showed up as visibly wavy / jagged
   hatch boundaries on edges that crossed the long axis of the drawing.
   Re-run the full bulge → arc math in f64; only cast at the to_xy
   call that feeds the (also f64) boundary buffer.

2. **Pattern phase** — the per-fragment `perp / perp_step` modulo blew
   precision when `xz` was at offset-rel WCS magnitude and perp_step
   was sub-cm. Move the precision-critical bits out of f32 by keeping
   each hatch's anchor in f64 on the model side: `HatchModel` now
   carries a `world_origin: [f64; 2]` and the `boundary: [f32; 2]`
   vertices are stored as small offsets from it. The GPU vertex shader
   adds `origin` back inside the view_proj multiply for clip position
   (sub-pixel-lossy at large hatches but invisible); the `xz` varying
   that the fragment shader uses for in_polygon + pattern math stays
   in hatch-local space at full f32 precision.

   For pattern alignment across adjacent hatches with the same family,
   the GPU pipeline snaps the anchor to the family's perp/along grid
   in f64 before sending it down. Snap uses the same QCAD-PAT
   convention (`fam.dy = perp_step`, `fam.dx = along_step`) the
   shader's check_family uses — an earlier attempt used the world-frame
   `(-dx*sin + dy*cos)` formula, which produced a meaningless grid and
   no alignment.

   Trade-off: pattern phase is "global modulo per-family-grid" rather
   than tied to WCS (0,0). Same hatches still align with each other;
   the absolute phase relative to the WCS origin may differ from the
   pre-fix behaviour on small drawings.

Consumers that read `model.boundary` directly (hit_test, paper_canvas,
add_hatch round-trip) reconstruct WCS via `boundary[i] + world_origin`.
The hatch wgsl bind-group entry for binding 0 was bumped to
`VERTEX | FRAGMENT` because the vertex shader now reads `h.origin`.

The fragment-shader hatch LOD substitute + sub-pixel skip from e7675be
and the scissor plumbing from fbbac42 still work — those touched
different fields of the same struct.

Co-Authored-By: Claude Opus 4.7 <[email protected]>

src/modules/home/draw/hatch.rs

@@ -119,6 +119,7 @@ impl HatchCommand {
             color: [0.75, 0.75, 0.75, 0.85],
             angle_offset: 0.0,
             scale: 1.0,
+            world_origin: [0.0; 2],
             vp_scissor: None,
         }
     }
@@ -255,6 +256,7 @@ impl GradientCommand {
             color: [0.30, 0.60, 0.95, 0.80],
             angle_offset: 0.0,
             scale: 1.0,
+            world_origin: [0.0; 2],
             vp_scissor: None,
         }
     }
@@ -404,6 +406,7 @@ impl CadCommand for BoundaryCommand {
                     color: [0.45, 0.45, 0.45, 0.60],
                     angle_offset: 0.0,
                     scale: 1.0,
+                    world_origin: [0.0; 2],
                     vp_scissor: None,
                 };
                 return CmdResult::CommitHatch(model);

src/scene/hatch_model.rs

@@ -43,8 +43,22 @@ pub enum HatchPattern {
 /// A hatched region defined by a closed polygon boundary.
 #[derive(Clone, Debug)]
 pub struct HatchModel {
-    /// World XY coordinates of the boundary polygon vertices.
-    /// Arc so HatchModel::clone() is a pointer bump rather than a full Vec copy.
+    /// World XY anchor (in the same offset-relative coordinate space as
+    /// the rest of the scene — `world_offset` already subtracted, but
+    /// kept at f64 precision). Boundary vertices are stored as f32
+    /// offsets from this anchor so that:
+    ///   1) hit-test / paper_canvas can still read small-magnitude f32
+    ///      coords without precision loss from the f64 → f32 cast that
+    ///      would otherwise happen at large drawing extents (UTM, etc.).
+    ///   2) the GPU pipeline can pre-shift the quad in hatch-local
+    ///      space (so the fragment shader's `xz` varying stays small)
+    ///      and add `world_origin` back inside the view_proj multiply.
+    /// Reconstruct WCS-relative coords as `(world_origin.x + v.x as f64,
+    /// world_origin.y + v.y as f64)`.
+    pub world_origin: [f64; 2],
+    /// World-XY coordinates of the boundary polygon vertices, stored as
+    /// f32 offsets from `world_origin`. NaN-NaN sentinels separate
+    /// disconnected paths and must be preserved un-shifted by consumers.
     pub boundary: Arc<Vec<[f32; 2]>>,
     /// Fill pattern.
     pub pattern: HatchPattern,

src/scene/hit_test.rs

@@ -277,10 +277,15 @@ pub fn click_hit_hatch(
     bounds: Rectangle,
 ) -> Option<Handle> {
     for (&handle, hatch) in hatches {
+        // boundary verts are stored as small f32 offsets from
+        // `world_origin` (f64). Reconstruct offset-rel WCS before
+        // projecting to screen.
+        let ox = hatch.world_origin[0] as f32;
+        let oy = hatch.world_origin[1] as f32;
         let screen: Vec<Point> = hatch
             .boundary
             .iter()
-            .map(|&[x, y]| world_to_screen(Vec3::new(x, y, 0.0), view_proj, bounds))
+            .map(|&[x, y]| world_to_screen(Vec3::new(x + ox, y + oy, 0.0), view_proj, bounds))
             .collect();
         if screen.len() >= 3 && point_in_polygon(cursor, &screen) {
             return Some(handle);
@@ -315,10 +320,12 @@ pub fn box_hit_hatch(
             if hatch.boundary.is_empty() {
                 return None;
             }
+            let ox = hatch.world_origin[0] as f32;
+            let oy = hatch.world_origin[1] as f32;
             let screen: Vec<Point> = hatch
                 .boundary
                 .iter()
-                .map(|&[x, y]| world_to_screen(Vec3::new(x, y, 0.0), view_proj, bounds))
+                .map(|&[x, y]| world_to_screen(Vec3::new(x + ox, y + oy, 0.0), view_proj, bounds))
                 .collect();
             let hit = if crossing {
                 screen.iter().any(|&sp| inside(sp))
@@ -352,10 +359,12 @@ pub fn poly_hit_hatch(
             if hatch.boundary.is_empty() {
                 return None;
             }
+            let ox = hatch.world_origin[0] as f32;
+            let oy = hatch.world_origin[1] as f32;
             let screen: Vec<Point> = hatch
                 .boundary
                 .iter()
-                .map(|&[x, y]| world_to_screen(Vec3::new(x, y, 0.0), view_proj, bounds))
+                .map(|&[x, y]| world_to_screen(Vec3::new(x + ox, y + oy, 0.0), view_proj, bounds))
                 .collect();
             let hit = if crossing {
                 screen.iter().any(|&sp| point_in_polygon(sp, poly))

src/scene/mod.rs

@@ -2307,6 +2307,7 @@ impl Scene {
                     color,
                     angle_offset: 0.0,
                     scale: 1.0,
+                    world_origin: [0.0; 2],
                     vp_scissor: None,
                 });
             }
@@ -2360,6 +2361,7 @@ impl Scene {
                     color: fill_color,
                     angle_offset: 0.0,
                     scale: 1.0,
+                    world_origin: [0.0; 2],
                     vp_scissor: None,
                 });
             }
@@ -2418,21 +2420,26 @@ impl Scene {
     ) -> Option<HatchModel> {
         let [ox, oy, _oz] = world_offset;
         let normal = (dxf.normal.x, dxf.normal.y, dxf.normal.z);
-        let to_xy = |x: f64, y: f64| -> [f32; 2] {
+        // Build the boundary in f64 first so the precision-preserving
+        // origin computation below sees full WCS precision. We only cast
+        // to f32 once at the end, after subtracting the AABB centre, so
+        // the stored offsets are small-magnitude with high f32 precision
+        // even on large UTM-scale drawings.
+        let to_xy = |x: f64, y: f64| -> [f64; 2] {
             let (wx, wy, _) =
                 crate::scene::transform::ocs_point_to_wcs((x, y, dxf.elevation), normal);
-            [(wx - ox) as f32, (wy - oy) as f32]
+            [wx - ox, wy - oy]
         };
         if dxf.paths.is_empty() {
             return None;
         }
 
-        let mut boundary: Vec<[f32; 2]> = Vec::new();
+        let mut boundary: Vec<[f64; 2]> = Vec::new();
 
         for path in &dxf.paths {
             let before_path = boundary.len();
             if !boundary.is_empty() {
-                boundary.push([f32::NAN, f32::NAN]);
+                boundary.push([f64::NAN, f64::NAN]);
             }
             let path_start = boundary.len();
 
@@ -2456,54 +2463,50 @@ impl Scene {
                             if bulge.abs() < 1e-9 {
                                 boundary.push(to_xy(v0.x, v0.y));
                             } else {
-                                let p0 = [v0.x as f32, v0.y as f32];
-                                let p1 = [v1.x as f32, v1.y as f32];
-                                let theta = 4.0 * (bulge as f32).atan();
-                                let dx = p1[0] - p0[0];
-                                let dy = p1[1] - p0[1];
+                                // Run the whole bulge → arc tessellation in
+                                // f64. The old f32 path lost ~1 cm of
+                                // precision at UTM-scale WCS coordinates
+                                // (1e5 magnitude × 1 ULP), which showed up
+                                // as visibly wavy hatch-boundary arcs.
+                                let theta = 4.0 * bulge.atan();
+                                let dx = v1.x - v0.x;
+                                let dy = v1.y - v0.y;
                                 let d = (dx * dx + dy * dy).sqrt();
-                                if d < 1e-9 {
+                                if d < 1e-12 {
                                     boundary.push(to_xy(v0.x, v0.y));
                                     continue;
                                 }
                                 let r = (d * 0.5) / (theta * 0.5).sin().abs();
-                                let mx = (p0[0] + p1[0]) * 0.5;
-                                let my = (p0[1] + p1[1]) * 0.5;
+                                let mx = (v0.x + v1.x) * 0.5;
+                                let my = (v0.y + v1.y) * 0.5;
                                 let px = -dy / d;
                                 let py = dx / d;
-                                let sign = if bulge > 0.0 { 1.0_f32 } else { -1.0_f32 };
+                                let sign = if bulge > 0.0 { 1.0_f64 } else { -1.0_f64 };
                                 let center_offset = r * (theta * 0.5).cos();
                                 let cx = mx + sign * px * center_offset;
                                 let cy = my + sign * py * center_offset;
-                                let a0 = (p0[1] - cy).atan2(p0[0] - cx);
-                                let a1 = (p1[1] - cy).atan2(p1[0] - cx);
+                                let a0 = (v0.y - cy).atan2(v0.x - cx);
+                                let a1 = (v1.y - cy).atan2(v1.x - cx);
                                 let mut sweep = a1 - a0;
+                                const TAU: f64 = std::f64::consts::TAU;
                                 if bulge > 0.0 {
                                     if sweep <= 0.0 {
-                                        sweep += std::f32::consts::TAU;
+                                        sweep += TAU;
                                     }
                                 } else if sweep >= 0.0 {
-                                    sweep -= std::f32::consts::TAU;
+                                    sweep -= TAU;
                                 }
-                                if sweep.abs() < 1e-7 {
-                                    sweep = if bulge > 0.0 {
-                                        std::f32::consts::TAU
-                                    } else {
-                                        -std::f32::consts::TAU
-                                    };
+                                if sweep.abs() < 1e-9 {
+                                    sweep = if bulge > 0.0 { TAU } else { -TAU };
                                 }
-                                let r64 = r as f64;
                                 let segs = tessellate::arc_segments(
-                                    r64,
-                                    sweep.abs() as f64,
-                                    tessellate::fill_chord_tol(r64),
+                                    r,
+                                    sweep.abs(),
+                                    tessellate::fill_chord_tol(r),
                                 );
                                 for j in 0..segs {
-                                    let a = a0 + sweep * (j as f32 / segs as f32);
-                                    boundary.push(to_xy(
-                                        (cx + r * a.cos()) as f64,
-                                        (cy + r * a.sin()) as f64,
-                                    ));
+                                    let a = a0 + sweep * (j as f64 / segs as f64);
+                                    boundary.push(to_xy(cx + r * a.cos(), cy + r * a.sin()));
                                 }
                             }
                         }
@@ -2656,13 +2659,47 @@ impl Scene {
         } else {
             dxf.pattern.name.clone()
         };
+
+        // Precision-preserving cast f64 → f32: pick an `world_origin`
+        // anchor (boundary AABB centre in f64) and store every vertex
+        // as a small f32 offset from it. NaN separators are preserved
+        // so the in_polygon ray-cast still sees the path breaks.
+        let mut min_x = f64::INFINITY;
+        let mut min_y = f64::INFINITY;
+        let mut max_x = f64::NEG_INFINITY;
+        let mut max_y = f64::NEG_INFINITY;
+        for &[x, y] in &boundary {
+            if x.is_finite() && y.is_finite() {
+                if x < min_x { min_x = x; }
+                if y < min_y { min_y = y; }
+                if x > max_x { max_x = x; }
+                if y > max_y { max_y = y; }
+            }
+        }
+        let world_origin = if min_x.is_finite() && min_y.is_finite() {
+            [(min_x + max_x) * 0.5, (min_y + max_y) * 0.5]
+        } else {
+            [0.0, 0.0]
+        };
+        let boundary_f32: Vec<[f32; 2]> = boundary
+            .iter()
+            .map(|&[x, y]| {
+                if x.is_finite() && y.is_finite() {
+                    [(x - world_origin[0]) as f32, (y - world_origin[1]) as f32]
+                } else {
+                    [f32::NAN, f32::NAN]
+                }
+            })
+            .collect();
+
         Some(HatchModel {
-            boundary: std::sync::Arc::new(boundary),
+            boundary: std::sync::Arc::new(boundary_f32),
             pattern,
             name,
             color,
             angle_offset: dxf.pattern_angle as f32,
             scale: dxf.pattern_scale as f32,
+            world_origin,
             vp_scissor: None,
         })
     }
@@ -2812,6 +2849,7 @@ impl Scene {
             color,
             angle_offset: 0.0,
             scale: 1.0,
+            world_origin: [0.0; 2],
             vp_scissor: None,
         }
     }
@@ -2822,10 +2860,12 @@ impl Scene {
             model.pattern,
             crate::scene::hatch_model::HatchPattern::Solid
         );
+        let [wx, wy] = model.world_origin;
         let verts: Vec<Vector2> = model
             .boundary
             .iter()
-            .map(|&[x, y]| Vector2::new(x as f64, y as f64))
+            .filter(|v| v[0].is_finite() && v[1].is_finite())
+            .map(|&[x, y]| Vector2::new(x as f64 + wx, y as f64 + wy))
             .collect();
         let edge = PolylineEdge::new(verts, true);
         let mut path = BoundaryPath::external();

src/scene/paper_canvas.rs

@@ -171,11 +171,14 @@ fn draw_hatch(frame: &mut canvas::Frame, hatch: &HatchModel, to_px: &impl Fn(f32
     let [r, g, b, a] = hatch.color;
     let color = Color { r, g, b, a };
 
+    // Reconstruct offset-rel WCS from stored small offsets + f64 anchor.
+    let ox = hatch.world_origin[0] as f32;
+    let oy = hatch.world_origin[1] as f32;
     let path = canvas::Path::new(|builder| {
-        let first = to_px(hatch.boundary[0][0], hatch.boundary[0][1]);
+        let first = to_px(hatch.boundary[0][0] + ox, hatch.boundary[0][1] + oy);
         builder.move_to(first);
         for &[x, y] in &hatch.boundary[1..] {
-            builder.line_to(to_px(x, y));
+            builder.line_to(to_px(x + ox, y + oy));
         }
         builder.close();
     });

src/scene/pipeline/hatch_gpu.rs

@@ -39,7 +39,7 @@ impl HatchVertex {
 
 // ── Uniform structs ───────────────────────────────────────────────────────
 
-/// Per-hatch parameters (binding 0).  Must be 64 bytes.
+/// Per-hatch parameters (binding 0).  Must be 80 bytes (16-byte aligned).
 ///
 /// `mode` encoding:
 ///   0 → Pattern  (families in FamilyBatchData)
@@ -48,17 +48,24 @@ impl HatchVertex {
 #[repr(C)]
 #[derive(Copy, Clone, bytemuck::Pod, bytemuck::Zeroable)]
 pub struct HatchUniformData {
-    pub color: [f32; 4],   //  0: primary RGBA / gradient start
-    pub color2: [f32; 4],  // 16: gradient end color
-    pub mode: u32,         // 32: 0=pattern, 1=solid, 2=gradient
-    pub vertex_count: u32, // 36: boundary vertex count
-    pub angle_offset: f32, // 40: pattern rotation (radians)
-    pub scale: f32,        // 44: pattern scale multiplier
-    pub grad_cos: f32,     // 48: gradient direction cos
-    pub grad_sin: f32,     // 52: gradient direction sin
-    pub grad_min: f32,     // 56: gradient proj_min
-    pub grad_range: f32,   // 60: gradient proj_range
-} // total 64 bytes
+    pub color: [f32; 4],     //  0: primary RGBA / gradient start
+    pub color2: [f32; 4],    // 16: gradient end color
+    pub mode: u32,           // 32: 0=pattern, 1=solid, 2=gradient
+    pub vertex_count: u32,   // 36: boundary vertex count
+    pub angle_offset: f32,   // 40: pattern rotation (radians)
+    pub scale: f32,          // 44: pattern scale multiplier
+    pub grad_cos: f32,       // 48: gradient direction cos
+    pub grad_sin: f32,       // 52: gradient direction sin
+    pub grad_min: f32,       // 56: gradient proj_min
+    pub grad_range: f32,     // 60: gradient proj_range
+    pub origin: [f32; 2],    // 64: hatch-local origin (boundary AABB centre).
+    //                       //     The pattern fragment shader subtracts this
+    //                       //     from xz before the perp/perp_step quotient
+    //                       //     so f32 doesn't catastrophically cancel when
+    //                       //     world coords are large and pattern spacing
+    //                       //     is small (e.g. UTM drawing + sub-mm hatch).
+    pub _pad: [f32; 2],      // 72: 16-byte alignment
+} // total 80 bytes
 
 /// Boundary polygon (binding 1).  Matches WGSL `Boundary`.
 #[repr(C)]
@@ -153,7 +160,68 @@ impl HatchGpu {
         };
         let diag = ((max_x - min_x).powi(2) + (max_y - min_y).powi(2)).sqrt();
         let pad = (diag * 0.8 + max_spacing * 2.0 * model.scale).max(1.0);
-        let (x0, x1, y0, y1) = (min_x - pad, max_x + pad, min_y - pad, max_y + pad);
+        // ── Pattern phase anchor.
+        //
+        // Boundary verts are already stored as small f32 offsets from
+        // `model.world_origin` (an f64 anchor near the AABB centre), so
+        // the in_polygon ray-cast and the pattern math both run at full
+        // f32 precision in the fragment shader. The remaining question
+        // is *which* offset-rel point we hand to the vertex shader as
+        // `h.origin` (so the quad lands at the correct WCS position
+        // after `view_proj * (v.pos + h.origin)`).
+        //
+        // For pattern alignment across adjacent hatches with the same
+        // family, snap `h.origin` to the family's perp/along grid in
+        // f64. Without the snap, each hatch's pattern phase is per-hatch
+        // and stripes don't continue across hatch boundaries (you see
+        // pattern misalignment along a road etc).
+        let mut origin = model.world_origin;
+        if let HatchPattern::Pattern(families) = &model.pattern {
+            if let Some(fam) = families.first() {
+                // Must mirror the shader's pattern math exactly. QCAD PAT
+                // convention (matching build_family_batch): `dy` is the
+                // perpendicular spacing and `dx` is the along-line phase
+                // shift, both in family-local space. Multiplied by
+                // `model.scale` in the fragment shader. Combined angle
+                // (family + pattern_angle) rotates the perp/along axes.
+                let angle = (fam.angle_deg.to_radians() + model.angle_offset) as f64;
+                let cos_a = angle.cos();
+                let sin_a = angle.sin();
+                let scale = model.scale as f64;
+                let perp_step = fam.dy as f64 * scale;
+                let along_step = fam.dx as f64 * scale;
+                let o_perp = -origin[0] * sin_a + origin[1] * cos_a;
+                let o_along = origin[0] * cos_a + origin[1] * sin_a;
+                let snapped_perp = if perp_step.abs() > 1e-12 {
+                    (o_perp / perp_step).round() * perp_step
+                } else {
+                    o_perp
+                };
+                let snapped_along = if along_step.abs() > 1e-12 {
+                    (o_along / along_step).round() * along_step
+                } else {
+                    o_along
+                };
+                origin = [
+                    -snapped_perp * sin_a + snapped_along * cos_a,
+                    snapped_perp * cos_a + snapped_along * sin_a,
+                ];
+            }
+        }
+        // Boundary buffer + quad now live in "snap-shifted" hatch-local
+        // coords: subtract any drift between model.world_origin and the
+        // snapped origin so the in_polygon ray-cast in fragment shader
+        // (which reads the boundary verts straight) lines up with v.xz.
+        let drift = [
+            (model.world_origin[0] - origin[0]) as f32,
+            (model.world_origin[1] - origin[1]) as f32,
+        ];
+        let (x0, x1, y0, y1) = (
+            min_x + drift[0] - pad,
+            max_x + drift[0] + pad,
+            min_y + drift[1] - pad,
+            max_y + drift[1] + pad,
+        );
 
         let quad = [
             HatchVertex {
@@ -187,13 +255,13 @@ impl HatchGpu {
             usage: wgpu::BufferUsages::VERTEX,
         });
 
-        // ── Gradient: projection range ───────────────────────────────────
+        // ── Gradient: projection range (in snapped-local space) ──────────
         let (grad_min, grad_range) = if mode == 2 {
             let projs: Vec<f32> = model
                 .boundary
                 .iter()
                 .filter(|v| v[0].is_finite() && v[1].is_finite())
-                .map(|&[x, y]| x * grad_cos + y * grad_sin)
+                .map(|&[x, y]| (x + drift[0]) * grad_cos + (y + drift[1]) * grad_sin)
                 .collect();
             if projs.is_empty() {
                 (0.0, 1.0)
@@ -219,9 +287,11 @@ impl HatchGpu {
             grad_sin,
             grad_min,
             grad_range,
+            origin: [origin[0] as f32, origin[1] as f32],
+            _pad: [0.0; 2],
         };
 
-        // ── BoundaryData ─────────────────────────────────────────────────
+        // ── BoundaryData (in snapped-local space) ────────────────────────
         let mut boundary_data = BoundaryData {
             verts: [[0.0; 4]; MAX_HATCH_BOUNDARY_VERTS],
         };
@@ -231,7 +301,13 @@ impl HatchGpu {
             .take(MAX_HATCH_BOUNDARY_VERTS)
             .enumerate()
         {
-            boundary_data.verts[i] = [x, y, 0.0, 0.0];
+            // Preserve NaN separators — they signal path breaks to the
+            // in_polygon ray-cast, must not be shifted into finite numbers.
+            if x.is_finite() && y.is_finite() {
+                boundary_data.verts[i] = [x + drift[0], y + drift[1], 0.0, 0.0];
+            } else {
+                boundary_data.verts[i] = [f32::NAN, f32::NAN, 0.0, 0.0];
+            }
         }
 
         // ── FamilyBatchData ───────────────────────────────────────────────

src/scene/pipeline/mod.rs

@@ -220,9 +220,13 @@ impl Pipeline {
         });
 
         // ── Hatch pipeline ─────────────────────────────────────────────────
-        let hatch_entry = |binding: u32| wgpu::BindGroupLayoutEntry {
+        // binding 0 (HatchUniforms) is read by the vertex shader too — it
+        // pulls `origin` to undo the CPU-side hatch-local pre-shift when
+        // computing clip position. bindings 1 (Boundary) and 2
+        // (FamilyBatch) stay fragment-only.
+        let hatch_entry = |binding: u32, vis: wgpu::ShaderStages| wgpu::BindGroupLayoutEntry {
             binding,
-            visibility: wgpu::ShaderStages::FRAGMENT,
+            visibility: vis,
             ty: wgpu::BindingType::Buffer {
                 ty: wgpu::BufferBindingType::Uniform,
                 has_dynamic_offset: false,
@@ -230,9 +234,15 @@ impl Pipeline {
             },
             count: None,
         };
+        let frag = wgpu::ShaderStages::FRAGMENT;
+        let vert_frag = wgpu::ShaderStages::VERTEX | wgpu::ShaderStages::FRAGMENT;
         let hatch_bgl1 = device.create_bind_group_layout(&wgpu::BindGroupLayoutDescriptor {
             label: Some("hatch.bgl1"),
-            entries: &[hatch_entry(0), hatch_entry(1), hatch_entry(2)],
+            entries: &[
+                hatch_entry(0, vert_frag),
+                hatch_entry(1, frag),
+                hatch_entry(2, frag),
+            ],
         });
 
         let hatch_layout = device.create_pipeline_layout(&wgpu::PipelineLayoutDescriptor {

src/shaders/hatch.wgsl

@@ -39,6 +39,12 @@ struct HatchUniforms {
     grad_sin:     f32,        // 52: gradient direction sin
     grad_min:     f32,        // 56: gradient proj_min
     grad_range:   f32,        // 60: gradient proj_range
+    origin:       vec2<f32>,  // 64: hatch-local pattern origin (boundary
+    //                        //     centre). Pattern math runs against
+    //                        //     `xz - origin` so the f32 modulo doesn't
+    //                        //     catastrophically cancel when world coords
+    //                        //     are large and pattern spacing is small.
+    _pad:         vec2<f32>,  // 72: 16-byte alignment
 }
 @group(1) @binding(0) var<uniform> h: HatchUniforms;
 
@@ -81,7 +87,16 @@ struct VOut {
 
 @vertex fn vs_main(v: VIn) -> VOut {
     var o: VOut;
-    o.clip = u.view_proj * vec4<f32>(v.pos, 1.0);
+    // v.pos comes in pre-shifted to hatch-local space (CPU subtracted
+    // h.origin from the quad). Add origin back inside the view_proj
+    // multiply to land at the correct world-space clip position. The
+    // f32+f32 addition can lose sub-mm precision at very far hatches
+    // but the result feeds view_proj for screen NDC where sub-pixel
+    // jitter is invisible. The local-space `xz` we export to the
+    // fragment shader stays at full f32 precision so the in_polygon /
+    // pattern math doesn't catastrophically cancel.
+    let world = vec3<f32>(v.pos.x + h.origin.x, v.pos.y + h.origin.y, v.pos.z);
+    o.clip = u.view_proj * vec4<f32>(world, 1.0);
     o.xz   = vec2<f32>(v.pos.x, v.pos.y);
     return o;
 }
@@ -213,7 +228,12 @@ fn check_family(
         }
     }
 
-    // 4. Pattern: evaluate each line family.
+    // 4. Pattern: v.xz is already in hatch-local space (CPU pre-shifted
+    //    quad + BoundaryData by h.origin). The PAT family origin
+    //    (typically (0,0)) is treated as relative to h.origin, so
+    //    pattern phase is per-hatch rather than world-aligned — that
+    //    trade-off is the price of f32 precision at large drawing
+    //    extents.
     let cos_off = cos(h.angle_offset);
     let sin_off = sin(h.angle_offset);
     for (var i = 0u; i < f.n_families; i++) {