diff --git a/export/src/lib.rs b/export/src/lib.rs
index 6db4161c..167e3559 100644
--- a/export/src/lib.rs
+++ b/export/src/lib.rs
@@ -54,7 +54,7 @@ fn export_svg<W: Write>(scene: &Scene, writer: &mut W) -> io::Result<()> {
         view_box.size().y()
     )?;
     for &DrawPath { paint: paint_id, ref outline, ref name, .. } in scene.draw_paths() {
-        let paint = scene.palette().paints.get(paint_id.0 as usize).unwrap();
+        let paint = scene.get_paint(paint_id);
         write!(writer, "    <path")?;
         if !name.is_empty() {
             write!(writer, " id=\"{}\"", name)?;
@@ -78,7 +78,7 @@ fn export_pdf<W: Write>(scene: &Scene, writer: &mut W) -> io::Result<()> {
 
     for &DrawPath { paint: paint_id, ref outline, .. } in scene.draw_paths() {
         // TODO(pcwalton): Gradients and patterns.
-        let paint = scene.palette().paints.get(paint_id.0 as usize).unwrap();
+        let paint = scene.get_paint(paint_id);
         if paint.is_color() {
             pdf.set_fill_color(paint.base_color());
         }
@@ -182,7 +182,7 @@ fn export_ps<W: Write>(scene: &Scene, writer: &mut W) -> io::Result<()> {
         }
 
         // TODO(pcwalton): Gradients and patterns.
-        let paint = scene.palette().paints.get(paint_id.0 as usize).unwrap();
+        let paint = scene.get_paint(paint_id);
         if paint.is_color() {
             let color = paint.base_color();
             writeln!(writer, "{} {} {} setrgbcolor", color.r, color.g, color.b)?;
diff --git a/renderer/src/concurrent/executor.rs b/renderer/src/concurrent/executor.rs
index 150f4d32..dc32afa1 100644
--- a/renderer/src/concurrent/executor.rs
+++ b/renderer/src/concurrent/executor.rs
@@ -21,6 +21,7 @@ pub trait Executor {
                           where T: Send, F: Fn(usize) -> T + Send + Sync;
 }
 
+/// An executor that simply executes tasks sequentially in the same thread.
 pub struct SequentialExecutor;
 
 impl Executor for SequentialExecutor {
diff --git a/renderer/src/concurrent/rayon.rs b/renderer/src/concurrent/rayon.rs
index e11d28b1..f61724de 100644
--- a/renderer/src/concurrent/rayon.rs
+++ b/renderer/src/concurrent/rayon.rs
@@ -8,11 +8,12 @@
 // option. This file may not be copied, modified, or distributed
 // except according to those terms.
 
-//! An implementation of the executor using the Rayon library.
+//! An executor that parallelizes tasks across all CPUs using the Rayon library.
 
 use crate::concurrent::executor::Executor;
 use rayon::iter::{IntoParallelIterator, ParallelIterator};
 
+/// An executor that parallelizes tasks across all CPUs using the Rayon library.
 pub struct RayonExecutor;
 
 impl Executor for RayonExecutor {
diff --git a/renderer/src/concurrent/scene_proxy.rs b/renderer/src/concurrent/scene_proxy.rs
index e5a88db1..d0079bd3 100644
--- a/renderer/src/concurrent/scene_proxy.rs
+++ b/renderer/src/concurrent/scene_proxy.rs
@@ -32,17 +32,30 @@ use std::thread;
 
 const MAX_MESSAGES_IN_FLIGHT: usize = 1024;
 
+/// A version of `Scene` that proxies all method calls out to a separate thread.
 pub struct SceneProxy {
     sender: Sender<MainToWorkerMsg>,
     receiver: Receiver<RenderCommand>,
 }
 
 impl SceneProxy {
+    /// Creates a new scene proxy using the given renderer GPU API level and executor to execute
+    /// CPU tasks.
+    /// 
+    /// If you want to use multiple threads, you typically pass in a `RayonExecutor` here. If you
+    /// want to use a single thread (perhaps because you're in a WebAssembly environment), pass a
+    /// `SequentialExecutor`.
     pub fn new<E>(renderer_level: RendererLevel, executor: E) -> SceneProxy
                   where E: Executor + Send + 'static {
         SceneProxy::from_scene(Scene::new(), renderer_level, executor)
     }
 
+    /// Wraps an existing scene in a scene proxy using the given renderer GPU API level an executor
+    /// to execute CPU tasks.
+    /// 
+    /// If you want to use multiple threads, you typically pass in a `RayonExecutor` here. If you
+    /// want to use a single thread (perhaps because you're in a WebAssembly environment), pass a
+    /// `SequentialExecutor`.
     pub fn from_scene<E>(scene: Scene, renderer_level: RendererLevel, executor: E)
                          -> SceneProxy
                          where E: Executor + Send + 'static {
@@ -58,22 +71,25 @@ impl SceneProxy {
         SceneProxy { sender: main_to_worker_sender, receiver: worker_to_main_receiver }
     }
 
+    /// Replaces the wrapped scene with a new one, discarding the old scene.
     #[inline]
     pub fn replace_scene(&self, new_scene: Scene) {
         self.sender.send(MainToWorkerMsg::ReplaceScene(new_scene)).unwrap();
     }
 
+    /// Sets the view box of the scene, which defines the visible rectangle.
     #[inline]
     pub fn set_view_box(&self, new_view_box: RectF) {
         self.sender.send(MainToWorkerMsg::SetViewBox(new_view_box)).unwrap();
     }
 
+    /// Constructs a scene and queues up the commands needed to render it.
     #[inline]
     pub fn build(&self, options: BuildOptions) {
         self.sender.send(MainToWorkerMsg::Build(options)).unwrap();
     }
 
-    /// Sends all queued commands to the given renderer.
+    /// Sends all queued commands to the given renderer to render the wrapped scene.
     #[inline]
     pub fn render<D>(&mut self, renderer: &mut Renderer<D>) where D: Device {
         renderer.begin_scene();
@@ -104,6 +120,7 @@ impl SceneProxy {
         self.render(renderer);
     }
 
+    /// Returns a copy of the wrapped scene.
     #[inline]
     pub fn copy_scene(&self) -> Scene {
         let (sender, receiver) = crossbeam_channel::bounded(MAX_MESSAGES_IN_FLIGHT);
diff --git a/renderer/src/gpu/d3d11/mod.rs b/renderer/src/gpu/d3d11/mod.rs
index eebeb578..b3168830 100644
--- a/renderer/src/gpu/d3d11/mod.rs
+++ b/renderer/src/gpu/d3d11/mod.rs
@@ -8,5 +8,9 @@
 // option. This file may not be copied, modified, or distributed
 // except according to those terms.
 
+//! A GPU compute-based renderer that uses functionality available in Direct3D 11.
+//! 
+//! This renderer supports OpenGL at least 4.3, OpenGL ES at least 3.1, and Metal of any version.
+
 pub mod renderer;
 pub mod shaders;
diff --git a/renderer/src/gpu/d3d11/renderer.rs b/renderer/src/gpu/d3d11/renderer.rs
index f8982f9d..fe7414b3 100644
--- a/renderer/src/gpu/d3d11/renderer.rs
+++ b/renderer/src/gpu/d3d11/renderer.rs
@@ -8,6 +8,10 @@
 // option. This file may not be copied, modified, or distributed
 // except according to those terms.
 
+//! A GPU compute-based renderer that uses functionality available in Direct3D 11.
+//! 
+//! This renderer supports OpenGL at least 4.3, OpenGL ES at least 3.1, and Metal of any version.
+
 use crate::gpu::d3d11::shaders::{BOUND_WORKGROUP_SIZE, DICE_WORKGROUP_SIZE};
 use crate::gpu::d3d11::shaders::{PROPAGATE_WORKGROUP_SIZE, ProgramsD3D11, SORT_WORKGROUP_SIZE};
 use crate::gpu::perf::TimeCategory;
diff --git a/renderer/src/gpu/d3d11/shaders.rs b/renderer/src/gpu/d3d11/shaders.rs
index 52fe5f24..d5efdf5e 100644
--- a/renderer/src/gpu/d3d11/shaders.rs
+++ b/renderer/src/gpu/d3d11/shaders.rs
@@ -8,29 +8,31 @@
 // option. This file may not be copied, modified, or distributed
 // except according to those terms.
 
+//! Shaders and vertex specifications for the Direct3D 11-level renderer.
+
 use crate::gpu::shaders::TileProgramCommon;
 use crate::tiles::{TILE_HEIGHT, TILE_WIDTH};
 use pathfinder_gpu::{ComputeDimensions, Device};
 use pathfinder_resources::ResourceLoader;
 
-pub const BOUND_WORKGROUP_SIZE: u32 = 64;
-pub const DICE_WORKGROUP_SIZE: u32 = 64;
-pub const BIN_WORKGROUP_SIZE: u32 = 64;
-pub const PROPAGATE_WORKGROUP_SIZE: u32 = 64;
-pub const SORT_WORKGROUP_SIZE: u32 = 64;
+pub(crate) const BOUND_WORKGROUP_SIZE: u32 = 64;
+pub(crate) const DICE_WORKGROUP_SIZE: u32 = 64;
+pub(crate) const BIN_WORKGROUP_SIZE: u32 = 64;
+pub(crate) const PROPAGATE_WORKGROUP_SIZE: u32 = 64;
+pub(crate) const SORT_WORKGROUP_SIZE: u32 = 64;
 
-pub struct ProgramsD3D11<D> where D: Device {
-    pub bound_program: BoundProgramD3D11<D>,
-    pub dice_program: DiceProgramD3D11<D>,
-    pub bin_program: BinProgramD3D11<D>,
-    pub propagate_program: PropagateProgramD3D11<D>,
-    pub sort_program: SortProgramD3D11<D>,
-    pub fill_program: FillProgramD3D11<D>,
-    pub tile_program: TileProgramD3D11<D>,
+pub(crate) struct ProgramsD3D11<D> where D: Device {
+    pub(crate) bound_program: BoundProgramD3D11<D>,
+    pub(crate) dice_program: DiceProgramD3D11<D>,
+    pub(crate) bin_program: BinProgramD3D11<D>,
+    pub(crate) propagate_program: PropagateProgramD3D11<D>,
+    pub(crate) sort_program: SortProgramD3D11<D>,
+    pub(crate) fill_program: FillProgramD3D11<D>,
+    pub(crate) tile_program: TileProgramD3D11<D>,
 }
 
 impl<D> ProgramsD3D11<D> where D: Device {
-    pub fn new(device: &D, resources: &dyn ResourceLoader) -> ProgramsD3D11<D> {
+    pub(crate) fn new(device: &D, resources: &dyn ResourceLoader) -> ProgramsD3D11<D> {
         ProgramsD3D11 {
             bound_program: BoundProgramD3D11::new(device, resources),
             dice_program: DiceProgramD3D11::new(device, resources),
@@ -43,23 +45,23 @@ impl<D> ProgramsD3D11<D> where D: Device {
     }
 }
 
-pub struct PropagateProgramD3D11<D> where D: Device {
-    pub program: D::Program,
-    pub framebuffer_tile_size_uniform: D::Uniform,
-    pub column_count_uniform: D::Uniform,
-    pub first_alpha_tile_index_uniform: D::Uniform,
-    pub draw_metadata_storage_buffer: D::StorageBuffer,
-    pub clip_metadata_storage_buffer: D::StorageBuffer,
-    pub backdrops_storage_buffer: D::StorageBuffer,
-    pub draw_tiles_storage_buffer: D::StorageBuffer,
-    pub clip_tiles_storage_buffer: D::StorageBuffer,
-    pub z_buffer_storage_buffer: D::StorageBuffer,
-    pub first_tile_map_storage_buffer: D::StorageBuffer,
-    pub alpha_tiles_storage_buffer: D::StorageBuffer,
+pub(crate) struct PropagateProgramD3D11<D> where D: Device {
+    pub(crate) program: D::Program,
+    pub(crate) framebuffer_tile_size_uniform: D::Uniform,
+    pub(crate) column_count_uniform: D::Uniform,
+    pub(crate) first_alpha_tile_index_uniform: D::Uniform,
+    pub(crate) draw_metadata_storage_buffer: D::StorageBuffer,
+    pub(crate) clip_metadata_storage_buffer: D::StorageBuffer,
+    pub(crate) backdrops_storage_buffer: D::StorageBuffer,
+    pub(crate) draw_tiles_storage_buffer: D::StorageBuffer,
+    pub(crate) clip_tiles_storage_buffer: D::StorageBuffer,
+    pub(crate) z_buffer_storage_buffer: D::StorageBuffer,
+    pub(crate) first_tile_map_storage_buffer: D::StorageBuffer,
+    pub(crate) alpha_tiles_storage_buffer: D::StorageBuffer,
 }
 
 impl<D> PropagateProgramD3D11<D> where D: Device {
-    pub fn new(device: &D, resources: &dyn ResourceLoader) -> PropagateProgramD3D11<D> {
+    pub(crate) fn new(device: &D, resources: &dyn ResourceLoader) -> PropagateProgramD3D11<D> {
         let mut program = device.create_compute_program(resources, "d3d11/propagate");
         let local_size = ComputeDimensions { x: PROPAGATE_WORKGROUP_SIZE, y: 1, z: 1 };
         device.set_compute_program_local_size(&mut program, local_size);
@@ -93,18 +95,18 @@ impl<D> PropagateProgramD3D11<D> where D: Device {
     }
 }
 
-pub struct FillProgramD3D11<D> where D: Device {
-    pub program: D::Program,
-    pub dest_image: D::ImageParameter,
-    pub area_lut_texture: D::TextureParameter,
-    pub alpha_tile_range_uniform: D::Uniform,
-    pub fills_storage_buffer: D::StorageBuffer,
-    pub tiles_storage_buffer: D::StorageBuffer,
-    pub alpha_tiles_storage_buffer: D::StorageBuffer,
+pub(crate) struct FillProgramD3D11<D> where D: Device {
+    pub(crate) program: D::Program,
+    pub(crate) dest_image: D::ImageParameter,
+    pub(crate) area_lut_texture: D::TextureParameter,
+    pub(crate) alpha_tile_range_uniform: D::Uniform,
+    pub(crate) fills_storage_buffer: D::StorageBuffer,
+    pub(crate) tiles_storage_buffer: D::StorageBuffer,
+    pub(crate) alpha_tiles_storage_buffer: D::StorageBuffer,
 }
 
 impl<D> FillProgramD3D11<D> where D: Device {
-    pub fn new(device: &D, resources: &dyn ResourceLoader) -> FillProgramD3D11<D> {
+    pub(crate) fn new(device: &D, resources: &dyn ResourceLoader) -> FillProgramD3D11<D> {
         let mut program = device.create_compute_program(resources, "d3d11/fill");
         let local_size = ComputeDimensions { x: TILE_WIDTH, y: TILE_HEIGHT / 4, z: 1 };
         device.set_compute_program_local_size(&mut program, local_size);
@@ -128,14 +130,14 @@ impl<D> FillProgramD3D11<D> where D: Device {
     }
 }
 
-pub struct TileProgramD3D11<D> where D: Device {
-    pub common: TileProgramCommon<D>,
-    pub load_action_uniform: D::Uniform,
-    pub clear_color_uniform: D::Uniform,
-    pub framebuffer_tile_size_uniform: D::Uniform,
-    pub dest_image: D::ImageParameter,
-    pub tiles_storage_buffer: D::StorageBuffer,
-    pub first_tile_map_storage_buffer: D::StorageBuffer,
+pub(crate) struct TileProgramD3D11<D> where D: Device {
+    pub(crate) common: TileProgramCommon<D>,
+    pub(crate) load_action_uniform: D::Uniform,
+    pub(crate) clear_color_uniform: D::Uniform,
+    pub(crate) framebuffer_tile_size_uniform: D::Uniform,
+    pub(crate) dest_image: D::ImageParameter,
+    pub(crate) tiles_storage_buffer: D::StorageBuffer,
+    pub(crate) first_tile_map_storage_buffer: D::StorageBuffer,
 }
 
 impl<D> TileProgramD3D11<D> where D: Device {
@@ -164,20 +166,20 @@ impl<D> TileProgramD3D11<D> where D: Device {
     }
 }
 
-pub struct BinProgramD3D11<D> where D: Device {
-    pub program: D::Program,
-    pub microline_count_uniform: D::Uniform,
-    pub max_fill_count_uniform: D::Uniform,
-    pub microlines_storage_buffer: D::StorageBuffer,
-    pub metadata_storage_buffer: D::StorageBuffer,
-    pub indirect_draw_params_storage_buffer: D::StorageBuffer,
-    pub fills_storage_buffer: D::StorageBuffer,
-    pub tiles_storage_buffer: D::StorageBuffer,
-    pub backdrops_storage_buffer: D::StorageBuffer,
+pub(crate) struct BinProgramD3D11<D> where D: Device {
+    pub(crate) program: D::Program,
+    pub(crate) microline_count_uniform: D::Uniform,
+    pub(crate) max_fill_count_uniform: D::Uniform,
+    pub(crate) microlines_storage_buffer: D::StorageBuffer,
+    pub(crate) metadata_storage_buffer: D::StorageBuffer,
+    pub(crate) indirect_draw_params_storage_buffer: D::StorageBuffer,
+    pub(crate) fills_storage_buffer: D::StorageBuffer,
+    pub(crate) tiles_storage_buffer: D::StorageBuffer,
+    pub(crate) backdrops_storage_buffer: D::StorageBuffer,
 }
 
 impl<D> BinProgramD3D11<D> where D: Device {
-    pub fn new(device: &D, resources: &dyn ResourceLoader) -> BinProgramD3D11<D> {
+    pub(crate) fn new(device: &D, resources: &dyn ResourceLoader) -> BinProgramD3D11<D> {
         let mut program = device.create_compute_program(resources, "d3d11/bin");
         let dimensions = ComputeDimensions { x: BIN_WORKGROUP_SIZE, y: 1, z: 1 };
         device.set_compute_program_local_size(&mut program, dimensions);
@@ -207,22 +209,22 @@ impl<D> BinProgramD3D11<D> where D: Device {
     }
 }
 
-pub struct DiceProgramD3D11<D> where D: Device {
-    pub program: D::Program,
-    pub transform_uniform: D::Uniform,
-    pub translation_uniform: D::Uniform,
-    pub path_count_uniform: D::Uniform,
-    pub last_batch_segment_index_uniform: D::Uniform,
-    pub max_microline_count_uniform: D::Uniform,
-    pub compute_indirect_params_storage_buffer: D::StorageBuffer,
-    pub dice_metadata_storage_buffer: D::StorageBuffer,
-    pub points_storage_buffer: D::StorageBuffer,
-    pub input_indices_storage_buffer: D::StorageBuffer,
-    pub microlines_storage_buffer: D::StorageBuffer,
+pub(crate) struct DiceProgramD3D11<D> where D: Device {
+    pub(crate) program: D::Program,
+    pub(crate) transform_uniform: D::Uniform,
+    pub(crate) translation_uniform: D::Uniform,
+    pub(crate) path_count_uniform: D::Uniform,
+    pub(crate) last_batch_segment_index_uniform: D::Uniform,
+    pub(crate) max_microline_count_uniform: D::Uniform,
+    pub(crate) compute_indirect_params_storage_buffer: D::StorageBuffer,
+    pub(crate) dice_metadata_storage_buffer: D::StorageBuffer,
+    pub(crate) points_storage_buffer: D::StorageBuffer,
+    pub(crate) input_indices_storage_buffer: D::StorageBuffer,
+    pub(crate) microlines_storage_buffer: D::StorageBuffer,
 }
 
 impl<D> DiceProgramD3D11<D> where D: Device {
-    pub fn new(device: &D, resources: &dyn ResourceLoader) -> DiceProgramD3D11<D> {
+    pub(crate) fn new(device: &D, resources: &dyn ResourceLoader) -> DiceProgramD3D11<D> {
         let mut program = device.create_compute_program(resources, "d3d11/dice");
         let dimensions = ComputeDimensions { x: DICE_WORKGROUP_SIZE, y: 1, z: 1 };
         device.set_compute_program_local_size(&mut program, dimensions);
@@ -257,16 +259,16 @@ impl<D> DiceProgramD3D11<D> where D: Device {
     }
 }
 
-pub struct BoundProgramD3D11<D> where D: Device {
-    pub program: D::Program,
-    pub path_count_uniform: D::Uniform,
-    pub tile_count_uniform: D::Uniform,
-    pub tile_path_info_storage_buffer: D::StorageBuffer,
-    pub tiles_storage_buffer: D::StorageBuffer,
+pub(crate) struct BoundProgramD3D11<D> where D: Device {
+    pub(crate) program: D::Program,
+    pub(crate) path_count_uniform: D::Uniform,
+    pub(crate) tile_count_uniform: D::Uniform,
+    pub(crate) tile_path_info_storage_buffer: D::StorageBuffer,
+    pub(crate) tiles_storage_buffer: D::StorageBuffer,
 }
 
 impl<D> BoundProgramD3D11<D> where D: Device {
-    pub fn new(device: &D, resources: &dyn ResourceLoader) -> BoundProgramD3D11<D> {
+    pub(crate) fn new(device: &D, resources: &dyn ResourceLoader) -> BoundProgramD3D11<D> {
         let mut program = device.create_compute_program(resources, "d3d11/bound");
         let dimensions = ComputeDimensions { x: BOUND_WORKGROUP_SIZE, y: 1, z: 1 };
         device.set_compute_program_local_size(&mut program, dimensions);
@@ -287,16 +289,16 @@ impl<D> BoundProgramD3D11<D> where D: Device {
     }
 }
 
-pub struct SortProgramD3D11<D> where D: Device {
-    pub program: D::Program,
-    pub tile_count_uniform: D::Uniform,
-    pub tiles_storage_buffer: D::StorageBuffer,
-    pub first_tile_map_storage_buffer: D::StorageBuffer,
-    pub z_buffer_storage_buffer: D::StorageBuffer,
+pub(crate) struct SortProgramD3D11<D> where D: Device {
+    pub(crate) program: D::Program,
+    pub(crate) tile_count_uniform: D::Uniform,
+    pub(crate) tiles_storage_buffer: D::StorageBuffer,
+    pub(crate) first_tile_map_storage_buffer: D::StorageBuffer,
+    pub(crate) z_buffer_storage_buffer: D::StorageBuffer,
 }
 
 impl<D> SortProgramD3D11<D> where D: Device {
-    pub fn new(device: &D, resources: &dyn ResourceLoader) -> SortProgramD3D11<D> {
+    pub(crate) fn new(device: &D, resources: &dyn ResourceLoader) -> SortProgramD3D11<D> {
         let mut program = device.create_compute_program(resources, "d3d11/sort");
         let dimensions = ComputeDimensions { x: SORT_WORKGROUP_SIZE, y: 1, z: 1 };
         device.set_compute_program_local_size(&mut program, dimensions);
diff --git a/renderer/src/gpu/d3d9/mod.rs b/renderer/src/gpu/d3d9/mod.rs
index 6f67f3bc..0383770c 100644
--- a/renderer/src/gpu/d3d9/mod.rs
+++ b/renderer/src/gpu/d3d9/mod.rs
@@ -8,5 +8,10 @@
 // option. This file may not be copied, modified, or distributed
 // except according to those terms.
 
+//! A hybrid CPU-GPU renderer that only relies on functionality available in Direct3D 9.
+//! 
+//! This renderer supports OpenGL at least 3.0, OpenGL ES at least 3.0, Metal of any version, and
+//! WebGL at least 2.0.
+
 pub mod renderer;
 pub mod shaders;
diff --git a/renderer/src/gpu/d3d9/renderer.rs b/renderer/src/gpu/d3d9/renderer.rs
index 5c134e97..d3295e78 100644
--- a/renderer/src/gpu/d3d9/renderer.rs
+++ b/renderer/src/gpu/d3d9/renderer.rs
@@ -8,6 +8,11 @@
 // option. This file may not be copied, modified, or distributed
 // except according to those terms.
 
+//! A hybrid CPU-GPU renderer that only relies on functionality available in Direct3D 9.
+//! 
+//! This renderer supports OpenGL at least 3.0, OpenGL ES at least 3.0, Metal of any version, and
+//! WebGL at least 2.0.
+
 use crate::gpu::blend::{BlendModeExt, ToBlendState};
 use crate::gpu::perf::TimeCategory;
 use crate::gpu::renderer::{FramebufferFlags, MASK_FRAMEBUFFER_HEIGHT, MASK_FRAMEBUFFER_WIDTH};
diff --git a/renderer/src/gpu/d3d9/shaders.rs b/renderer/src/gpu/d3d9/shaders.rs
index 82aabd7d..7e560ea0 100644
--- a/renderer/src/gpu/d3d9/shaders.rs
+++ b/renderer/src/gpu/d3d9/shaders.rs
@@ -8,6 +8,8 @@
 // option. This file may not be copied, modified, or distributed
 // except according to those terms.
 
+//! Shaders and vertex specifications for the Direct3D 9-level renderer.
+
 use crate::gpu::shaders::{TILE_INSTANCE_SIZE, TileProgramCommon};
 use pathfinder_gpu::{BufferTarget, Device, VertexAttrClass, VertexAttrDescriptor, VertexAttrType};
 use pathfinder_resources::ResourceLoader;
@@ -15,17 +17,17 @@ use pathfinder_resources::ResourceLoader;
 const FILL_INSTANCE_SIZE: usize = 12;
 const CLIP_TILE_INSTANCE_SIZE: usize = 16;
 
-pub struct FillVertexArrayD3D9<D> where D: Device {
-    pub vertex_array: D::VertexArray,
+pub(crate) struct FillVertexArrayD3D9<D> where D: Device {
+    pub(crate) vertex_array: D::VertexArray,
 }
 
 impl<D> FillVertexArrayD3D9<D> where D: Device {
-    pub fn new(device: &D,
-               fill_program: &FillProgramD3D9<D>,
-               vertex_buffer: &D::Buffer,
-               quad_vertex_positions_buffer: &D::Buffer,
-               quad_vertex_indices_buffer: &D::Buffer)
-               -> FillVertexArrayD3D9<D> {
+    pub(crate) fn new(device: &D,
+                      fill_program: &FillProgramD3D9<D>,
+                      vertex_buffer: &D::Buffer,
+                      quad_vertex_positions_buffer: &D::Buffer,
+                      quad_vertex_indices_buffer: &D::Buffer)
+                      -> FillVertexArrayD3D9<D> {
         let vertex_array = device.create_vertex_array();
 
         let tess_coord_attr = device.get_vertex_attr(&fill_program.program, "TessCoord").unwrap();
@@ -68,17 +70,17 @@ impl<D> FillVertexArrayD3D9<D> where D: Device {
     }
 }
 
-pub struct TileVertexArrayD3D9<D> where D: Device {
-    pub vertex_array: D::VertexArray,
+pub(crate) struct TileVertexArrayD3D9<D> where D: Device {
+    pub(crate) vertex_array: D::VertexArray,
 }
 
 impl<D> TileVertexArrayD3D9<D> where D: Device {
-    pub fn new(device: &D,
-               tile_program: &TileProgramD3D9<D>,
-               tile_vertex_buffer: &D::Buffer,
-               quad_vertex_positions_buffer: &D::Buffer,
-               quad_vertex_indices_buffer: &D::Buffer)
-               -> TileVertexArrayD3D9<D> {
+    pub(crate) fn new(device: &D,
+                      tile_program: &TileProgramD3D9<D>,
+                      tile_vertex_buffer: &D::Buffer,
+                      quad_vertex_positions_buffer: &D::Buffer,
+                      quad_vertex_indices_buffer: &D::Buffer)
+                      -> TileVertexArrayD3D9<D> {
         let vertex_array = device.create_vertex_array();
 
         let tile_offset_attr =
@@ -155,17 +157,17 @@ impl<D> TileVertexArrayD3D9<D> where D: Device {
     }
 }
 
-pub struct ClipTileCopyVertexArrayD3D9<D> where D: Device {
-    pub vertex_array: D::VertexArray,
+pub(crate) struct ClipTileCopyVertexArrayD3D9<D> where D: Device {
+    pub(crate) vertex_array: D::VertexArray,
 }
 
 impl<D> ClipTileCopyVertexArrayD3D9<D> where D: Device {
-    pub fn new(device: &D,
-               clip_tile_copy_program: &ClipTileCopyProgramD3D9<D>,
-               vertex_buffer: &D::Buffer,
-               quad_vertex_positions_buffer: &D::Buffer,
-               quad_vertex_indices_buffer: &D::Buffer)
-               -> ClipTileCopyVertexArrayD3D9<D> {
+    pub(crate) fn new(device: &D,
+                      clip_tile_copy_program: &ClipTileCopyProgramD3D9<D>,
+                      vertex_buffer: &D::Buffer,
+                      quad_vertex_positions_buffer: &D::Buffer,
+                      quad_vertex_indices_buffer: &D::Buffer)
+                      -> ClipTileCopyVertexArrayD3D9<D> {
         let vertex_array = device.create_vertex_array();
 
         let tile_offset_attr =
@@ -199,17 +201,17 @@ impl<D> ClipTileCopyVertexArrayD3D9<D> where D: Device {
     }
 }
 
-pub struct ClipTileCombineVertexArrayD3D9<D> where D: Device {
-    pub vertex_array: D::VertexArray,
+pub(crate) struct ClipTileCombineVertexArrayD3D9<D> where D: Device {
+    pub(crate) vertex_array: D::VertexArray,
 }
 
 impl<D> ClipTileCombineVertexArrayD3D9<D> where D: Device {
-    pub fn new(device: &D,
-               clip_tile_combine_program: &ClipTileCombineProgramD3D9<D>,
-               vertex_buffer: &D::Buffer,
-               quad_vertex_positions_buffer: &D::Buffer,
-               quad_vertex_indices_buffer: &D::Buffer)
-               -> ClipTileCombineVertexArrayD3D9<D> {
+    pub(crate) fn new(device: &D,
+                      clip_tile_combine_program: &ClipTileCombineProgramD3D9<D>,
+                      vertex_buffer: &D::Buffer,
+                      quad_vertex_positions_buffer: &D::Buffer,
+                      quad_vertex_indices_buffer: &D::Buffer)
+                      -> ClipTileCombineVertexArrayD3D9<D> {
         let vertex_array = device.create_vertex_array();
 
         let tile_offset_attr =
@@ -276,16 +278,16 @@ impl<D> ClipTileCombineVertexArrayD3D9<D> where D: Device {
     }
 }
 
-pub struct CopyTileVertexArray<D> where D: Device {
-    pub vertex_array: D::VertexArray,
+pub(crate) struct CopyTileVertexArray<D> where D: Device {
+    pub(crate) vertex_array: D::VertexArray,
 }
 
 impl<D> CopyTileVertexArray<D> where D: Device {
-    pub fn new(device: &D,
-               copy_tile_program: &CopyTileProgram<D>,
-               copy_tile_vertex_buffer: &D::Buffer,
-               quads_vertex_indices_buffer: &D::Buffer)
-               -> CopyTileVertexArray<D> {
+    pub(crate) fn new(device: &D,
+                      copy_tile_program: &CopyTileProgram<D>,
+                      copy_tile_vertex_buffer: &D::Buffer,
+                      quads_vertex_indices_buffer: &D::Buffer)
+                      -> CopyTileVertexArray<D> {
         let vertex_array = device.create_vertex_array();
 
         let tile_position_attr =
@@ -307,11 +309,11 @@ impl<D> CopyTileVertexArray<D> where D: Device {
     }
 }
 
-pub struct FillProgramD3D9<D> where D: Device {
-    pub program: D::Program,
-    pub framebuffer_size_uniform: D::Uniform,
-    pub tile_size_uniform: D::Uniform,
-    pub area_lut_texture: D::TextureParameter,
+pub(crate) struct FillProgramD3D9<D> where D: Device {
+    pub(crate) program: D::Program,
+    pub(crate) framebuffer_size_uniform: D::Uniform,
+    pub(crate) tile_size_uniform: D::Uniform,
+    pub(crate) area_lut_texture: D::TextureParameter,
 }
 
 impl<D> FillProgramD3D9<D> where D: Device {
@@ -329,10 +331,10 @@ impl<D> FillProgramD3D9<D> where D: Device {
     }
 }
 
-pub struct TileProgramD3D9<D> where D: Device {
-    pub common: TileProgramCommon<D>,
-    pub dest_texture: D::TextureParameter,
-    pub transform_uniform: D::Uniform,
+pub(crate) struct TileProgramD3D9<D> where D: Device {
+    pub(crate) common: TileProgramCommon<D>,
+    pub(crate) dest_texture: D::TextureParameter,
+    pub(crate) transform_uniform: D::Uniform,
 }
 
 impl<D> TileProgramD3D9<D> where D: Device {
@@ -345,14 +347,15 @@ impl<D> TileProgramD3D9<D> where D: Device {
     }
 }
 
-pub struct ClipTileCombineProgramD3D9<D> where D: Device {
-    pub program: D::Program,
-    pub src_texture: D::TextureParameter,
-    pub framebuffer_size_uniform: D::Uniform,
+pub(crate) struct ClipTileCombineProgramD3D9<D> where D: Device {
+    pub(crate) program: D::Program,
+    pub(crate) src_texture: D::TextureParameter,
+    pub(crate) framebuffer_size_uniform: D::Uniform,
 }
 
 impl<D> ClipTileCombineProgramD3D9<D> where D: Device {
-    pub fn new(device: &D, resources: &dyn ResourceLoader) -> ClipTileCombineProgramD3D9<D> {
+    pub(crate) fn new(device: &D, resources: &dyn ResourceLoader)
+                      -> ClipTileCombineProgramD3D9<D> {
         let program = device.create_raster_program(resources, "d3d9/tile_clip_combine");
         let src_texture = device.get_texture_parameter(&program, "Src");
         let framebuffer_size_uniform = device.get_uniform(&program, "FramebufferSize");
@@ -360,14 +363,14 @@ impl<D> ClipTileCombineProgramD3D9<D> where D: Device {
     }
 }
 
-pub struct ClipTileCopyProgramD3D9<D> where D: Device {
-    pub program: D::Program,
-    pub src_texture: D::TextureParameter,
-    pub framebuffer_size_uniform: D::Uniform,
+pub(crate) struct ClipTileCopyProgramD3D9<D> where D: Device {
+    pub(crate) program: D::Program,
+    pub(crate) src_texture: D::TextureParameter,
+    pub(crate) framebuffer_size_uniform: D::Uniform,
 }
 
 impl<D> ClipTileCopyProgramD3D9<D> where D: Device {
-    pub fn new(device: &D, resources: &dyn ResourceLoader) -> ClipTileCopyProgramD3D9<D> {
+    pub(crate) fn new(device: &D, resources: &dyn ResourceLoader) -> ClipTileCopyProgramD3D9<D> {
         let program = device.create_raster_program(resources, "d3d9/tile_clip_copy");
         let src_texture = device.get_texture_parameter(&program, "Src");
         let framebuffer_size_uniform = device.get_uniform(&program, "FramebufferSize");
@@ -375,16 +378,16 @@ impl<D> ClipTileCopyProgramD3D9<D> where D: Device {
     }
 }
 
-pub struct CopyTileProgram<D> where D: Device {
-    pub program: D::Program,
-    pub transform_uniform: D::Uniform,
-    pub tile_size_uniform: D::Uniform,
-    pub framebuffer_size_uniform: D::Uniform,
-    pub src_texture: D::TextureParameter,
+pub(crate) struct CopyTileProgram<D> where D: Device {
+    pub(crate) program: D::Program,
+    pub(crate) transform_uniform: D::Uniform,
+    pub(crate) tile_size_uniform: D::Uniform,
+    pub(crate) framebuffer_size_uniform: D::Uniform,
+    pub(crate) src_texture: D::TextureParameter,
 }
 
 impl<D> CopyTileProgram<D> where D: Device {
-    pub fn new(device: &D, resources: &dyn ResourceLoader) -> CopyTileProgram<D> {
+    pub(crate) fn new(device: &D, resources: &dyn ResourceLoader) -> CopyTileProgram<D> {
         let program = device.create_raster_program(resources, "d3d9/tile_copy");
         let transform_uniform = device.get_uniform(&program, "Transform");
         let tile_size_uniform = device.get_uniform(&program, "TileSize");
@@ -400,36 +403,16 @@ impl<D> CopyTileProgram<D> where D: Device {
     }
 }
 
-pub struct D3D9Programs<D> where D: Device {
-    pub fill_program: FillProgramD3D9<D>,
-    pub tile_program: TileProgramD3D9<D>,
-    pub tile_clip_combine_program: ClipTileCombineProgramD3D9<D>,
-    pub tile_clip_copy_program: ClipTileCopyProgramD3D9<D>,
-    pub tile_copy_program: CopyTileProgram<D>,
-}
-
-impl<D> D3D9Programs<D> where D: Device {
-    pub fn new(device: &D, resources: &dyn ResourceLoader) -> D3D9Programs<D> {
-        D3D9Programs {
-            fill_program: FillProgramD3D9::new(device, resources),
-            tile_program: TileProgramD3D9::new(device, resources),
-            tile_clip_combine_program: ClipTileCombineProgramD3D9::new(device, resources),
-            tile_clip_copy_program: ClipTileCopyProgramD3D9::new(device, resources),
-            tile_copy_program: CopyTileProgram::new(device, resources),
-        }
-    }
-}
-
-pub struct ProgramsD3D9<D> where D: Device {
-    pub fill_program: FillProgramD3D9<D>,
-    pub tile_program: TileProgramD3D9<D>,
-    pub tile_clip_copy_program: ClipTileCopyProgramD3D9<D>,
-    pub tile_clip_combine_program: ClipTileCombineProgramD3D9<D>,
-    pub tile_copy_program: CopyTileProgram<D>,
+pub(crate) struct ProgramsD3D9<D> where D: Device {
+    pub(crate) fill_program: FillProgramD3D9<D>,
+    pub(crate) tile_program: TileProgramD3D9<D>,
+    pub(crate) tile_clip_copy_program: ClipTileCopyProgramD3D9<D>,
+    pub(crate) tile_clip_combine_program: ClipTileCombineProgramD3D9<D>,
+    pub(crate) tile_copy_program: CopyTileProgram<D>,
 }
 
 impl<D> ProgramsD3D9<D> where D: Device {
-    pub fn new(device: &D, resources: &dyn ResourceLoader) -> ProgramsD3D9<D> {
+    pub(crate) fn new(device: &D, resources: &dyn ResourceLoader) -> ProgramsD3D9<D> {
         ProgramsD3D9 {
             fill_program: FillProgramD3D9::new(device, resources),
             tile_program: TileProgramD3D9::new(device, resources),
diff --git a/renderer/src/gpu/debug.rs b/renderer/src/gpu/debug.rs
index ea7e9e31..665e8f21 100644
--- a/renderer/src/gpu/debug.rs
+++ b/renderer/src/gpu/debug.rs
@@ -39,8 +39,13 @@ const PERFORMANCE_WINDOW_HEIGHT_D3D11: i32 = LINE_HEIGHT * 10 + PADDING + 2;
 const INFO_WINDOW_WIDTH: i32 = 425;
 const INFO_WINDOW_HEIGHT: i32 = LINE_HEIGHT * 2 + PADDING + 2;
 
+/// Manages the debug UI.
 pub struct DebugUIPresenter<D> where D: Device {
+    /// The general UI presenter object.
+    /// 
+    /// You can use this to draw your own application-specific debug widgets.
     pub ui_presenter: UIPresenter<D>,
+
     cpu_samples: SampleBuffer<RenderStats>,
     gpu_samples: SampleBuffer<RenderTime>,
     backend_name: &'static str,
@@ -49,11 +54,11 @@ pub struct DebugUIPresenter<D> where D: Device {
 }
 
 impl<D> DebugUIPresenter<D> where D: Device {
-    pub fn new(device: &D,
-               resources: &dyn ResourceLoader,
-               framebuffer_size: Vector2I,
-               renderer_level: RendererLevel)
-               -> DebugUIPresenter<D> {
+    pub(crate) fn new(device: &D,
+                      resources: &dyn ResourceLoader,
+                      framebuffer_size: Vector2I,
+                      renderer_level: RendererLevel)
+                      -> DebugUIPresenter<D> {
         let ui_presenter = UIPresenter::new(device, resources, framebuffer_size);
         DebugUIPresenter {
             ui_presenter,
@@ -65,19 +70,19 @@ impl<D> DebugUIPresenter<D> where D: Device {
         }
     }
 
-    pub fn add_sample(&mut self, stats: RenderStats, rendering_time: RenderTime) {
+    pub(crate) fn add_sample(&mut self, stats: RenderStats, rendering_time: RenderTime) {
         self.cpu_samples.push(stats);
         self.gpu_samples.push(rendering_time);
     }
 
-    pub fn draw(&self, device: &D, allocator: &mut GPUMemoryAllocator<D>) {
+    pub(crate) fn draw(&self, device: &D, allocator: &mut GPUMemoryAllocator<D>) {
         self.draw_stats_window(device, allocator);
         self.draw_performance_window(device, allocator);
         self.draw_info_window(device, allocator);
     }
 
     #[inline]
-    pub fn set_framebuffer_size(&mut self, new_framebuffer_size: Vector2I) {
+    pub(crate) fn set_framebuffer_size(&mut self, new_framebuffer_size: Vector2I) {
         self.ui_presenter.set_framebuffer_size(new_framebuffer_size)
     }
 
diff --git a/renderer/src/gpu/options.rs b/renderer/src/gpu/options.rs
index e5b6aa97..1a8f6051 100644
--- a/renderer/src/gpu/options.rs
+++ b/renderer/src/gpu/options.rs
@@ -8,6 +8,8 @@
 // option. This file may not be copied, modified, or distributed
 // except according to those terms.
 
+//! Various options that control how the renderer behaves.
+
 use pathfinder_color::ColorF;
 use pathfinder_geometry::rect::RectI;
 use pathfinder_geometry::vector::Vector2I;
@@ -57,6 +59,7 @@ impl<D> Default for RendererOptions<D> where D: Device {
 }
 
 impl RendererLevel {
+    /// Returns a suitable renderer level for the given device.
     pub fn default_for_device<D>(device: &D) -> RendererLevel where D: Device {
         match device.feature_level() {
             FeatureLevel::D3D10 => RendererLevel::D3D9,
@@ -65,12 +68,17 @@ impl RendererLevel {
     }
 }
 
+/// Where the rendered content should go.
 #[derive(Clone)]
 pub enum DestFramebuffer<D> where D: Device {
+    /// The rendered content should go to the default framebuffer (e.g. the window in OpenGL).
     Default {
+        /// The rectangle within the window to draw in, in device pixels.
         viewport: RectI,
+        /// The total size of the window in device pixels.
         window_size: Vector2I,
     },
+    /// The rendered content should go to a non-default framebuffer (off-screen, typically).
     Other(D::Framebuffer),
 }
 
@@ -81,16 +89,18 @@ impl<D> Default for DestFramebuffer<D> where D: Device {
     }
 }
 
-impl<D> DestFramebuffer<D>
-where
-    D: Device,
-{
+impl<D> DestFramebuffer<D> where D: Device {
+    /// Returns a `DestFramebuffer` object that renders to the entire contents of the default
+    /// framebuffer.
+    /// 
+    /// The `window_size` parameter specifies the size of the window in device pixels.
     #[inline]
     pub fn full_window(window_size: Vector2I) -> DestFramebuffer<D> {
         let viewport = RectI::new(Vector2I::default(), window_size);
         DestFramebuffer::Default { viewport, window_size }
     }
 
+    /// Returns the size of the destination buffer, in device pixels.
     #[inline]
     pub fn window_size(&self, device: &D) -> Vector2I {
         match *self {
diff --git a/renderer/src/gpu/perf.rs b/renderer/src/gpu/perf.rs
index e891ffdb..81db829e 100644
--- a/renderer/src/gpu/perf.rs
+++ b/renderer/src/gpu/perf.rs
@@ -16,15 +16,30 @@ use std::mem;
 use std::ops::{Add, Div};
 use std::time::Duration;
 
+/// Various GPU-side statistics about rendering.
 #[derive(Clone, Copy, Debug, Default)]
 pub struct RenderStats {
+    /// The total number of path objects in the scene.
     pub path_count: usize,
+    /// The number of fill operations it took to render the scene.
+    /// 
+    /// A fill operation is a single edge in a 16x16 device pixel tile.
     pub fill_count: usize,
+    /// The total number of 16x16 device pixel tile masks generated.
     pub alpha_tile_count: usize,
+    /// The total number of 16x16 tiles needed to render the scene, including both alpha tiles and
+    /// solid-color tiles.
     pub total_tile_count: usize,
+    /// The amount of CPU time it took to build the scene.
     pub cpu_build_time: Duration,
+    /// The number of GPU API draw calls it took to render the scene.
     pub drawcall_count: u32,
+    /// The number of bytes of VRAM Pathfinder has allocated.
+    /// 
+    /// This may be higher than `gpu_bytes_committed` because Pathfinder caches some data for
+    /// faster reuse.
     pub gpu_bytes_allocated: u64,
+    /// The number of bytes of VRAM Pathfinder actually used for the frame.
     pub gpu_bytes_committed: u64,
 }
 
@@ -203,16 +218,28 @@ fn total_time_of_timer_futures<D>(futures: &[TimerFuture<D>]) -> Option<Duration
     Some(total)
 }
 
+/// The amount of GPU time it took to render the scene, broken up into stages.
 #[derive(Clone, Copy, Debug)]
 pub struct RenderTime {
+    /// How much GPU time it took to divide all edges in the scene into small lines.
+    /// 
+    /// This will be zero in the D3D9-level backend, since in that backend dicing is done on CPU.
     pub dice_time: Duration,
+    /// How much GPU time it took to assign those diced microlines to tiles.
+    /// 
+    /// This will be zero in the D3D9-level backend, since in that backend binning is done on CPU.
     pub bin_time: Duration,
+    /// How much GPU time it took to draw fills (i.e. render edges) to masks.
     pub fill_time: Duration,
+    /// How much GPU time it took to draw the contents of the tiles to the output.
     pub composite_time: Duration,
+    /// How much GPU time it took to execute miscellaneous tasks other than dicing, binning,
+    /// filling, and compositing.
     pub other_time: Duration,
 }
 
 impl RenderTime {
+    /// The total GPU time it took to render the scene.
     #[inline]
     pub fn total_time(&self) -> Duration {
         self.dice_time + self.bin_time + self.fill_time + self.composite_time + self.other_time
diff --git a/renderer/src/gpu/renderer.rs b/renderer/src/gpu/renderer.rs
index d3ed432b..ce434dc7 100644
--- a/renderer/src/gpu/renderer.rs
+++ b/renderer/src/gpu/renderer.rs
@@ -144,6 +144,23 @@ pub(crate) struct MaskStorage {
 }
 
 impl<D> Renderer<D> where D: Device {
+    /// Creates a new renderer ready to render Pathfinder content.
+    /// 
+    /// Arguments:
+    /// 
+    /// * `device`: The GPU device to render with. This effectively specifies the system GPU API
+    ///   Pathfinder will use (OpenGL, Metal, etc.)
+    /// 
+    /// * `resources`: Where Pathfinder should find shaders, lookup tables, and other data.
+    ///   This is typically either an `EmbeddedResourceLoader` to use resources included in the
+    ///   Pathfinder library or (less commonly) a `FilesystemResourceLoader` to use resources
+    ///   stored in a directory on disk.
+    /// 
+    /// * `mode`: Renderer options that can't be changed after the renderer is created. Most
+    ///   notably, this specifies the API level (D3D9 or D3D11).
+    /// 
+    /// * `options`: Renderer options that can be changed after the renderer is created. Most
+    ///   importantly, this specifies where the output should go (to a window or off-screen).
     pub fn new(device: D,
                resources: &dyn ResourceLoader,
                mode: RendererMode,
@@ -290,10 +307,14 @@ impl<D> Renderer<D> where D: Device {
         }
     }
 
+    /// Destroys this renderer and returns the embedded GPU device.
     pub fn destroy(self) -> D {
         self.core.device
     }
 
+    /// Performs work necessary to begin rendering a scene.
+    /// 
+    /// This must be called before `render_command()`.
     pub fn begin_scene(&mut self) {
         self.core.framebuffer_flags = FramebufferFlags::empty();
 
@@ -304,6 +325,11 @@ impl<D> Renderer<D> where D: Device {
         self.core.alpha_tile_count = 0;
     }
 
+    /// Issues a rendering command to the renderer.
+    /// 
+    /// These commands are generated from methods like `Scene::build()`.
+    /// 
+    /// `begin_scene()` must have been called first.
     pub fn render_command(&mut self, command: &RenderCommand) {
         debug!("render command: {:?}", command);
         match *command {
@@ -352,6 +378,13 @@ impl<D> Renderer<D> where D: Device {
         }
     }
 
+    /// Finishes rendering a scene.
+    /// 
+    /// `begin_scene()` and all `render_command()` calls must have been issued before calling this
+    /// method.
+    /// 
+    /// Note that, after calling this method, you might need to flush the output to the screen via
+    /// `swap_buffers()`, `present()`, or a similar method that your windowing library offers.
     pub fn end_scene(&mut self) {
         self.clear_dest_framebuffer_if_necessary();
         self.blit_intermediate_dest_framebuffer_if_necessary();
@@ -451,34 +484,69 @@ impl<D> Renderer<D> where D: Device {
         self.last_rendering_time = None;
     }
 
+    /// Returns GPU timing information for the last frame, if present.
     pub fn last_rendering_time(&self) -> Option<RenderTime> {
         self.last_rendering_time
     }
 
+    /// Returns a reference to the GPU device.
+    /// 
+    /// This can be useful to issue GPU commands manually via the low-level `pathfinder_gpu`
+    /// abstraction. (Of course, you can also use your platform API such as OpenGL directly
+    /// alongside Pathfinder.)
     #[inline]
     pub fn device(&self) -> &D {
         &self.core.device
     }
 
+    /// Returns a mutable reference to the GPU device.
+    /// 
+    /// This can be useful to issue GPU commands manually via the low-level `pathfinder_gpu`
+    /// abstraction. (Of course, you can also use your platform API such as OpenGL directly
+    /// alongside Pathfinder.)
     #[inline]
     pub fn device_mut(&mut self) -> &mut D {
         &mut self.core.device
     }
 
+    /// Returns the `RendererMode` this renderer was created with.
     #[inline]
     pub fn mode(&self) -> &RendererMode {
         &self.core.mode
     }
 
+    /// Returns the current rendering options.
     #[inline]
     pub fn options(&self) -> &RendererOptions<D> {
         &self.core.options
     }
 
+    /// Returns a mutable reference to the current rendering options, allowing them to be changed.
+    /// 
+    /// Among other things, you can use this function to change the destination of rendering
+    /// output without having to recreate the renderer.
+    /// 
+    /// After changing the destination framebuffer size, you must call
+    /// `dest_framebuffer_size_changed()`.
     pub fn options_mut(&mut self) -> &mut RendererOptions<D> {
         &mut self.core.options
     }
 
+    /// Notifies Pathfinder that the size of the output framebuffer has changed.
+    /// 
+    /// You must call this function after changing the `dest_framebuffer` member of
+    /// `RendererOptions` to a target with a different size.
+    #[inline]
+    pub fn dest_framebuffer_size_changed(&mut self) {
+        let new_framebuffer_size = self.core.main_viewport().size();
+        if let Some(ref mut debug_ui_presenter) = self.debug_ui_presenter {
+            debug_ui_presenter.ui_presenter.set_framebuffer_size(new_framebuffer_size);
+        }
+    }
+
+    /// Returns a mutable reference to the debug UI.
+    /// 
+    /// You can use this function to draw custom debug widgets on screen, as the demo does.
     #[inline]
     pub fn debug_ui_presenter_mut(&mut self) -> DebugUIPresenterInfo<D> {
         DebugUIPresenterInfo {
@@ -488,34 +556,38 @@ impl<D> Renderer<D> where D: Device {
         }
     }
 
+    /// Turns off Pathfinder's use of the depth buffer.
     #[inline]
-    pub fn dest_framebuffer_size_changed(&mut self) {
-        let new_framebuffer_size = self.core.main_viewport().size();
-        if let Some(ref mut debug_ui_presenter) = self.debug_ui_presenter {
-            debug_ui_presenter.ui_presenter.set_framebuffer_size(new_framebuffer_size);
-        }
-    }
-
-    #[inline]
+    #[deprecated]
     pub fn disable_depth(&mut self) {
         self.core.renderer_flags.remove(RendererFlags::USE_DEPTH);
     }
 
+    /// Turns on Pathfinder's use of the depth buffer.
     #[inline]
+    #[deprecated]
     pub fn enable_depth(&mut self) {
         self.core.renderer_flags.insert(RendererFlags::USE_DEPTH);
     }
 
+    /// Returns various GPU-side statistics about rendering, averaged over the last few frames.
     #[inline]
     pub fn stats(&self) -> &RenderStats {
         &self.core.stats
     }
 
+    /// Returns a GPU-side vertex buffer containing 2D vertices of a unit square.
+    /// 
+    /// This can be handy for custom rendering.
     #[inline]
     pub fn quad_vertex_positions_buffer(&self) -> &D::Buffer {
         self.core.allocator.get_general_buffer(self.core.quad_vertex_positions_buffer_id)
     }
 
+    /// Returns a GPU-side 32-bit unsigned index buffer of triangles necessary to render a quad
+    /// with the buffer returned by `quad_vertex_positions_buffer()`.
+    /// 
+    /// This can be handy for custom rendering.
     #[inline]
     pub fn quad_vertex_indices_buffer(&self) -> &D::Buffer {
         self.core.allocator.get_index_buffer(self.core.quad_vertex_indices_buffer_id)
@@ -684,6 +756,10 @@ impl<D> Renderer<D> where D: Device {
         self.core.stats.drawcall_count += 1;
     }
 
+
+    /// Draws a texture that was originally drawn with `old_transform` with `new_transform` by
+    /// transforming in screen space.
+    #[deprecated]
     pub fn reproject_texture(&mut self,
                              texture: &D::Texture,
                              old_transform: &Transform4F,
@@ -814,21 +890,19 @@ impl<D> Renderer<D> where D: Device {
         self.core.stats.drawcall_count += 1;
     }
 
+    /// Returns the output viewport in the destination framebuffer, as specified in the render
+    /// options.
     #[inline]
     pub fn draw_viewport(&self) -> RectI {
         self.core.draw_viewport()
     }
 
+    /// Returns the destination framebuffer, wrapped in a render target.
     #[inline]
     pub fn draw_render_target(&self) -> RenderTarget<D> {
         self.core.draw_render_target()
     }
 
-    #[inline]
-    pub fn render_stats(&self) -> &RenderStats {
-        &self.core.stats
-    }
-
     fn compute_filter_params(&self,
                              filter: &Filter,
                              blend_mode: BlendMode,
@@ -1085,14 +1159,14 @@ impl<D> RendererCore<D> where D: Device {
         }
     }
 
-    pub fn draw_viewport(&self) -> RectI {
+    pub(crate) fn draw_viewport(&self) -> RectI {
         match self.render_target_stack.last() {
             Some(&render_target_id) => self.render_target_location(render_target_id).rect,
             None => self.main_viewport(),
         }
     }
 
-    pub fn draw_render_target(&self) -> RenderTarget<D> {
+    pub(crate) fn draw_render_target(&self) -> RenderTarget<D> {
         match self.render_target_stack.last() {
             Some(&render_target_id) => {
                 let texture_page_id = self.render_target_location(render_target_id).page;
@@ -1248,8 +1322,14 @@ pub(crate) struct PatternTexturePage {
     pub(crate) must_preserve_contents: bool,
 }
 
+/// A mutable reference to the debug UI presenter.
+/// 
+/// You can use this structure to draw custom debug widgets on screen, as the demo does.
 pub struct DebugUIPresenterInfo<'a, D> where D: Device {
+    /// The GPU device.
     pub device: &'a mut D,
+    /// The GPU memory allocator.
     pub allocator: &'a mut GPUMemoryAllocator<D>,
+    /// The debug UI presenter, useful for drawing custom debug widgets on screen.
     pub debug_ui_presenter: &'a mut DebugUIPresenter<D>,
 }
diff --git a/renderer/src/gpu/shaders.rs b/renderer/src/gpu/shaders.rs
index 84bd0482..ee85c000 100644
--- a/renderer/src/gpu/shaders.rs
+++ b/renderer/src/gpu/shaders.rs
@@ -15,16 +15,16 @@ use pathfinder_resources::ResourceLoader;
 // TODO(pcwalton): Replace with `mem::size_of` calls?
 pub(crate) const TILE_INSTANCE_SIZE: usize = 16;
 
-pub struct BlitVertexArray<D> where D: Device {
-    pub vertex_array: D::VertexArray,
+pub(crate) struct BlitVertexArray<D> where D: Device {
+    pub(crate) vertex_array: D::VertexArray,
 }
 
 impl<D> BlitVertexArray<D> where D: Device {
-    pub fn new(device: &D,
-               blit_program: &BlitProgram<D>,
-               quad_vertex_positions_buffer: &D::Buffer,
-               quad_vertex_indices_buffer: &D::Buffer)
-               -> BlitVertexArray<D> {
+    pub(crate) fn new(device: &D,
+                      blit_program: &BlitProgram<D>,
+                      quad_vertex_positions_buffer: &D::Buffer,
+                      quad_vertex_indices_buffer: &D::Buffer)
+                      -> BlitVertexArray<D> {
         let vertex_array = device.create_vertex_array();
         let position_attr = device.get_vertex_attr(&blit_program.program, "Position").unwrap();
 
@@ -44,12 +44,12 @@ impl<D> BlitVertexArray<D> where D: Device {
     }
 }
 
-pub struct VertexArraysCore<D> where D: Device {
-    pub blit_vertex_array: BlitVertexArray<D>,
+pub(crate) struct VertexArraysCore<D> where D: Device {
+    pub(crate) blit_vertex_array: BlitVertexArray<D>,
 }
 
 impl<D> VertexArraysCore<D> where D: Device {
-    pub fn new(device: &D,
+    pub(crate) fn new(device: &D,
                programs: &ProgramsCore<D>,
                quad_vertex_positions_buffer: &D::Buffer,
                quad_vertex_indices_buffer: &D::Buffer)
@@ -63,16 +63,16 @@ impl<D> VertexArraysCore<D> where D: Device {
     }
 }
 
-pub struct ClearVertexArray<D> where D: Device {
-    pub vertex_array: D::VertexArray,
+pub(crate) struct ClearVertexArray<D> where D: Device {
+    pub(crate) vertex_array: D::VertexArray,
 }
 
 impl<D> ClearVertexArray<D> where D: Device {
-    pub fn new(device: &D,
-               clear_program: &ClearProgram<D>,
-               quad_vertex_positions_buffer: &D::Buffer,
-               quad_vertex_indices_buffer: &D::Buffer)
-               -> ClearVertexArray<D> {
+    pub(crate) fn new(device: &D,
+                      clear_program: &ClearProgram<D>,
+                      quad_vertex_positions_buffer: &D::Buffer,
+                      quad_vertex_indices_buffer: &D::Buffer)
+                      -> ClearVertexArray<D> {
         let vertex_array = device.create_vertex_array();
         let position_attr = device.get_vertex_attr(&clear_program.program, "Position").unwrap();
 
@@ -92,15 +92,15 @@ impl<D> ClearVertexArray<D> where D: Device {
     }
 }
 
-pub struct BlitProgram<D> where D: Device {
-    pub program: D::Program,
-    pub dest_rect_uniform: D::Uniform,
-    pub framebuffer_size_uniform: D::Uniform,
-    pub src_texture: D::TextureParameter,
+pub(crate) struct BlitProgram<D> where D: Device {
+    pub(crate) program: D::Program,
+    pub(crate) dest_rect_uniform: D::Uniform,
+    pub(crate) framebuffer_size_uniform: D::Uniform,
+    pub(crate) src_texture: D::TextureParameter,
 }
 
 impl<D> BlitProgram<D> where D: Device {
-    pub fn new(device: &D, resources: &dyn ResourceLoader) -> BlitProgram<D> {
+    pub(crate) fn new(device: &D, resources: &dyn ResourceLoader) -> BlitProgram<D> {
         let program = device.create_raster_program(resources, "blit");
         let dest_rect_uniform = device.get_uniform(&program, "DestRect");
         let framebuffer_size_uniform = device.get_uniform(&program, "FramebufferSize");
@@ -109,27 +109,27 @@ impl<D> BlitProgram<D> where D: Device {
     }
 }
 
-pub struct ProgramsCore<D> where D: Device {
-    pub blit_program: BlitProgram<D>,
+pub(crate) struct ProgramsCore<D> where D: Device {
+    pub(crate) blit_program: BlitProgram<D>,
 }
 
 impl<D> ProgramsCore<D> where D: Device {
-    pub fn new(device: &D, resources: &dyn ResourceLoader) -> ProgramsCore<D> {
+    pub(crate) fn new(device: &D, resources: &dyn ResourceLoader) -> ProgramsCore<D> {
         ProgramsCore {
             blit_program: BlitProgram::new(device, resources),
         }
     }
 }
 
-pub struct ClearProgram<D> where D: Device {
-    pub program: D::Program,
-    pub rect_uniform: D::Uniform,
-    pub framebuffer_size_uniform: D::Uniform,
-    pub color_uniform: D::Uniform,
+pub(crate) struct ClearProgram<D> where D: Device {
+    pub(crate) program: D::Program,
+    pub(crate) rect_uniform: D::Uniform,
+    pub(crate) framebuffer_size_uniform: D::Uniform,
+    pub(crate) color_uniform: D::Uniform,
 }
 
 impl<D> ClearProgram<D> where D: Device {
-    pub fn new(device: &D, resources: &dyn ResourceLoader) -> ClearProgram<D> {
+    pub(crate) fn new(device: &D, resources: &dyn ResourceLoader) -> ClearProgram<D> {
         let program = device.create_raster_program(resources, "clear");
         let rect_uniform = device.get_uniform(&program, "Rect");
         let framebuffer_size_uniform = device.get_uniform(&program, "FramebufferSize");
@@ -138,20 +138,20 @@ impl<D> ClearProgram<D> where D: Device {
     }
 }
 
-pub struct TileProgramCommon<D> where D: Device {
-    pub program: D::Program,
-    pub tile_size_uniform: D::Uniform,
-    pub texture_metadata_texture: D::TextureParameter,
-    pub texture_metadata_size_uniform: D::Uniform,
-    pub z_buffer_texture: D::TextureParameter,
-    pub z_buffer_texture_size_uniform: D::Uniform,
-    pub color_texture_0: D::TextureParameter,
-    pub color_texture_size_0_uniform: D::Uniform,
-    pub color_texture_1: D::TextureParameter,
-    pub mask_texture_0: D::TextureParameter,
-    pub mask_texture_size_0_uniform: D::Uniform,
-    pub gamma_lut_texture: D::TextureParameter,
-    pub framebuffer_size_uniform: D::Uniform,
+pub(crate) struct TileProgramCommon<D> where D: Device {
+    pub(crate) program: D::Program,
+    pub(crate) tile_size_uniform: D::Uniform,
+    pub(crate) texture_metadata_texture: D::TextureParameter,
+    pub(crate) texture_metadata_size_uniform: D::Uniform,
+    pub(crate) z_buffer_texture: D::TextureParameter,
+    pub(crate) z_buffer_texture_size_uniform: D::Uniform,
+    pub(crate) color_texture_0: D::TextureParameter,
+    pub(crate) color_texture_size_0_uniform: D::Uniform,
+    pub(crate) color_texture_1: D::TextureParameter,
+    pub(crate) mask_texture_0: D::TextureParameter,
+    pub(crate) mask_texture_size_0_uniform: D::Uniform,
+    pub(crate) gamma_lut_texture: D::TextureParameter,
+    pub(crate) framebuffer_size_uniform: D::Uniform,
 }
 
 impl<D> TileProgramCommon<D> where D: Device {
@@ -187,25 +187,25 @@ impl<D> TileProgramCommon<D> where D: Device {
     }
 }
 
-pub struct StencilProgram<D> where D: Device {
-    pub program: D::Program,
+pub(crate) struct StencilProgram<D> where D: Device {
+    pub(crate) program: D::Program,
 }
 
 impl<D> StencilProgram<D> where D: Device {
-    pub fn new(device: &D, resources: &dyn ResourceLoader) -> StencilProgram<D> {
+    pub(crate) fn new(device: &D, resources: &dyn ResourceLoader) -> StencilProgram<D> {
         let program = device.create_raster_program(resources, "stencil");
         StencilProgram { program }
     }
 }
 
-pub struct StencilVertexArray<D> where D: Device {
-    pub vertex_array: D::VertexArray,
-    pub vertex_buffer: D::Buffer,
-    pub index_buffer: D::Buffer,
+pub(crate) struct StencilVertexArray<D> where D: Device {
+    pub(crate) vertex_array: D::VertexArray,
+    pub(crate) vertex_buffer: D::Buffer,
+    pub(crate) index_buffer: D::Buffer,
 }
 
 impl<D> StencilVertexArray<D> where D: Device {
-    pub fn new(device: &D, stencil_program: &StencilProgram<D>) -> StencilVertexArray<D> {
+    pub(crate) fn new(device: &D, stencil_program: &StencilProgram<D>) -> StencilVertexArray<D> {
         let vertex_array = device.create_vertex_array();
         let vertex_buffer = device.create_buffer(BufferUploadMode::Static);
         let index_buffer = device.create_buffer(BufferUploadMode::Static);
@@ -228,15 +228,15 @@ impl<D> StencilVertexArray<D> where D: Device {
     }
 }
 
-pub struct ReprojectionProgram<D> where D: Device {
-    pub program: D::Program,
-    pub old_transform_uniform: D::Uniform,
-    pub new_transform_uniform: D::Uniform,
-    pub texture: D::TextureParameter,
+pub(crate) struct ReprojectionProgram<D> where D: Device {
+    pub(crate) program: D::Program,
+    pub(crate) old_transform_uniform: D::Uniform,
+    pub(crate) new_transform_uniform: D::Uniform,
+    pub(crate) texture: D::TextureParameter,
 }
 
 impl<D> ReprojectionProgram<D> where D: Device {
-    pub fn new(device: &D, resources: &dyn ResourceLoader) -> ReprojectionProgram<D> {
+    pub(crate) fn new(device: &D, resources: &dyn ResourceLoader) -> ReprojectionProgram<D> {
         let program = device.create_raster_program(resources, "reproject");
         let old_transform_uniform = device.get_uniform(&program, "OldTransform");
         let new_transform_uniform = device.get_uniform(&program, "NewTransform");
@@ -245,17 +245,16 @@ impl<D> ReprojectionProgram<D> where D: Device {
     }
 }
 
-pub struct ReprojectionVertexArray<D> where D: Device {
-    pub vertex_array: D::VertexArray,
+pub(crate) struct ReprojectionVertexArray<D> where D: Device {
+    pub(crate) vertex_array: D::VertexArray,
 }
 
 impl<D> ReprojectionVertexArray<D> where D: Device {
-    pub fn new(
-        device: &D,
-        reprojection_program: &ReprojectionProgram<D>,
-        quad_vertex_positions_buffer: &D::Buffer,
-        quad_vertex_indices_buffer: &D::Buffer,
-    ) -> ReprojectionVertexArray<D> {
+    pub(crate) fn new(device: &D,
+                      reprojection_program: &ReprojectionProgram<D>,
+                      quad_vertex_positions_buffer: &D::Buffer,
+                      quad_vertex_indices_buffer: &D::Buffer)
+                      -> ReprojectionVertexArray<D> {
         let vertex_array = device.create_vertex_array();
         let position_attr = device.get_vertex_attr(&reprojection_program.program, "Position")
                                   .unwrap();
diff --git a/renderer/src/lib.rs b/renderer/src/lib.rs
index ac43b530..d6ad7b8e 100644
--- a/renderer/src/lib.rs
+++ b/renderer/src/lib.rs
@@ -17,13 +17,13 @@ extern crate log;
 
 pub mod concurrent;
 pub mod gpu;
-pub mod gpu_data;
 pub mod options;
 pub mod paint;
 pub mod scene;
 
 mod allocator;
 mod builder;
+mod gpu_data;
 mod tile_map;
 mod tiler;
 mod tiles;
diff --git a/renderer/src/options.rs b/renderer/src/options.rs
index 646bdda4..cc739cad 100644
--- a/renderer/src/options.rs
+++ b/renderer/src/options.rs
@@ -18,20 +18,30 @@ use pathfinder_geometry::transform3d::Perspective;
 use pathfinder_geometry::vector::{Vector2F, Vector4F};
 use pathfinder_content::clip::PolygonClipper3D;
 
+/// A sink for the render commands that scenes build.
+/// 
+/// In single-threaded operation, this object typically buffers commands into an array and then,
+/// once scene building is complete, commands are all sent to the output at once. In multithreaded
+/// operation, on the other hand, commands are sent to the renderer on the fly as they're built.
+/// The latter is generally preferable for performance, because it allows the CPU and GPU to run
+/// concurrently. However, it requires a multithreaded environment, which may not always be
+/// available.
 pub struct RenderCommandListener<'a> {
     send_fn: RenderCommandSendFunction<'a>,
 }
 
+/// The callback function that receives the render commands from the scene builder.
 pub type RenderCommandSendFunction<'a> = Box<dyn Fn(RenderCommand) + Send + Sync + 'a>;
 
 impl<'a> RenderCommandListener<'a> {
+    /// Wraps a render command callback in a `RenderCommandListener`.
     #[inline]
     pub fn new(send_fn: RenderCommandSendFunction<'a>) -> RenderCommandListener<'a> {
         RenderCommandListener { send_fn }
     }
 
     #[inline]
-    pub fn send(&self, render_command: RenderCommand) {
+    pub(crate) fn send(&self, render_command: RenderCommand) {
         (self.send_fn)(render_command)
     }
 }
@@ -39,8 +49,12 @@ impl<'a> RenderCommandListener<'a> {
 /// Options that influence scene building.
 #[derive(Clone, Default)]
 pub struct BuildOptions {
+    /// A global transform to be applied to the scene.
     pub transform: RenderTransform,
+    /// Expands outlines by the given number of device pixels. This is useful to perform *stem
+    /// darkening* for fonts, to mitigate the thinness of gamma-corrected fonts.
     pub dilation: Vector2F,
+    /// True if subpixel antialiasing for LCD screens is to be performed.
     pub subpixel_aa_enabled: bool,
 }
 
@@ -54,9 +68,12 @@ impl BuildOptions {
     }
 }
 
+/// A global transform to apply to the scene.
 #[derive(Clone)]
 pub enum RenderTransform {
+    /// A 2D transform.
     Transform2D(Transform2F),
+    /// A perspective transform. (This will soon be removed in favor of a revised 3D approach.)
     Perspective(Perspective),
 }
 
diff --git a/renderer/src/paint.rs b/renderer/src/paint.rs
index 878f9caf..052a3111 100644
--- a/renderer/src/paint.rs
+++ b/renderer/src/paint.rs
@@ -34,8 +34,8 @@ use std::sync::Arc;
 const GRADIENT_TILE_LENGTH: u32 = 256;
 
 #[derive(Clone)]
-pub struct Palette {
-    pub paints: Vec<Paint>,
+pub(crate) struct Palette {
+    pub(crate) paints: Vec<Paint>,
     render_targets: Vec<RenderTargetData>,
     cache: HashMap<Paint, PaintId>,
     allocator: TextureAllocator,
@@ -90,7 +90,7 @@ impl Debug for PaintContents {
 
 impl Palette {
     #[inline]
-    pub fn new(scene_id: SceneId) -> Palette {
+    pub(crate) fn new(scene_id: SceneId) -> Palette {
         Palette {
             paints: vec![],
             render_targets: vec![],
@@ -266,62 +266,62 @@ impl PaintOverlay {
     }
 }
 
-pub struct PaintInfo {
+pub(crate) struct PaintInfo {
     /// The render commands needed to prepare the textures.
-    pub render_commands: Vec<RenderCommand>,
+    pub(crate) render_commands: Vec<RenderCommand>,
     /// The metadata for each paint.
     ///
     /// The indices of this vector are paint IDs.
-    pub paint_metadata: Vec<PaintMetadata>,
+    pub(crate) paint_metadata: Vec<PaintMetadata>,
     /// The metadata for each render target.
     ///
     /// The indices of this vector are render target IDs.
-    pub render_target_metadata: Vec<RenderTargetMetadata>,
+    pub(crate) render_target_metadata: Vec<RenderTargetMetadata>,
 }
 
 #[derive(Debug)]
-pub struct PaintMetadata {
+pub(crate) struct PaintMetadata {
     /// Metadata associated with the color texture, if applicable.
-    pub color_texture_metadata: Option<PaintColorTextureMetadata>,
+    pub(crate) color_texture_metadata: Option<PaintColorTextureMetadata>,
     /// The base color that the color texture gets mixed into.
-    pub base_color: ColorU,
-    pub blend_mode: BlendMode,
+    pub(crate) base_color: ColorU,
+    pub(crate) blend_mode: BlendMode,
     /// True if this paint is fully opaque.
-    pub is_opaque: bool,
+    pub(crate) is_opaque: bool,
 }
 
 #[derive(Debug)]
-pub struct PaintColorTextureMetadata {
+pub(crate) struct PaintColorTextureMetadata {
     /// The location of the paint.
-    pub location: TextureLocation,
+    pub(crate) location: TextureLocation,
     /// The scale for the page this paint is on.
-    pub page_scale: Vector2F,
+    pub(crate) page_scale: Vector2F,
     /// The transform to apply to screen coordinates to translate them into UVs.
-    pub transform: Transform2F,
+    pub(crate) transform: Transform2F,
     /// The sampling mode for the texture.
-    pub sampling_flags: TextureSamplingFlags,
+    pub(crate) sampling_flags: TextureSamplingFlags,
     /// The filter to be applied to this paint.
-    pub filter: PaintFilter,
+    pub(crate) filter: PaintFilter,
     /// How the color texture is to be composited over the base color.
-    pub composite_op: PaintCompositeOp,
+    pub(crate) composite_op: PaintCompositeOp,
 }
 
 #[derive(Clone, Copy, Debug)]
-pub struct RadialGradientMetadata {
+pub(crate) struct RadialGradientMetadata {
     /// The line segment that connects the two circles.
-    pub line: LineSegment2F,
+    pub(crate) line: LineSegment2F,
     /// The radii of the two circles.
-    pub radii: F32x2,
+    pub(crate) radii: F32x2,
 }
 
 #[derive(Clone, Copy, Debug)]
-pub struct RenderTargetMetadata {
+pub(crate) struct RenderTargetMetadata {
     /// The location of the render target.
-    pub location: TextureLocation,
+    pub(crate) location: TextureLocation,
 }
 
 #[derive(Debug)]
-pub enum PaintFilter {
+pub(crate) enum PaintFilter {
     None,
     RadialGradient {
         /// The line segment that connects the two circles.
@@ -334,7 +334,7 @@ pub enum PaintFilter {
 
 impl Palette {
     #[allow(clippy::trivially_copy_pass_by_ref)]
-    pub fn push_paint(&mut self, paint: &Paint) -> PaintId {
+    pub(crate) fn push_paint(&mut self, paint: &Paint) -> PaintId {
         if let Some(paint_id) = self.cache.get(paint) {
             return *paint_id;
         }
@@ -345,7 +345,7 @@ impl Palette {
         paint_id
     }
 
-    pub fn push_render_target(&mut self, render_target: RenderTarget) -> RenderTargetId {
+    pub(crate) fn push_render_target(&mut self, render_target: RenderTarget) -> RenderTargetId {
         let id = self.render_targets.len() as u32;
 
         let metadata = RenderTargetMetadata {
@@ -356,7 +356,7 @@ impl Palette {
         RenderTargetId { scene: self.scene_id.0, render_target: id }
     }
 
-    pub fn build_paint_info(&mut self, render_transform: Transform2F) -> PaintInfo {
+    pub(crate) fn build_paint_info(&mut self, render_transform: Transform2F) -> PaintInfo {
         let mut paint_metadata = vec![];
 
         // Assign paint locations.
diff --git a/renderer/src/scene.rs b/renderer/src/scene.rs
index 693abb13..ac032bcc 100644
--- a/renderer/src/scene.rs
+++ b/renderer/src/scene.rs
@@ -8,7 +8,7 @@
 // option. This file may not be copied, modified, or distributed
 // except according to those terms.
 
-//! A set of paths to be rendered.
+//! The vector scene to be rendered.
 
 use crate::builder::SceneBuilder;
 use crate::concurrent::executor::Executor;
@@ -34,6 +34,7 @@ use std::u64;
 
 static NEXT_SCENE_ID: AtomicUsize = AtomicUsize::new(0);
 
+/// The vector scene to be rendered.
 #[derive(Clone)]
 pub struct Scene {
     display_list: Vec<DisplayItem>,
@@ -156,7 +157,7 @@ impl Scene {
     }
 
     #[inline]
-    pub fn build_paint_info(&mut self, render_transform: Transform2F) -> PaintInfo {
+    pub(crate) fn build_paint_info(&mut self, render_transform: Transform2F) -> PaintInfo {
         self.palette.build_paint_info(render_transform)
     }
 
@@ -281,8 +282,8 @@ impl Scene {
     }
 
     #[inline]
-    pub fn palette(&self) -> &Palette {
-        &self.palette
+    pub fn get_paint(&self, paint_id: PaintId) -> &Paint {
+        self.palette.paints.get(paint_id.0 as usize).expect("No paint with that ID!")
     }
 
     #[inline]