@ikaros-arch/react-3dhop-iiif 0.1.0 → 0.2.0

package/README.md

@@ -1,381 +1,381 @@
-# @ikaros-arch/react-3dhop-iiif
-
-Renders [IIIF Presentation 4.0 / IIIF 3D](https://github.com/IIIF/3d/blob/main/temp-draft-4.md)
-manifests with [3DHOP](https://3dhop.net/), on top of
-[`@ikaros-arch/react-3dhop`](https://github.com/ikaros-arch/react-3dhop/tree/main/packages/react-3dhop#readme).
-
-The IIIF 3D specification is still a draft. Keeping it in its own package means the core viewer
-does not have to move every time the draft does.
-
-```bash
-npm install @ikaros-arch/react-3dhop-iiif @ikaros-arch/react-3dhop @ikaros-arch/3dhop
-```
-
-`react`, `react-dom` and `@ikaros-arch/react-3dhop` are peer dependencies (and `@ikaros-arch/3dhop`
-is a peer of that, holding the 3DHOP runtime assets). The package itself has **no runtime
-dependencies**.
-
-This README is the reference for manifest authoring and specification coverage.
-[docs/iiif.md](https://github.com/ikaros-arch/react-3dhop/blob/main/docs/iiif.md) explains how the
-package is put together and where to extend it.
-
----
-
-## Quick start
-
-```tsx
-import { IIIFViewer, IIIFMetadataPanel, IIIFSavedViewsPanel } from '@ikaros-arch/react-3dhop-iiif';
-
-export function Viewer() {
-  return (
-    <IIIFViewer
-      manifest="https://example.org/iiif/object/manifest.json"
-      assetBaseUrl="/3dhop"
-      width={800}
-      height={600}
-    >
-      <IIIFMetadataPanel />
-      <IIIFSavedViewsPanel />
-    </IIIFViewer>
-  );
-}
-```
-
-`<IIIFViewer>` accepts everything `<ThreeDHopViewer>` does, except `models` and `modelUrl`, which
-it derives from the manifest. `manifest` takes either a URL to fetch or an already-parsed manifest
-object.
-
-Panels must be rendered **inside** `<IIIFViewer>` to reach its context. The viewer renders its
-children into a fixed-size, clipped box intended for canvas overlays, so to place panels elsewhere
-on the page, portal them out — React context passes through portals. See
-[`examples/react-3dhop-demo/src/IIIFDemo.tsx`](https://github.com/ikaros-arch/react-3dhop/blob/main/examples/react-3dhop-demo/src/IIIFDemo.tsx).
-
-### Reading the manifest yourself
-
-```tsx
-const { metadata, models, cameras, language, setLanguage, goToCamera, saveCurrentView } =
-  useIIIFManifest();
-```
-
-Every parsing and geometry function is also exported standalone, with no React and no dependency on
-3DHOP's globals — usable in Node, and unit-tested there:
-
-```ts
-import { loadManifest, sceneFromManifest, buildModelMatrix, view2track } from '@ikaros-arch/react-3dhop-iiif';
-
-const parsed = await loadManifest(url);
-const scene = sceneFromManifest(parsed, { displayUnit: 'cm' });
-```
-
----
-
-## Manifest authoring
-
-### Model formats
-
-3DHOP streams [Nexus](https://vcg.isti.cnr.it/nexus/) (`.nxz`/`.nxs`) and loads `.ply`. When one
-annotation offers several sources, the best supported one wins:
-
-| Rank | `format` | Support |
-|---|---|---|
-| 1 | `nexus`, `nxz`, `nxs`, `application/octet-stream+nexus` | Streamed progressively — the format to publish |
-| 2 | `ply`, `application/ply` | Loaded whole |
-| 3 | `obj`, `model/obj` | **Cannot be rendered** — ranked here only so it is preferred over glTF |
-| 4 | `glb`, `gltf`, `model/gltf-binary`, `model/gltf+json` | **Cannot be rendered** |
-
-Comparison is case-insensitive, so `"format": "Nexus"` works. An unrecognised format ranks below
-every known one rather than being discarded.
-
-3DHOP has loaders for Nexus and PLY only. A model in any other format is still parsed, so the
-manifest continues to describe the object, and a diagnostic says why nothing appeared. Adding an
-OBJ or glTF loader is out of scope; convert to Nexus with
-[`nxsbuild`](https://vcg.isti.cnr.it/nexus/) instead.
-
-### Units
-
-Two non-standard metadata fields control scale, recognised by their English or Norwegian labels:
-
-| Label | Meaning | Default |
-|---|---|---|
-| `Measure Unit` / `Måleenhet` | The unit the manifest's own coordinates are authored in | `mm` |
-| `Display Unit` / `Visningsenhet` | The unit the scene is rendered and measured in | the measure unit |
-
-```json
-{ "label": { "en": ["Measure Unit"] }, "value": { "en": ["m"] } }
-```
-
-A single model whose mesh is authored in a different unit from the rest of the manifest can say so
-on its source, without disturbing anything else:
-
-```json
-{ "id": "…/alvim.nxz", "type": "Model", "format": "Nexus", "measureUnit": "mm" }
-```
-
-Recognised units are `km`, `m`, `cm`, `mm`, `um` (or `µm`) and `nm`. An unrecognised unit is
-treated as a factor of 1, so a typo renders the object unscaled rather than failing.
-
-The `Display Unit` also sets the label on measurements taken with the measuring tool, and picks the
-clipping-border width, which otherwise looks invisible on a millimetre-scale object and enormous on
-a metre-scale one.
-
-### Placement and transforms
-
-A model's position comes from the `PointSelector` on the annotation's target:
-
-```json
-"target": {
-  "type": "SpecificResource",
-  "source": [{ "id": "…/scene/1", "type": "Scene" }],
-  "selector": [{ "type": "PointSelector", "x": 2.0, "y": 0.0, "z": 0.0 }]
-}
-```
-
-`transform` is an **ordered list**, applied first entry first, and this package composes it in
-order rather than flattening it. Order is significant: scaling then translating moves the model by
-the unscaled offset, while translating then scaling moves it by the scaled one. The same holds for
-rotation.
-
-```json
-"transform": [
-  { "type": "ScaleTransform", "x": 1.5, "y": 1.5, "z": 1.5 },
-  { "type": "RotateTransform", "x": -90.0, "y": -90.0, "z": 0.0 },
-  { "type": "TranslateTransform", "x": 0.0, "y": 0.5, "z": 0.0 }
-]
-```
-
-`RotateTransform` accepts either per-axis Euler angles in degrees (`x`/`y`/`z`, composed Z then Y
-then X, matching 3DHOP's own rotation triple) or a single axis-angle (`{ "axis": "y", "angle": 45 }`).
-
-The complete model matrix is:
-
-```text
-M = S(displayUnit ← measureUnit) · T(pointSelector) · L(transform list) · S(geometry unit override)
-```
-
-Read right to left: bring the mesh's vertices into the manifest's unit, apply the transform list,
-place the result at the point selector, then convert the whole scene into the display unit. Keeping
-the unit conversions outermost and innermost is what makes positions and translations scale
-alongside the geometry.
-
-Models sharing a source URL are collapsed onto one mesh, so placing the same object several times
-downloads it once.
-
-### Cameras
-
-Camera annotations become buttons in `<IIIFSavedViewsPanel>`, and the first one is applied as the
-opening view unless `applyInitialCamera={false}`:
-
-```json
-{
-  "type": "Annotation",
-  "body": {
-    "type": "PerspectiveCamera",
-    "label": { "en": ["Front"] },
-    "fieldOfView": 50.0,
-    "lookAt": { "type": "PointSelector", "x": 0, "y": 0.5, "z": 0 }
-  },
-  "target": {
-    "type": "SpecificResource",
-    "source": [{ "id": "…/scene/1", "type": "Scene" }],
-    "selector": [{ "type": "PointSelector", "x": 0.0, "y": 3.0, "z": -8.0 }]
-  }
-}
-```
-
-The target's `PointSelector` is the camera position; `lookAt` is what it points at. `lookAt` may
-instead reference another annotation by `id`, in which case the camera aims at that model's
-position. Both `fieldOfView` and the shorthand `fov` are read. With no `lookAt`, the camera aims at
-the scene centre.
-
-`saveCurrentView()` returns an annotation in exactly this shape, ready to paste back into a
-manifest.
-
-### Language
-
-Any IIIF language map is resolved against the active language, falling back through `en`, `no`,
-`nb`, `nn` and then any remaining language, so a partially translated manifest still renders. The
-language defaults to the browser's and can be overridden with the `language` prop or switched at
-runtime through `<IIIFLanguageSwitcher>`, which hides itself when there is only one language to
-choose from.
-
-Metadata fields are recognised by their **English** label even when another language is displayed,
-so switching language does not lose the unit or inventory fields.
-
----
-
-## Panels
-
-All read from `useIIIFManifest()` unless noted otherwise, use semantic markup with no CSS
-framework, and accept `className` props on every element. There is no bundled stylesheet; the
-demo's [`IIIFDemo.css`](https://github.com/ikaros-arch/react-3dhop/blob/main/examples/react-3dhop-demo/src/IIIFDemo.css) is a starting point.
-
-| Component | Shows |
-|---|---|
-| `<IIIFSummary>` | `summary` and the `requiredStatement` attribution |
-| `<IIIFMetadataPanel>` | Recognised fields first, then the rest in manifest order |
-| `<IIIFModelsPanel>` | Per-model visibility and transparency toggles |
-| `<IIIFSavedViewsPanel>` | The manifest's cameras, plus "save current view" |
-| `<IIIFLanguageSwitcher>` | Language selector; hidden when the manifest has one language |
-| `<IIIFCollectionPicker>` | A `<select>` of every manifest in a collection |
-| `<IIIFCollectionCarousel>` | A thumbnail strip of every manifest in a collection, with prev/next |
-| `<IIIFMultiManifestModelsPanel>` | Visibility/transparency toggles grouped by manifest; reads `useIIIFMultiManifest()`, for use inside `<IIIFMultiManifestViewer>` instead of `<IIIFViewer>` |
-
----
-
-## Collections
-
-A IIIF Collection lists several manifests — e.g. every object in a museum sub-catalogue — so a
-picker or carousel can switch between them. `<IIIFCollectionProvider>` fetches and parses one, and
-tracks which manifest is selected; it renders nothing itself, so it composes with `<IIIFViewer>`
-rather than replacing it:
-
-```tsx
-import {
-  IIIFCollectionProvider,
-  IIIFCollectionPicker,
-  IIIFCollectionCarousel,
-  IIIFViewer,
-  useIIIFCollection
-} from '@ikaros-arch/react-3dhop-iiif';
-
-function CollectionBrowser() {
-  return (
-    <IIIFCollectionProvider collection="/manifests/bitfrost/collection.json">
-      <IIIFCollectionPicker />
-      <IIIFCollectionCarousel />
-      <SelectedManifestViewer />
-    </IIIFCollectionProvider>
-  );
-}
-
-function SelectedManifestViewer() {
-  const { selectedId } = useIIIFCollection();
-  if (!selectedId) return null;
-  // Remounting on selection change avoids carrying one object's camera into the next.
-  return <IIIFViewer key={selectedId} manifest={selectedId} assetBaseUrl="/3dhop" width={760} height={620} />;
-}
-```
-
-`collection` accepts a URL or an already-fetched object, mirroring `<IIIFViewer manifest>`.
-`useIIIFCollection()` also exposes `items` (id, label, thumbnail), `status`/`error`, and
-`next()`/`previous()` for building a custom control. See
-[`examples/react-3dhop-demo/src/IIIFDemo.tsx`](https://github.com/ikaros-arch/react-3dhop/blob/main/examples/react-3dhop-demo/src/IIIFDemo.tsx)
-for a full example, including the [generator
-script](https://github.com/ikaros-arch/react-3dhop/blob/main/examples/react-3dhop-demo/scripts/generate-bitfrost-collection.mjs) that builds its
-sample collection from a museum catalogue export.
-
----
-
-## Multiple manifests in one viewer
-
-`<IIIFMultiManifestViewer>` renders several manifests together in a single `<ThreeDHopViewer>`
-instance, laid out side by side rather than switched between one at a time:
-
-```tsx
-import { IIIFMultiManifestViewer, IIIFMultiManifestModelsPanel } from '@ikaros-arch/react-3dhop-iiif';
-
-function Comparison() {
-  return (
-    <IIIFMultiManifestViewer
-      manifests={['/manifests/a.json', '/manifests/b.json']}
-      assetBaseUrl="/3dhop"
-      width={760}
-      height={620}
-    >
-      <IIIFMultiManifestModelsPanel />
-    </IIIFMultiManifestViewer>
-  );
-}
-```
-
-It is a parallel component, not an extension of `<IIIFViewer>`: it fetches and parses each manifest
-independently, and fails fast if any one of them can't be loaded, since a partially-failed shared
-scene isn't a state worth trying to render. Its context, `useIIIFMultiManifest()`, only covers
-per-model visibility and transparency, scoped by manifest — cameras, saved views, metadata and
-language are per-manifest concepts without a defined multi-manifest behaviour yet, so `<IIIFViewer>`
-remains the way to reach those.
-
-There is no mesh geometry available at layout time — meshes stream in later, and the manifest only
-says where each annotation is *placed*, not how large it is — so the side-by-side layout is a
-heuristic: each manifest's own placement spread (the spread between its own annotations), floored by
-`minRadius` and separated by `gap`. A manifest with a single model at the origin (the common case)
-has a spread of zero and falls back entirely to `minRadius`. Tune both options for your data:
-
-```tsx
-<IIIFMultiManifestViewer manifests={ids} minRadius={150} gap={80} columns={3} />
-```
-
-`useIIIFMultiManifest()` addresses models with a `(manifestKey, modelId)` pair rather than the bare
-IIIF annotation id `useIIIFManifest()` uses, because two manifests can otherwise declare colliding
-ids:
-
-```tsx
-const { manifests, isModelVisible, setModelVisible } = useIIIFMultiManifest();
-// manifests: [{ key, sourceId, label, models }, …]
-```
-
-See [`examples/react-3dhop-demo/src/IIIFDemo.tsx`](https://github.com/ikaros-arch/react-3dhop/blob/main/examples/react-3dhop-demo/src/IIIFDemo.tsx)
-(`"Multi-manifest viewer"` mode) for a full example that lets the user check off any number of
-objects from a collection.
-
----
-
-## Error handling
-
-A manifest that cannot be fetched or parsed puts the viewer into an error state and renders
-`errorFallback` — it does not throw or leave a blank canvas.
-
-Problems affecting only part of a manifest are reported as **diagnostics** and the rest still
-renders: an annotation with no usable source, a glTF-only model, an unrecognised body type, a
-manifest with no `Scene`, or one with several. Diagnostics reach you through the `onDiagnostic`
-prop or the `diagnostics` array on the context.
-
----
-
-## Supported IIIF 3D subset
-
-Against the [draft specification](https://github.com/IIIF/3d/blob/main/temp-draft-4.md). ✅ full,
-⚠️ partial, ❌ not implemented.
-
-| Feature | Parsed | Rendered | Notes |
-|---|---|---|---|
-| **Containers** |
-| `Scene` | ✅ | ✅ | |
-| Multiple scenes | ✅ | ⚠️ | Merged into one view, with a diagnostic |
-| Canvas in Scene | ❌ | ❌ | |
-| Nested scenes | ❌ | ❌ | |
-| **Resources** |
-| `Model` as body | ✅ | ✅ | |
-| `Model` via `SpecificResource` | ✅ | ✅ | Required for transforms |
-| Several sources per annotation | ✅ | ✅ | Best supported format wins |
-| `PerspectiveCamera` | ✅ | ✅ | Position, `lookAt`, `fieldOfView` |
-| `OrthographicCamera` | ✅ | ✅ | Position, `lookAt` |
-| **Lights** (`Ambient`/`Directional`/`Point`/`Spot`) | ❌ | ❌ | 3DHOP has one directional light, controlled by the viewer |
-| **Transforms** |
-| `ScaleTransform` | ✅ | ✅ | |
-| `TranslateTransform` | ✅ | ✅ | |
-| `RotateTransform` | ✅ | ✅ | Euler and axis-angle forms |
-| Ordered composition | ✅ | ✅ | Composed as a matrix, not flattened |
-| **Selectors** |
-| `PointSelector` | ✅ | ✅ | Positions models and cameras |
-| `PolygonZSelector` | ❌ | ❌ | For Canvas placement |
-| **Properties** |
-| `label`, `summary`, `metadata` | ✅ | ✅ | Multilingual |
-| `requiredStatement` | ✅ | ✅ | Shown as attribution |
-| `fieldOfView` / `fov` | ✅ | ✅ | |
-| `lookAt` | ✅ | ✅ | Point or annotation reference |
-| `backgroundColor` | ❌ | ❌ | Use the viewer's `backgroundUrl` |
-| `duration` | ❌ | ❌ | Temporal scenes |
-| `near` / `far` | ❌ | ❌ | Derived from scene extents instead |
-| `exclude` | ❌ | ❌ | |
-| **Formats** |
-| Nexus (`.nxz`/`.nxs`) | ✅ | ✅ | Streamed progressively |
-| PLY | ✅ | ✅ | |
-| OBJ | ✅ | ❌ | Reported as a diagnostic; convert to Nexus |
-| glTF / GLB | ✅ | ❌ | Reported as a diagnostic; convert to Nexus |
-
----
-
-## Licence
-
-GPL-3.0-or-later, matching 3DHOP.
+# @ikaros-arch/react-3dhop-iiif
+
+Renders [IIIF Presentation 4.0 / IIIF 3D](https://github.com/IIIF/3d/blob/main/temp-draft-4.md)
+manifests with [3DHOP](https://3dhop.net/), on top of
+[`@ikaros-arch/react-3dhop`](https://github.com/ikaros-arch/react-3dhop/tree/main/packages/react-3dhop#readme).
+
+The IIIF 3D specification is still a draft. Keeping it in its own package means the core viewer
+does not have to move every time the draft does.
+
+```bash
+npm install @ikaros-arch/react-3dhop-iiif @ikaros-arch/react-3dhop @ikaros-arch/3dhop
+```
+
+`react`, `react-dom` and `@ikaros-arch/react-3dhop` are peer dependencies (and `@ikaros-arch/3dhop`
+is a peer of that, holding the 3DHOP runtime assets). The package itself has **no runtime
+dependencies**.
+
+This README is the reference for manifest authoring and specification coverage.
+[docs/iiif.md](https://github.com/ikaros-arch/react-3dhop/blob/main/docs/iiif.md) explains how the
+package is put together and where to extend it.
+
+---
+
+## Quick start
+
+```tsx
+import { IIIFViewer, IIIFMetadataPanel, IIIFSavedViewsPanel } from '@ikaros-arch/react-3dhop-iiif';
+
+export function Viewer() {
+  return (
+    <IIIFViewer
+      manifest="https://example.org/iiif/object/manifest.json"
+      assetBaseUrl="/3dhop"
+      width={800}
+      height={600}
+    >
+      <IIIFMetadataPanel />
+      <IIIFSavedViewsPanel />
+    </IIIFViewer>
+  );
+}
+```
+
+`<IIIFViewer>` accepts everything `<ThreeDHopViewer>` does, except `models` and `modelUrl`, which
+it derives from the manifest. `manifest` takes either a URL to fetch or an already-parsed manifest
+object.
+
+Panels must be rendered **inside** `<IIIFViewer>` to reach its context. The viewer renders its
+children into a fixed-size, clipped box intended for canvas overlays, so to place panels elsewhere
+on the page, portal them out — React context passes through portals. See
+[`examples/react-3dhop-demo/src/IIIFDemo.tsx`](https://github.com/ikaros-arch/react-3dhop/blob/main/examples/react-3dhop-demo/src/IIIFDemo.tsx).
+
+### Reading the manifest yourself
+
+```tsx
+const { metadata, models, cameras, language, setLanguage, goToCamera, saveCurrentView } =
+  useIIIFManifest();
+```
+
+Every parsing and geometry function is also exported standalone, with no React and no dependency on
+3DHOP's globals — usable in Node, and unit-tested there:
+
+```ts
+import { loadManifest, sceneFromManifest, buildModelMatrix, view2track } from '@ikaros-arch/react-3dhop-iiif';
+
+const parsed = await loadManifest(url);
+const scene = sceneFromManifest(parsed, { displayUnit: 'cm' });
+```
+
+---
+
+## Manifest authoring
+
+### Model formats
+
+3DHOP streams [Nexus](https://vcg.isti.cnr.it/nexus/) (`.nxz`/`.nxs`) and loads `.ply`. When one
+annotation offers several sources, the best supported one wins:
+
+| Rank | `format` | Support |
+|---|---|---|
+| 1 | `nexus`, `nxz`, `nxs`, `application/octet-stream+nexus` | Streamed progressively — the format to publish |
+| 2 | `ply`, `application/ply` | Loaded whole |
+| 3 | `obj`, `model/obj` | **Cannot be rendered** — ranked here only so it is preferred over glTF |
+| 4 | `glb`, `gltf`, `model/gltf-binary`, `model/gltf+json` | **Cannot be rendered** |
+
+Comparison is case-insensitive, so `"format": "Nexus"` works. An unrecognised format ranks below
+every known one rather than being discarded.
+
+3DHOP has loaders for Nexus and PLY only. A model in any other format is still parsed, so the
+manifest continues to describe the object, and a diagnostic says why nothing appeared. Adding an
+OBJ or glTF loader is out of scope; convert to Nexus with
+[`nxsbuild`](https://vcg.isti.cnr.it/nexus/) instead.
+
+### Units
+
+Two non-standard metadata fields control scale, recognised by their English or Norwegian labels:
+
+| Label | Meaning | Default |
+|---|---|---|
+| `Measure Unit` / `Måleenhet` | The unit the manifest's own coordinates are authored in | `mm` |
+| `Display Unit` / `Visningsenhet` | The unit the scene is rendered and measured in | the measure unit |
+
+```json
+{ "label": { "en": ["Measure Unit"] }, "value": { "en": ["m"] } }
+```
+
+A single model whose mesh is authored in a different unit from the rest of the manifest can say so
+on its source, without disturbing anything else:
+
+```json
+{ "id": "…/alvim.nxz", "type": "Model", "format": "Nexus", "measureUnit": "mm" }
+```
+
+Recognised units are `km`, `m`, `cm`, `mm`, `um` (or `µm`) and `nm`. An unrecognised unit is
+treated as a factor of 1, so a typo renders the object unscaled rather than failing.
+
+The `Display Unit` also sets the label on measurements taken with the measuring tool, and picks the
+clipping-border width, which otherwise looks invisible on a millimetre-scale object and enormous on
+a metre-scale one.
+
+### Placement and transforms
+
+A model's position comes from the `PointSelector` on the annotation's target:
+
+```json
+"target": {
+  "type": "SpecificResource",
+  "source": [{ "id": "…/scene/1", "type": "Scene" }],
+  "selector": [{ "type": "PointSelector", "x": 2.0, "y": 0.0, "z": 0.0 }]
+}
+```
+
+`transform` is an **ordered list**, applied first entry first, and this package composes it in
+order rather than flattening it. Order is significant: scaling then translating moves the model by
+the unscaled offset, while translating then scaling moves it by the scaled one. The same holds for
+rotation.
+
+```json
+"transform": [
+  { "type": "ScaleTransform", "x": 1.5, "y": 1.5, "z": 1.5 },
+  { "type": "RotateTransform", "x": -90.0, "y": -90.0, "z": 0.0 },
+  { "type": "TranslateTransform", "x": 0.0, "y": 0.5, "z": 0.0 }
+]
+```
+
+`RotateTransform` accepts either per-axis Euler angles in degrees (`x`/`y`/`z`, composed Z then Y
+then X, matching 3DHOP's own rotation triple) or a single axis-angle (`{ "axis": "y", "angle": 45 }`).
+
+The complete model matrix is:
+
+```text
+M = S(displayUnit ← measureUnit) · T(pointSelector) · L(transform list) · S(geometry unit override)
+```
+
+Read right to left: bring the mesh's vertices into the manifest's unit, apply the transform list,
+place the result at the point selector, then convert the whole scene into the display unit. Keeping
+the unit conversions outermost and innermost is what makes positions and translations scale
+alongside the geometry.
+
+Models sharing a source URL are collapsed onto one mesh, so placing the same object several times
+downloads it once.
+
+### Cameras
+
+Camera annotations become buttons in `<IIIFSavedViewsPanel>`, and the first one is applied as the
+opening view unless `applyInitialCamera={false}`:
+
+```json
+{
+  "type": "Annotation",
+  "body": {
+    "type": "PerspectiveCamera",
+    "label": { "en": ["Front"] },
+    "fieldOfView": 50.0,
+    "lookAt": { "type": "PointSelector", "x": 0, "y": 0.5, "z": 0 }
+  },
+  "target": {
+    "type": "SpecificResource",
+    "source": [{ "id": "…/scene/1", "type": "Scene" }],
+    "selector": [{ "type": "PointSelector", "x": 0.0, "y": 3.0, "z": -8.0 }]
+  }
+}
+```
+
+The target's `PointSelector` is the camera position; `lookAt` is what it points at. `lookAt` may
+instead reference another annotation by `id`, in which case the camera aims at that model's
+position. Both `fieldOfView` and the shorthand `fov` are read. With no `lookAt`, the camera aims at
+the scene centre.
+
+`saveCurrentView()` returns an annotation in exactly this shape, ready to paste back into a
+manifest.
+
+### Language
+
+Any IIIF language map is resolved against the active language, falling back through `en`, `no`,
+`nb`, `nn` and then any remaining language, so a partially translated manifest still renders. The
+language defaults to the browser's and can be overridden with the `language` prop or switched at
+runtime through `<IIIFLanguageSwitcher>`, which hides itself when there is only one language to
+choose from.
+
+Metadata fields are recognised by their **English** label even when another language is displayed,
+so switching language does not lose the unit or inventory fields.
+
+---
+
+## Panels
+
+All read from `useIIIFManifest()` unless noted otherwise, use semantic markup with no CSS
+framework, and accept `className` props on every element. There is no bundled stylesheet; the
+demo's [`IIIFDemo.css`](https://github.com/ikaros-arch/react-3dhop/blob/main/examples/react-3dhop-demo/src/IIIFDemo.css) is a starting point.
+
+| Component | Shows |
+|---|---|
+| `<IIIFSummary>` | `summary` and the `requiredStatement` attribution |
+| `<IIIFMetadataPanel>` | Recognised fields first, then the rest in manifest order |
+| `<IIIFModelsPanel>` | Per-model visibility and transparency toggles |
+| `<IIIFSavedViewsPanel>` | The manifest's cameras, plus "save current view" |
+| `<IIIFLanguageSwitcher>` | Language selector; hidden when the manifest has one language |
+| `<IIIFCollectionPicker>` | A `<select>` of every manifest in a collection |
+| `<IIIFCollectionCarousel>` | A thumbnail strip of every manifest in a collection, with prev/next |
+| `<IIIFMultiManifestModelsPanel>` | Visibility/transparency toggles grouped by manifest; reads `useIIIFMultiManifest()`, for use inside `<IIIFMultiManifestViewer>` instead of `<IIIFViewer>` |
+
+---
+
+## Collections
+
+A IIIF Collection lists several manifests — e.g. every object in a museum sub-catalogue — so a
+picker or carousel can switch between them. `<IIIFCollectionProvider>` fetches and parses one, and
+tracks which manifest is selected; it renders nothing itself, so it composes with `<IIIFViewer>`
+rather than replacing it:
+
+```tsx
+import {
+  IIIFCollectionProvider,
+  IIIFCollectionPicker,
+  IIIFCollectionCarousel,
+  IIIFViewer,
+  useIIIFCollection
+} from '@ikaros-arch/react-3dhop-iiif';
+
+function CollectionBrowser() {
+  return (
+    <IIIFCollectionProvider collection="/manifests/bitfrost/collection.json">
+      <IIIFCollectionPicker />
+      <IIIFCollectionCarousel />
+      <SelectedManifestViewer />
+    </IIIFCollectionProvider>
+  );
+}
+
+function SelectedManifestViewer() {
+  const { selectedId } = useIIIFCollection();
+  if (!selectedId) return null;
+  // Remounting on selection change avoids carrying one object's camera into the next.
+  return <IIIFViewer key={selectedId} manifest={selectedId} assetBaseUrl="/3dhop" width={760} height={620} />;
+}
+```
+
+`collection` accepts a URL or an already-fetched object, mirroring `<IIIFViewer manifest>`.
+`useIIIFCollection()` also exposes `items` (id, label, thumbnail), `status`/`error`, and
+`next()`/`previous()` for building a custom control. See
+[`examples/react-3dhop-demo/src/IIIFDemo.tsx`](https://github.com/ikaros-arch/react-3dhop/blob/main/examples/react-3dhop-demo/src/IIIFDemo.tsx)
+for a full example, including the [generator
+script](https://github.com/ikaros-arch/react-3dhop/blob/main/examples/react-3dhop-demo/scripts/generate-bitfrost-collection.mjs) that builds its
+sample collection from a museum catalogue export.
+
+---
+
+## Multiple manifests in one viewer
+
+`<IIIFMultiManifestViewer>` renders several manifests together in a single `<ThreeDHopViewer>`
+instance, laid out side by side rather than switched between one at a time:
+
+```tsx
+import { IIIFMultiManifestViewer, IIIFMultiManifestModelsPanel } from '@ikaros-arch/react-3dhop-iiif';
+
+function Comparison() {
+  return (
+    <IIIFMultiManifestViewer
+      manifests={['/manifests/a.json', '/manifests/b.json']}
+      assetBaseUrl="/3dhop"
+      width={760}
+      height={620}
+    >
+      <IIIFMultiManifestModelsPanel />
+    </IIIFMultiManifestViewer>
+  );
+}
+```
+
+It is a parallel component, not an extension of `<IIIFViewer>`: it fetches and parses each manifest
+independently, and fails fast if any one of them can't be loaded, since a partially-failed shared
+scene isn't a state worth trying to render. Its context, `useIIIFMultiManifest()`, only covers
+per-model visibility and transparency, scoped by manifest — cameras, saved views, metadata and
+language are per-manifest concepts without a defined multi-manifest behaviour yet, so `<IIIFViewer>`
+remains the way to reach those.
+
+There is no mesh geometry available at layout time — meshes stream in later, and the manifest only
+says where each annotation is *placed*, not how large it is — so the side-by-side layout is a
+heuristic: each manifest's own placement spread (the spread between its own annotations), floored by
+`minRadius` and separated by `gap`. A manifest with a single model at the origin (the common case)
+has a spread of zero and falls back entirely to `minRadius`. Tune both options for your data:
+
+```tsx
+<IIIFMultiManifestViewer manifests={ids} minRadius={150} gap={80} columns={3} />
+```
+
+`useIIIFMultiManifest()` addresses models with a `(manifestKey, modelId)` pair rather than the bare
+IIIF annotation id `useIIIFManifest()` uses, because two manifests can otherwise declare colliding
+ids:
+
+```tsx
+const { manifests, isModelVisible, setModelVisible } = useIIIFMultiManifest();
+// manifests: [{ key, sourceId, label, models }, …]
+```
+
+See [`examples/react-3dhop-demo/src/IIIFDemo.tsx`](https://github.com/ikaros-arch/react-3dhop/blob/main/examples/react-3dhop-demo/src/IIIFDemo.tsx)
+(`"Multi-manifest viewer"` mode) for a full example that lets the user check off any number of
+objects from a collection.
+
+---
+
+## Error handling
+
+A manifest that cannot be fetched or parsed puts the viewer into an error state and renders
+`errorFallback` — it does not throw or leave a blank canvas.
+
+Problems affecting only part of a manifest are reported as **diagnostics** and the rest still
+renders: an annotation with no usable source, a glTF-only model, an unrecognised body type, a
+manifest with no `Scene`, or one with several. Diagnostics reach you through the `onDiagnostic`
+prop or the `diagnostics` array on the context.
+
+---
+
+## Supported IIIF 3D subset
+
+Against the [draft specification](https://github.com/IIIF/3d/blob/main/temp-draft-4.md). ✅ full,
+⚠️ partial, ❌ not implemented.
+
+| Feature | Parsed | Rendered | Notes |
+|---|---|---|---|
+| **Containers** |
+| `Scene` | ✅ | ✅ | |
+| Multiple scenes | ✅ | ⚠️ | Merged into one view, with a diagnostic |
+| Canvas in Scene | ❌ | ❌ | |
+| Nested scenes | ❌ | ❌ | |
+| **Resources** |
+| `Model` as body | ✅ | ✅ | |
+| `Model` via `SpecificResource` | ✅ | ✅ | Required for transforms |
+| Several sources per annotation | ✅ | ✅ | Best supported format wins |
+| `PerspectiveCamera` | ✅ | ✅ | Position, `lookAt`, `fieldOfView` |
+| `OrthographicCamera` | ✅ | ✅ | Position, `lookAt` |
+| **Lights** (`Ambient`/`Directional`/`Point`/`Spot`) | ❌ | ❌ | 3DHOP has one directional light, controlled by the viewer |
+| **Transforms** |
+| `ScaleTransform` | ✅ | ✅ | |
+| `TranslateTransform` | ✅ | ✅ | |
+| `RotateTransform` | ✅ | ✅ | Euler and axis-angle forms |
+| Ordered composition | ✅ | ✅ | Composed as a matrix, not flattened |
+| **Selectors** |
+| `PointSelector` | ✅ | ✅ | Positions models and cameras |
+| `PolygonZSelector` | ❌ | ❌ | For Canvas placement |
+| **Properties** |
+| `label`, `summary`, `metadata` | ✅ | ✅ | Multilingual |
+| `requiredStatement` | ✅ | ✅ | Shown as attribution |
+| `fieldOfView` / `fov` | ✅ | ✅ | |
+| `lookAt` | ✅ | ✅ | Point or annotation reference |
+| `backgroundColor` | ❌ | ❌ | Use the viewer's `backgroundUrl` |
+| `duration` | ❌ | ❌ | Temporal scenes |
+| `near` / `far` | ❌ | ❌ | Derived from scene extents instead |
+| `exclude` | ❌ | ❌ | |
+| **Formats** |
+| Nexus (`.nxz`/`.nxs`) | ✅ | ✅ | Streamed progressively |
+| PLY | ✅ | ✅ | |
+| OBJ | ✅ | ❌ | Reported as a diagnostic; convert to Nexus |
+| glTF / GLB | ✅ | ❌ | Reported as a diagnostic; convert to Nexus |
+
+---
+
+## Licence
+
+GPL-3.0-or-later, matching 3DHOP.