diff --git a/docs/plans/2026-03-12-refboard-product-expansion-plan.md b/docs/plans/2026-03-12-refboard-product-expansion-plan.md
new file mode 100644
index 0000000..b21f7ef
--- /dev/null
+++ b/docs/plans/2026-03-12-refboard-product-expansion-plan.md
@@ -0,0 +1,544 @@
+# RefBoard Product Expansion Plan — 2026-03-12
+
+## Goal
+
+Turn RefBoard from a collaborative media board into a stronger visual workspace that combines:
+
+- PureRef-style media handling
+- Figma-style review and comments
+- Milanote-style notes and structure
+- Kosmik-style mixed-media research and retrieval
+
+This plan is written for iterative implementation with code review after each slice.
+
+---
+
+## Working Model
+
+### Roles
+
+- **Builder session**: implements one scoped slice at a time
+- **Reviewer session**: reviews code for correctness, regressions, UX consistency, and plan adherence
+
+### Delivery rules
+
+- Ship in thin vertical slices, not broad unfinished scaffolding
+- Preserve existing board stability and current media workflows
+- Avoid overloading the current `text` element with unrelated behavior
+- Add first-class object types where semantics differ
+- Prefer migration-safe schema and scene-format extensions
+
+### Definition of done for each slice
+
+- Data model is explicit
+- UI behavior is testable manually
+- Socket sync behavior is defined
+- Save/load survives reload
+- Permissions are enforced
+- Existing image/video flows still work
+
+---
+
+## Current State Summary
+
+### Already present
+
+- Infinite canvas with images, GIFs, videos, drawings, text, frames/groups
+- Real-time sync
+- Upload pipeline with image/video support
+- Thread/comment backend and frontend groundwork
+- Votes groundwork
+- Mattermost import
+
+### Main gaps
+
+- Comments need to reach a polished Figma-like UX
+- No sticky notes / note cards
+- No PDF media type
+- No markdown/doc object
+- Current `text` object is too primitive for notes/docs
+- No strong retrieval layer for mixed media
+- No Milanote-style structural/workflow objects
+
+---
+
+## Product Principles
+
+1. **Labels are not documents**
+   Keep lightweight canvas text for labels and annotations.
+
+2. **Every content type needs its own semantics**
+   `text`, `sticky`, `pdf`, `markdown`, `comment`, and `task` should not collapse into one generic object.
+
+3. **Canvas should stay visually fast**
+   Large or dense content must degrade gracefully with zoom and viewport distance.
+
+4. **Review is a first-class layer**
+   Comments, pins, resolution state, and jump-to-context should feel native, not bolted on.
+
+5. **Mixed media must be searchable**
+   PDF text, markdown text, note text, filenames, and comments should converge in one retrieval model.
+
+---
+
+## Object Model Direction
+
+### Keep
+
+- `image`
+- `video`
+- `drawing`
+- `text` for lightweight labels
+- `group` / frame
+
+### Add next
+
+- `sticky` — visual note card
+- `pdf` — document or page-based asset object
+- `markdown` — rich document card with preview
+
+### Add later
+
+- `web` — URL/article/embed card
+- `task` — checklist/work item card
+- `connector` — relationship line
+- `column` or stronger structure container if needed beyond frames
+
+---
+
+## Zoom Behavior Rules
+
+### `text`
+
+- World-space label
+- Scales naturally with zoom
+- Best for captions and annotations
+
+### `sticky`
+
+- Canvas object with card chrome
+- Text remains readable longer than plain labels
+- Use clamped on-screen font sizing or simplified preview when zoomed out
+
+### `markdown`
+
+- Card preview on canvas
+- Full reading/editing in side panel or modal inspector
+- Do not render entire long documents at tiny zoom levels
+
+### `comment`
+
+- Review UI overlay behavior, not a regular scene text object
+
+---
+
+## Priority Plan
+
+## Phase 1 — Review Layer Completion
+
+### Objective
+
+Finish comments into a reliable Figma-style review workflow before adding more content types.
+
+### Scope
+
+- Object-level comments
+- Point/region-pinned comments
+- Thread list + detail panel polish
+- Resolve/reopen flow
+- Unread state
+- Jump-to-object
+- Better pin visibility and interactions
+- Notification/toast behavior for incoming comments
+- Permissions for comment vs edit vs view
+
+### Key tasks
+
+1. Audit existing comments/votes code paths and close gaps
+2. Stabilize socket event flows and sender behavior
+3. Ensure orphan handling for deleted objects is sane
+4. Add unread/new indicators and active-thread focus behavior
+5. Tighten panel UX and pin hit-testing
+
+### Acceptance criteria
+
+- User can comment on an object and on a specific point
+- Another user sees updates live without reload
+- Threads can be resolved and reopened
+- Deleted objects do not break thread history
+- Jump-to-object is reliable
+- Review flow feels coherent on a live board
+
+### Reviewer focus
+
+- Event duplication
+- Board permission leaks
+- Orphaned thread correctness
+- Socket race conditions
+- UI regressions in editor layout
+
+---
+
+## Phase 2 — Sticky Notes
+
+### Objective
+
+Add Milanote-style note cards without damaging the current label text tool.
+
+### Scope
+
+- New `sticky` scene object
+- Preset colors/themes
+- Title/body or single-body variant
+- Resize behavior
+- Duplicate/copy-paste
+- Inline editing
+- Selection, layering, grouping
+
+### Data model
+
+Suggested fields:
+
+```ts
+type StickyObject = {
+  id: string;
+  type: 'sticky';
+  x: number; y: number; w: number; h: number;
+  sx: number; sy: number; angle: number; z: number;
+  opacity: number; locked: boolean; visible: boolean; name: string;
+  title?: string;
+  text: string;
+  theme: string;
+  fontSize: number;
+};
+```
+
+### Key tasks
+
+1. Extend scene schema and serialization
+2. Add Pixi renderer for note card
+3. Add editing UX
+4. Add toolbar action and shortcuts
+5. Validate transform, selection, save/load, sync
+
+### Acceptance criteria
+
+- Sticky can be created in one action
+- It looks like a note card, not raw canvas text
+- It persists and syncs correctly
+- It remains readable enough across zoom levels
+
+### Reviewer focus
+
+- Scene-format backward compatibility
+- Hitbox and transform correctness
+- Editing UX under zoom
+- Performance with many notes
+
+---
+
+## Phase 3 — PDF Support
+
+### Objective
+
+Make PDFs a first-class media type with strong review utility.
+
+### Scope
+
+- Upload PDF files
+- Backend page rasterization
+- PDF metadata storage
+- Page thumbnails / previews
+- Place full doc card or individual pages on canvas
+- Comment on page/region
+- Basic PDF text extraction for search preparation
+
+### Product choice
+
+Start with **page-based ingestion**, not an embedded fully interactive PDF viewer on the canvas.
+
+Reason:
+
+- simpler scene model
+- easier comments
+- better performance predictability
+- closer to board workflows
+
+### Backend tasks
+
+1. Accept `application/pdf`
+2. Generate page previews
+3. Store doc metadata: page count, dimensions, extracted text status
+4. Add asset routing for preview pages
+
+### Frontend tasks
+
+1. Add PDF upload handling
+2. Add PDF card/import dialog
+3. Support dragging pages onto canvas as scene objects
+4. Allow comments pinned to page content
+
+### Acceptance criteria
+
+- User uploads a PDF successfully
+- User can place page previews on canvas
+- PDF pages persist and sync like other assets
+- Comments can be attached to a page object
+
+### Reviewer focus
+
+- Processing failures and user feedback
+- Storage layout and cleanup
+- Large PDF performance
+- Whether page objects and source document records stay consistent
+
+---
+
+## Phase 4 — Text System Overhaul
+
+### Objective
+
+Fix the current text element so it remains strong for labels while not pretending to be a note/doc system.
+
+### Scope
+
+- Better text box layout and wrapping
+- Alignment
+- Curated font set
+- Background/padding for label chips if needed
+- Improved resize semantics
+- Better zoom behavior
+
+### Explicit non-goal
+
+Do not turn `text` into markdown or sticky notes.
+
+### Acceptance criteria
+
+- Text boxes wrap predictably
+- Editing overlay aligns correctly under zoom and rotation
+- Font options are intentional, not random browser defaults
+- Labels remain lightweight
+
+### Reviewer focus
+
+- Text measurement drift
+- Serialization of layout fields
+- Rotation/edit overlay bugs
+- Cross-browser behavior
+
+---
+
+## Phase 5 — Markdown Cards
+
+### Objective
+
+Add richer document-like content while keeping the canvas uncluttered.
+
+### Scope
+
+- New `markdown` object
+- Canvas preview card with title/excerpt
+- Source edit mode
+- Render mode
+- Links, lists, headings, code blocks
+- Open-full-doc side panel
+
+### Product rule
+
+Canvas shows a compact preview, not the entire markdown body at all zoom levels.
+
+### Suggested fields
+
+```ts
+type MarkdownObject = {
+  id: string;
+  type: 'markdown';
+  x: number; y: number; w: number; h: number;
+  sx: number; sy: number; angle: number; z: number;
+  opacity: number; locked: boolean; visible: boolean; name: string;
+  title: string;
+  markdown: string;
+  previewMode: 'card' | 'expanded';
+};
+```
+
+### Acceptance criteria
+
+- Markdown card can be created and edited
+- Canvas preview is readable and compact
+- Full content opens in panel/modal
+- Save/load/sync works without corruption
+
+### Reviewer focus
+
+- Unsafe markdown rendering
+- Performance of live render
+- Whether the object should stay on-canvas or move to inspector sooner
+
+---
+
+## Phase 6 — Search and Retrieval
+
+### Objective
+
+Make mixed-media boards retrievable, not just visually navigable.
+
+### Scope
+
+- Search across board titles, filenames, notes, markdown, comments
+- PDF extracted text indexing
+- Later OCR for images
+- Filter by object type, author, tag, updated time
+
+### Acceptance criteria
+
+- A user can search content inside a board and find the right object
+- Results can jump to canvas context
+- Search does not require full reload of large boards
+
+### Reviewer focus
+
+- Query performance
+- Index strategy
+- Search result relevance vs complexity
+
+---
+
+## Phase 7 — Structure and Workflow
+
+### Objective
+
+Bring in selected Milanote-style organization features after the content core is solid.
+
+### Scope
+
+- Better frames/sections
+- Optional columns/stacks
+- Connectors
+- Task/checklist cards
+- Better share roles: view, comment, edit
+- Presentation / clean-share mode
+- Export improvements
+
+### Important constraint
+
+Do not add task/workflow objects until notes, docs, and comments are already stable.
+
+### Reviewer focus
+
+- Product sprawl
+- Interaction conflicts with selection/grouping
+- Share-mode permission correctness
+
+---
+
+## Cross-Cutting Technical Work
+
+### Scene format versioning
+
+- Extend scene schema conservatively
+- Keep old boards loadable
+- Add migration helpers where needed
+
+### Upload pipeline
+
+- Add clear status feedback for every media type
+- Unify queued/processing/ready/failed states
+
+### Performance
+
+- Avoid rendering full-detail content when zoomed out
+- Ensure many notes/pages/comments do not tank frame rate
+
+### Permissions
+
+- Formalize role behavior:
+  - `viewer`
+  - `commenter`
+  - `editor`
+  - `owner`
+
+### Export
+
+- Board export to image
+- Board export to PDF
+- Selection export
+
+---
+
+## Recommended Implementation Order
+
+1. Finish review/comments UX
+2. Add sticky notes
+3. Add PDF ingestion and page placement
+4. Overhaul plain text behavior
+5. Add markdown cards
+6. Add search and PDF text indexing
+7. Add structural/workflow features
+
+---
+
+## Suggested PR / Review Slices
+
+### Slice A
+
+- Review/comments gap audit and polish
+
+### Slice B
+
+- `sticky` scene object end-to-end
+
+### Slice C
+
+- PDF backend ingestion and metadata
+
+### Slice D
+
+- PDF page placement UI
+
+### Slice E
+
+- Text box overhaul
+
+### Slice F
+
+- Markdown card object
+
+### Slice G
+
+- Search and retrieval
+
+Each slice should be reviewable independently and should not mix unrelated feature families.
+
+---
+
+## Risks
+
+- Overloading the canvas with too many object semantics too early
+- Regressing existing image/video performance
+- Mixing review UI and content UI in confusing ways
+- Turning text into a catch-all object
+- Adding PDF support without strong processing/error handling
+
+---
+
+## Reviewer Checklist
+
+- Is the slice aligned with the current phase?
+- Does it preserve current board behavior?
+- Is the object schema explicit and future-safe?
+- Are save/load/sync semantics correct?
+- Are permissions and roles enforced?
+- Does the UI feel coherent at different zoom levels?
+- Is there a simpler implementation that preserves the same product outcome?
+
+---
+
+## Next Action
+
+Start with **Slice A / Phase 1**:
+
+- review the current comments implementation
+- enumerate gaps against the desired review workflow
+- close the highest-value UX and correctness issues before adding new content types