Here’s a slimmed-down, re-organized roadmap that groups related work into bigger milestones, highlights key deliverables (including a “demo” command and grep-style context snippets), and stages integrations for maximal developer velocity:

| Phase / Sprint                           | Timeline               | Focus & Rationale                                                                                         | Key Deliverables                                                             |
| ---------------------------------------- | ---------------------- | --------------------------------------------------------------------------------------------------------- | ---------------------------------------------------------------------------- |
| **Sprint α: Bedrock & Metadata Domains** | **2025-Q2 (now–6/01)** | Stabilize core schema, migrations, CI, and introduce the first metadata domains **with discoverability**. | - **CI:** `cargo test` + SQL migration dry-run coverage<br>- **Migrations:** |

* `links(src,dst,type)`
* `collections(name)+collection_files`
* `views(name,query)`<br>- **CLI stubs** for `marlin link|unlink|list|backlinks`, `marlin coll`, `marlin view`<br>- **`marlin demo`** command that walks you through a mini-drive-by tutorial of link/coll/view flows                                                                                               |
  \| **Epic 1: Scale & Reliability**                    | **2025-Q2**      | Prototype “dirty-row” FTS to avoid per-row triggers, benchmark at 100 k files, and shore up CI for edge cases.                                               | - Dirty-flag + `scan --dirty` reindex only changed rows<br>- Replace per-row triggers with periodic FTS rebuild<br>- End-to-end benchmarks on \~100 k files<br>- CI tests for dirty-scan edge-cases                                                                                                                                                                                                                           |
  \| **Epic 2: Live Mode & Self-Pruning Backups**       | **2025-Q2**      | Automate continuous indexing & backup hygiene so Marlin “just works” in a real workspace.                                                                   | - `marlin watch [dir]` via `notify` crate (create/modify/delete/rename)<br>- `backup --prune <N>` flag + post-scan auto-prune to latest N<br>- Daily/pr-merge auto-prune automation in CI                                                                                                                                                                                                                                    |
  \| **Phase 3: Content FTS & Annotations**             | **2025-Q3**      | Go beyond metadata—index file bodies, provide grep-style context, and add annotation support.                                                                | - New `files.content` column + migration<br>- Extend `files_fts` to include `content` and emit context snippets (`-C` style)<br>- `annotations` table + FTS triggers<br>- CLI: `marlin annotate add|list`                                                                                                                                                                                                                 |
  \| **Phase 4: Versioning & Deduplication**            | **2025-Q3**      | Enable history, diffs, and duplicate detection with content hashing.                                                                                         | - Compute & store `files.hash` (SHA256)<br>- `scan --rehash` to refresh hashes<br>- CLI: `marlin version diff <file>` to show changes between revisions                                                                                                                                                                                                                                                                   |
  \| **Phase 5: Tag Aliases & Semantic Enhancements**   | **2025-Q3**      | Tame tag sprawl and seed AI-powered suggestions via embeddings.                                                                                              | - Enforce `canonical_id` on `tags`; CLI: `marlin tag alias add|ls|rm`<br>- `embeddings` table + `scan --embed` to generate vectors<br>- CLI: `marlin tag suggest`, `marlin summary <file>`, `marlin similarity scan`                                                                                                                                                                                                       |
  \| **Phase 6: Search DSL v2 & Smart Views**           | **2025-Q4**      | Offer a robust query grammar and “virtual folders” powered by views.                                                                                        | - Swap ad-hoc parser for a `nom`-based grammar with `AND`, `OR`, parentheses, wildcards…<br>- CLI: `marlin view save|list|exec` with aliases and pagination                                                                                                                                                                                                                                                             |
  \| **Phase 7: Structured Workflows**                  | **2025-Q4**      | Unlock full task, state, reminder & event workflows directly on files.                                                                                      | - `templates` + `template_fields` + validation engine<br>- CLI:
* `marlin state set|transitions add|state log`
* `marlin task scan|task list`
* `marlin remind set <file> <ts> "<msg>"`
* `marlin event add <file> <date> "<desc>"` + `marlin timeline`                                                                                                                                                                                                                                |
  \| **Phase 8: Lightweight Integrations**              | **2026-Q1**      | Surface Marlin inside your editor/terminal before diving into a full GUI.                                                                                   | - **VSCode & terminal UI extension**: file-tree sidebar showing tags/attrs/links/annotations                                                                                                                                                                                                                                                                                                                              |
  \| **Phase 9: Dolphin Sidebar Plugin (MVP)**          | **2026-Q1**      | Prototype a read-only Qt sidebar for Linux file managers—peek metadata without leaving your file browser.                                                    | - Qt plugin showing tags, attributes, links, and annotations alongside files                                                                                                                                                                                                                                                                                                                                             |
  \| **Phase 10: Full Edit UI & Multi-Device Sync**     | **2026-Q2**      | Ship an in-place metadata editor and optional sync layer for distributed workflows.                                                                         | - Tag & view editors, task/reminder/event dialogs in GUI<br>- Choose/implement sync backend (rqlite, Litestream or custom) for optional read-only remote mounts                                                                                                                                                                                                                                                          |

---

### Why this order?

1. **Lock down core schema & domains** (links, collections, views) **with a “demo” helper** so users can explore right away.
2. **Scale & CI** unlocks safe indexing at volume, then
3. **Live mode & auto-prune** keep your index fresh without manual steps.
4. **Content FTS + annotations** builds on an efficient, reliable plumbing layer—you’ll love grep-style context snippets.
5. **Versioning & semantic layers** ride atop a stable full-text index and annotation system.
6. **Advanced queries & workflows** expand power users’ toolsets before branching into GUIs and sync.

This grouping ensures every new layer rests on a solid, tested foundation—maximizing both developer speed and user delight.