Roadmap

This is the product plan. Design notes live in the repo (roadmap_editor_documental_portable.md, roadmap_usabilidad_y_adaptadores.md). Status is honest against the code: [x] shipped · [~] partial · [ ] not started.

We are on 1.0.0-beta. A frozen 1.0.0 needs the items in Next, not a rewrite of the core.

Now (shipped in the beta)

Core. Versioned JSON AST, operations, normalize, canonical hash, history (document snapshot + ops), schema + validator.

Editor. contenteditable host, IME, paste allowlist, tables (cells, align, col/row resize, merge/split), columns, images (drop, resize, align), LaTeX subset, four themes.

Usability (track A — all [x]). Command palette and slash menu, selection bubble, variable/link popovers, find/replace, template preview, outline, image-as-product, page preview, autosave/restore hooks, keyboard/? sheet and narrow viewport.

Templates. 0, safe resolver, currency/date, repeat rows.

Export. exportDocument for PDF, ODT, and DOCX. PDF is the strongest path (layout, contiguous tables, PNG downscale). Office packages are product-complete at the API surface; visual parity vs PDF is still weaker.

Hosts. Vanilla / React / Vue / Svelte / Angular / Astro samples. No framework in the core package.

Ports. In-memory document repo + SQL migration; S3 SigV4 helpers. No pg driver and no live bucket in CI.

Next (toward 1.0.0)

Ordered so each slice is testable without expanding the AST. No new runtime npm deps. Optional entries stay split so the editor bundle never pulls export or LibreOffice.

1. velo-text/api-report (slots) — [x] map of variable / table / image / equation tags for the backend; dataFromSlotValues / assetsFromSlotValues. Front installs only editor-web / vanilla. Nothing gzip > 2 MB.
2. Office CI — open/save a golden ODT and DOCX with system LibreOffice on the runner (apt), not an npm WASM. Today we only have in-repo validators.
3. PDF / Office parity — tables, image align/size, columns, and equations in ODT/DOCX closer to the PDF boxes. SVG/WebP PNG fallback for old Word — our encoder, no extra suite.
4. PDF fonts — Standard-14 Helvetica/Symbol only. Optional embedded TTF behind a host-supplied font bytes port (license stays with the host; we do not ship font files).
5. Real PG / S3 in CI — optional job with pg and a fake S3 in the workflow, not in dependencies.
6. Color / fontSize in PDF — [x] AST marks (color, fontSizePt, bold/italic) map to /rg and Standard-14 faces in the same exportPdf path.
7. Perceptual goldens — [x] HTML snapshots plus CI raster diffs (pdftoppm / poppler-utils, not an npm dep): pnpm run test:pdf-pages.

Later (explicitly out of 1.0)

Item	Why it waits
CRDT / comments / track changes	Second source of truth; breaks the single-AST rule
Import arbitrary DOCX/ODT	Inverse of export; unbounded surface
OfficeMath / OMML	LaTeX subset is enough for v1
Native OS print / Word headers-footers	Page preview already uses document.page + layout
Extra themes, plugin marketplace, AI	Tokens + four presets are the contract
Puppeteer / LO-in-the-browser / fat PDF kits	Exceeds the ~2 MB gzip budget; export is our writers

How we take a slice

1. One user-visible path in the Playground.
2. At least Vanilla (or the adapter that owns the feature).
3. Export must not regress (exportDocument + smoke).
4. dependencies stays {}.
5. A short note on the matching guide page.

Source files stay ≤ 250 lines. New behavior goes through ports (Clock, IdGenerator, BinarySink, AssetResolver) — not window in src/core.