Changelog

Dates, releases and what actually changed in the engine. This page is the source for every “last verified” stamp on the site.

v0.5.0 — 2026-10-25 · new tool: Bionic Reader

• New Bionic Reader tool (T15): structure-aware, attention-guided bolding — read text or Markdown in a split workspace (fixation strength, saccade interval, punctuation emphasis) with an A/B toggle
• Five bionic exports, reusing the same engines: bionic HTML (self-contained styled page), bionic PDF (selectable text), bionic Word (.docx with real bold runs — reported with the usual fidelity report), bionic Markdown (**…** per line) and a bionic PNG/JPG card
• Structure-safe transform: fenced and indented code, link URLs, headings and HTML markup are never bolded, and astral Unicode is never split
• Honest by design: the tool carries a visible “What the research says” block — controlled studies find no speed or comprehension gain, and no page claims otherwise; preference measured via the built-in A/B toggle

v0.4.0 — 2026-10-24 · parity round (⚠️ partial → ✅ native)

• Word task lists → checkbox content controls: click-to-toggle ☑/☐ inside Word, replacing glyph characters (Markdown to Word)
• Word blockquotes → the “Quote” paragraph style (indented, italic, left rule) instead of plain indents
• Word → Markdown: Quote / Intense Quote styles now become > blockquotes instead of plain paragraphs
• Word → Markdown: merged cells no longer shift columns — rowspan/colspan expand over the grid (covered cells empty, exactly how GFM + pandoc handle merges)
• Excel → Markdown: 0-padded number formats keep their zeros (e.g. 00000), and rich-text bold/italic runs map to Markdown emphasis inside cells
• CSV → Markdown: header-row toggle — when your file has no header row, the first row stays data (toolbar switch)
• Table generator: numeric columns auto-right-align on pasted cells, matching spreadsheet convention
• PDF math is now typeset: real sub/superscripts, √ and ∑/∫ glyphs with bounds instead of raw LaTeX source (Markdown to PDF)
• Markdown → Image: long documents now render — the card is measured block by block and split at element boundaries into numbered pages delivered as a ZIP, so browser canvas limits can no longer produce blank or failed images; rasterization is hardened end-to-end (blob URLs, forced decode, pixel-level blank checks, per-page error detail) (Markdown to Image)
• HTML export renders math natively: real <sup>/<sub>, stacked CSS fractions and overlined √ — the live preview shows the same thing, no KaTeX needed

v0.3.0 — 2026-10-24 · engine update

• Native Word equations (OMML): supported LaTeX — fractions, roots, powers and indices, \sum/\int/\lim with bounds, greek letters and common operators — becomes real, editable Word equations. Unsupported constructs stay as styled math text, flagged per instance (Markdown to Word)
• Real Word footnotes: superscript reference markers with definitions at the bottom of the page (replacing the old end-of-document Notes list)
• Real Word numbered lists: decimal numbering that restarts for each list (previously bullet glyphs)
• Embedded images, privacy-first: data: PNG/JPEG images embed directly into the Word and PDF outputs; remote/relative files are still never fetched — labelled placeholders remain
• PDF footnotes: collected into a “Notes” section at the end of the PDF with the same superscript numbering (Markdown to PDF)
• Front-matter → document metadata: YAML title/author/description/subject/keywords map to Word core properties, PDF document info and the HTML <title>/meta tags; unknown keys are listed in the fidelity report

v0.2.0 — October 2026

• New tools: Markdown to JSON (tables → clean JSON arrays with type inference and leading-zero preservation) and Markdown to Image (shareable PNG/JPG cards rendered locally at 2×)
• Word → Markdown: embedded images are now recovered into a ZIP with relative media/ references instead of being dropped
• Safer, layered download system with origin-checked host relay for browsers that block in-frame downloads
• Smarter download names: files take the document's title (front-matter, first heading or first line) instead of a generic document.docx — unicode titles included
• Live workspace re-architecture: conversions run in a Web Worker, so exports no longer freeze the page on large documents; the preview parses the document once per edit (debounced)

v0.1.0 — October 2026

• Initial launch: 14 tools across Documents, Tables, and Text & Web — Markdown to Word, PDF, Word to Markdown, Excel/CSV table conversions, HTML both directions, plain text, formatter, JSON and image
• Fidelity Report on every conversion: per-element native / approximated / dropped with a 0–100 score — including honest approximations like numbered lists rendering as bullets in Word
• Privacy architecture: separate engine origin, network-isolated converter, zero server-side document handling
• PDF output: selectable text, embedded Roboto subset (Latin/Greek/Cyrillic), repeating table headers, page numbers

What's next

• Wider equation coverage: LaTeX environments (aligned matrices), more macros, and Mermaid diagram rendering
• True per-page footnotes in the PDF exporter, and PDF bookmarks / document outline
• Tracked-change semantics beyond accept-all in Word → Markdown (revision marks have no Markdown concept)
• Letter/landscape page sizes, batch conversion and a bare-fragment HTML mode

Current engine version: v0.5.0, October 2026. The format support matrix documents per-element behavior for this version.