Optimize fragment rendering and homepage link generation

README.md

@@ -49,8 +49,8 @@ The repo includes Playwright visual snapshots for the shipped viewer surfaces:
 ## Status
 
 - Markdown, code, diff, CSV, and JSON all render in the static shell
-- Fragment transport supports `plain`, `lz`, `deflate`, `arx`, and `arx2`, with automatic shortest-fragment selection across available wire formats
-- The `arx` substitution dictionary is served at `/arx-dictionary.json`; the `arx2` tuple-envelope overlay is served at `/arx2-dictionary.json`
+- Fragment transport supports `plain`, `lz`, `deflate`, `arx`, `arx2`, and `arx3`, with automatic shortest-fragment selection across available wire formats
+- The `arx` substitution dictionary is served at `/arx-dictionary.json` with a pre-compressed `/arx-dictionary.json.br` variant; the `arx2` tuple-envelope overlay is served at `/arx2-dictionary.json` with a pre-compressed `/arx2-dictionary.json.br` variant; `arx3` reuses those proven bytes and optimizes for compact visible Unicode fragments
 - The viewer toolbar copies artifact bodies to the clipboard, downloads them as files, and (for markdown) supports browser print-to-PDF
 - Deployment target: static hosting, including Cloudflare Pages
 
@@ -60,7 +60,7 @@ The repo includes Playwright visual snapshots for the shipped viewer surfaces:
 - `code` - read-only CodeMirror view with line numbers, wrap toggle, syntax-tree-aware rainbow brackets, and maintained indentation markers
 - `diff` - review-style multi-file git patch viewer with unified and split modes
 - `csv` - parsed table view with sticky headers and horizontal overflow handling
-- `json` - lightweight read-only tree view plus raw code view, with graceful malformed JSON fallback
+- `json` - lightweight read-only tree view plus native raw source view, with graceful malformed JSON fallback
 
 ## Principles
 
@@ -126,6 +126,7 @@ npm run preview
 ```
 
 Set `NEXT_PUBLIC_BASE_PATH` before `npm run build` when you want to preview a subpath deployment locally.
+The local preview server auto-detects the generated base path from the build manifest, so the same `npm run preview` command works for root and subpath exports.
 
 Set `NEXT_PUBLIC_SITE_URL` to your public origin before production builds so `sitemap.xml` and metadata use the correct canonical origin (see `docs/deployment.md`).
 

docs/architecture.md

@@ -25,7 +25,7 @@ The static export also emits `sitemap.xml` at the site root (and under `NEXT_PUB
 - `code` - read-only CodeMirror view with syntax-aware rendering and code affordances
 - `diff` - review-style diff view with unified and split modes
 - `csv` - table-focused data grid built from parsed rows and dynamic columns
-- `json` - lightweight read-only tree view plus a raw CodeMirror view
+- `json` - lightweight read-only tree view plus a native raw source view
 
 The viewer shell now routes all five artifact kinds through dynamically imported client-only renderers so the landing shell stays light and static-host friendly.
 
@@ -60,12 +60,13 @@ That keeps the viewer static-hosting friendly while removing the brittle parts o
 
 ## Bundle tradeoffs
 
-The largest remaining deferred cost is still the diff renderer stack, primarily `@git-diff-view/*` and its highlighting internals. It remains because it still provides the best review-style UX for multi-file git patches, split/unified modes, and syntax-aware rendering with less product code than a bespoke replacement.
+The largest remaining deferred cost is still the diff renderer stack, primarily `@git-diff-view/*` and its highlighting internals. Its vendor stylesheet is served from `public/vendor/diff-view-pure.css` with a precompressed `.css.br` variant and injected only when a rich diff mounts, so the empty shell and non-diff artifacts do not pay that CSS cost. The stack remains because it still provides the best review-style UX for multi-file git patches, split/unified modes, and syntax-aware rendering with less product code than a bespoke replacement.
 
 The JSON and markdown paths are now substantially lighter because:
 
 - `vanilla-jsoneditor` was removed in favor of a lighter read-only tree view
 - `rehype-highlight` and its Highlight.js stack were removed
+- raw markdown and CSV views use native source blocks instead of mounting CodeMirror
 - CodeMirror language support now loads on demand per active language
 
 ## Diff choice
@@ -76,7 +77,7 @@ The JSON and markdown paths are now substantially lighter because:
 - split and unified views are built in
 - syntax highlighting and diff affordances are stronger out of the box for artifact viewing
 - individual file patches can be rendered as a sequence while preserving filenames and boundaries
-- CodeMirror remains the better fit for raw source and raw JSON views
+- CodeMirror remains the better fit for full source artifacts and markdown code fences
 
 `@codemirror/merge` stays a reasonable future option if the project ever needs a more editor-centric comparison workflow, but it is not the best default for shareable review artifacts.
 
@@ -95,12 +96,13 @@ The fragment protocol keeps the JSON envelope stable and treats compression stri
 - `plain` stores base64url-encoded JSON for compatibility and debugging
 - `lz` stores compressed JSON via `lz-string` when it produces a smaller fragment
 - `deflate` stores deflate-compressed UTF-8 JSON bytes when it outperforms other codecs
-- `arx` applies domain-dictionary substitution, brotli compression (quality 11), and binary-to-text encoding for best-in-class compression. Four wire shapes are candidates: base76 (ASCII, 77 fragment-safe chars), base64url (RFC 4648 `A-Za-z0-9-_` with a `B.` prefix for detection), base1k (Unicode, 1774 chars from U+00A1–U+07FF), and baseBMP (high-density Unicode, ~62k safe BMP code points from U+00A1–U+FFEF, ~15.92 bits/char). The async encoder tries all four and picks the shortest **transport** length (percent-encoded UTF-8 length for non-ASCII), so base64url can win over Unicode encodings on chat-style surfaces. baseBMP produces ~32% fewer characters than base1k and ~55% fewer than base76 for the same compressed bytes, achieving ~70% smaller fragments than deflate on typical payloads (~6.1x compression ratio for 8k markdown). Full pipeline timing is on the order of ~8–14ms for 8k payloads depending on the wire encoding. The substitution dictionary is served as a static file at `/arx-dictionary.json` so agents can fetch it for local compression; a pre-compressed `/arx-dictionary.json.br` variant is also available. The viewer loads the dictionary on startup and falls back to a built-in table if the fetch fails.
-- `arx2` keeps the arx compression stack but replaces the JSON envelope with a compact tuple envelope and applies `/arx2-dictionary.json` as an overlay before the shared arx dictionary. It uses `v1.arx2.<dictVersion>.<payload>` and decodes back to the standard envelope before validation/rendering.
+- `arx` applies domain-dictionary substitution, brotli compression (quality 11), and binary-to-text encoding for best-in-class compression. Four wire shapes are candidates: base76 (ASCII, 77 fragment-safe chars), base64url (RFC 4648 `A-Za-z0-9-_` with a `B.` prefix for detection), base1k (Unicode, 1774 chars from U+00A1–U+07FF), and baseBMP (high-density Unicode, ~62k safe BMP code points from U+00A1–U+FFEF, ~15.92 bits/char). The async encoder tries all four and picks the shortest **transport** length (percent-encoded UTF-8 length for non-ASCII), so base64url can win over Unicode encodings on chat-style surfaces. baseBMP produces ~32% fewer characters than base1k and ~55% fewer than base76 for the same compressed bytes, achieving ~70% smaller fragments than deflate on typical payloads (~6.1x compression ratio for 8k markdown). Full pipeline timing is on the order of ~8–14ms for 8k payloads depending on the wire encoding. The substitution dictionary is served as a static file at `/arx-dictionary.json` so agents can fetch it for local compression; a pre-compressed `/arx-dictionary.json.br` variant is also available. The viewer tries the pre-compressed dictionary first on default ARX-family loads, falls back to the JSON file, and only loads external dictionaries when an ARX/ARX2/ARX3 encode or decode path needs them.
+- `arx2` keeps the arx compression stack but replaces the JSON envelope with a compact tuple envelope and applies `/arx2-dictionary.json` as an overlay before the shared arx dictionary. The viewer tries `/arx2-dictionary.json.br` first for default overlay loads and falls back to JSON. It uses `v1.arx2.<dictVersion>.<payload>` and decodes back to the standard envelope before validation/rendering.
+- `arx3` uses the same tuple envelope, overlay dictionary, shared arx dictionary, and brotli bytes as arx2, then allows the dense baseBMP wire to win by decoded visible character length. This deliberately optimizes copyable visible URL length for trusted Unicode-preserving surfaces; it is not a stronger compressed-byte format than arx2.
 - packed wire mode (`p: 1`) shortens transport keys before compression, then unpacks back to the standard envelope during decode
-- automatic async codec selection tries `arx2 -> arx -> deflate -> lz -> plain`; arx compares packed + non-packed candidates, while arx2 uses its tuple envelope
+- automatic async codec selection tries `arx3 -> arx2 -> arx -> deflate -> lz -> plain`; arx compares packed + non-packed candidates, while arx2/arx3 use tuple envelopes
 - sync codec selection (used by examples and legacy paths) tries `deflate -> lz -> plain`
-- decode enforces both fragment length and decoded payload size ceilings before UI rendering; arx/arx2 Brotli decompression uses a streaming output cap before final JSON or tuple parsing
+- decode enforces both visible fragment length and decoded payload size ceilings before UI rendering; arx/arx2/arx3 Brotli decompression uses a streaming output cap before final JSON or tuple parsing
 - invalid bundle state is normalized or rejected before renderers mount
 
 ## Zero-retention boundaries

docs/dependency-notes.md

@@ -12,7 +12,6 @@
 - `@replit/codemirror-indentation-markers` - MIT
 - `@git-diff-view/*` - MIT
 - `papaparse` - MIT
-- `@tanstack/react-table` - MIT
 - `lz-string` - MIT
 - `fflate` - MIT
 - `brotli-wasm` - Apache-2.0
@@ -28,16 +27,19 @@
 ## Why these libraries
 
 - `react-markdown` plus `remark-gfm` plus `rehype-sanitize` covers the markdown path without introducing unsafe raw HTML by default.
-- CodeMirror handles raw source and raw JSON well because it is excellent at read-only code presentation.
+- `next` pins its nested `postcss` dependency to `8.5.14` via `package.json` overrides so Tailwind CSS v4's `postcss ^8.5.6` peer range is satisfied in the Next.js toolchain.
+- CodeMirror handles source artifacts and markdown code fences because it is excellent at read-only code presentation; JSON, markdown raw, and CSV raw views use lighter native source blocks.
 - `@replit/codemirror-indentation-markers` replaces custom indent-guide logic with a maintained CM6 extension.
-- `@git-diff-view/*` fits review-style diffs better than a generic merge editor for the current viewer.
-- `papaparse` plus `@tanstack/react-table` keeps CSV parsing and rendering readable without coupling to a heavyweight data-grid framework.
+- `@git-diff-view/*` fits review-style diffs better than a generic merge editor for the current viewer. Its pure CSS file is mirrored into `public/vendor/diff-view-pure.css` with a Brotli-compressed `public/vendor/diff-view-pure.css.br` copy by `npm run assets:compress`, and loaded only by the diff renderer; `tests/diff-style-asset.test.ts` keeps those assets in sync with the package copy.
+- `papaparse` handles CSV parsing; CSV rendering uses a native read-only table to avoid a data-grid dependency for the shipped static viewer.
 - `fflate` provides portable deflate/inflate support across iOS Safari and Android Chromium without relying on browser-specific compression streams.
-- `brotli-wasm` provides the arx/arx2 Brotli compression layer, including streaming decompression used to cap expanded output before allocating oversized decoded payloads.
+- `brotli-wasm` provides the arx/arx2/arx3 Brotli compression layer, including streaming decompression used to cap expanded output before allocating oversized decoded payloads.
 - `mermaid` renders diagram definitions (flowcharts, sequence diagrams, etc.) to SVG client-side. Dynamically imported within the markdown renderer so it does not affect initial bundle size.
 - `better-sqlite3` provides synchronous SQLite access for the optional self-hosted server mode. Only used by `selfhosted/` code and not bundled into the static frontend export.
 
 ## Notable removals
 
 - `rehype-highlight` was removed after review because markdown fences now reuse the CodeMirror viewer stack directly.
 - `vanilla-jsoneditor` was removed because its bundle cost was too high for the default JSON tree-view use case in a viewer-first product.
+- `clsx` and `tailwind-merge` were removed because the app only needed simple conditional string joining, and the merge runtime was being pulled into shared client chunks.
+- `next-themes` was removed because the static shell only needs to preserve the `theme` localStorage key and synchronize the `html.dark` class.

docs/deployment.md

@@ -33,7 +33,7 @@ NEXT_PUBLIC_BASE_PATH=/agent-render npm run build
 npm run preview
 ```
 
-Then serve `out/` under `/agent-render/` and open the sample fragment links from the landing page.
+The preview server reads the generated build manifest, serves `out/` under `/agent-render/`, and keeps root requests working for convenience. Open the sample fragment links from the landing page.
 
 The preview server intentionally preserves the fragment payload and does not rely on hash-based in-page navigation for diff files.