Docs + MCP: dark_mode export param, visual-identity docs, translated READMEs

README.md

@@ -182,6 +182,24 @@ the template for any multi-paper search. The zh-tw companion is at
     are i18n'd across **14 languages**: English, 繁體中文, 简体中文,
     日本語, Español, Français, Deutsch, 한국어, Português, Русский,
     Italiano, Tiếng Việt, हिन्दी, Bahasa Indonesia.
+  - **Designed-deck visual identity** (not the default Calibri-on-white
+    look): per-language typography (Inter for Latin, Microsoft JhengHei
+    UI / YaHei UI / Yu Gothic UI / Malgun Gothic / Nirmala UI for CJK
+    + Hindi), programmatic accent geometry (top accent bar on every
+    content slide + left band on the cover), academic-style table
+    formatting (default grid stripped, navy header rule, soft inter-row
+    dividers, alternating row stripe, middle-vertical alignment, bold
+    row labels), and a five-colour palette discipline (navy / teal /
+    grey / light / white) with red **banned** for text (use bold +
+    teal `#0E7490` for emphasis instead).
+  - **Dark mode is the default render path.** Build runs on the light
+    palette, then a post-build pass swaps text + fill + cell-border
+    RGBs to a dark deck (slide bg `#12151B`, body text `#E5E7EB`,
+    accent teal swapped to a brighter `#2DD4BF`). OLED projectors and
+    low-light venues get the dark deck without authors having to think
+    about it; pass `--light-mode` (CLI), uncheck **Light mode** (GUI
+    Deck tab), or `ExportOptions(dark_mode=False)` (programmatic) to
+    opt out for projectors in well-lit rooms or for printed handouts.
   - `.xlsx` — Papers sheet + Query provenance sheet, hyperlinked URL /
     PDF, frozen header, auto column widths. Column 5 (**Source**) shows
     the real publication venue (e.g. "IEEE Access"); column 6
@@ -226,6 +244,13 @@ the template for any multi-paper search. The zh-tw companion is at
   limit (token bucket), `defusedxml` for any XML payload,
   path-traversal-safe export paths, no `eval` / `exec` / `pickle` on
   user input.
+- **zh-tw / zh-cn vocabulary guard**: ~244 regex patterns in
+  `tests/test_i18n.py::test_zh_tw_files_use_traditional_chinese_vocabulary`
+  catch Simplified-Chinese loan words rendered with Traditional hanzi
+  (e.g. `內存` → `記憶體`, `魯棒性` → `穩健性`, `軟件` → `軟體`,
+  `緩存` → `快取`). Same guard runs in reverse for zh-cn locale
+  strings. Full rule + the regex catalogue live in
+  `.claude/agents/language-vocabulary-check.md`.
 
 ## Quick start
 
@@ -369,7 +394,7 @@ Tools:
 | `fetch_paper` | arXiv / DOI / PMID / IEEE identifier → single paper. |
 | `fetch_pdf_text` | Download one PDF, return extracted body text. **The MCP path to "I read the paper".** |
 | `download_pdfs` | Batch-download a papers list's PDFs into `{out_dir}/pdfs/`. Returns per-paper results keyed by BibTeX key. |
-| `export` | Papers list + formats → writes `.pptx/.xlsx/.md/.bib/.json`. Accepts a `summary` field per paper for the rich thesis-style schema and `max_slides_per_paper` (default 25). |
+| `export` | Papers list + formats → writes `.pptx/.xlsx/.md/.bib/.json`. Accepts a `summary` field per paper for the rich thesis-style schema, `max_slides_per_paper` (default 25), and `dark_mode` (default `true` — the project's dark-deck post-pass; pass `false` for the printable light variant). |
 | `pptx_inspect` | Read slide / shape structure of an existing deck. |
 | `pptx_update_slide` | Replace `title` / `body` / `meta` (by shape name) or arbitrary shapes by index. |
 | `pptx_delete_slide` | Remove a slide and its part relationship. |
@@ -385,7 +410,7 @@ LLM-as-agent flow (no `ANTHROPIC_API_KEY` needed — the LLM is the agent):
 4. fetch_pdf_text(pdf_url=paper.pdf_url)           # per paper
 5. (the LLM reads body text, produces a structured `summary` dict)
 6. export(papers=[{...paper, "summary": {pain_points: [...], rq_results: [...]}}],
-          language="zh-tw", formats=["pptx","bib"], ...)
+          language="zh-tw", formats=["pptx","bib"], dark_mode=true, ...)
 ```
 
 Full reference in [`docs/mcp.md`](docs/mcp.md).
@@ -433,11 +458,13 @@ pip install autopapertoppt[gui]
 autopapertoppt-gui                 # or: autopapertoppt gui
 ```
 
-The window has four tabs — **Search** (functional), **Settings**
-(functional, persists API keys via QSettings), **Enrich**, and
-**Deck** (the latter two land in a follow-up). The Windows release
-zip ships the Nuitka-compiled bundle with PySide6 included, so
-`autopapertoppt.exe gui` works without a separate Python install.
+The window has four tabs — **Search**, **Settings** (persists API keys
+via QSettings), **Enrich** (drives the LLM-as-agent / Python-pipeline
+enrichment over a `collection_ready` signal), and **Deck** (the Light
+mode toggle + slide-cap + max-figures controls flow through to
+`ExportOptions`). The Windows release zip ships the Nuitka-compiled
+bundle with PySide6 included, so `autopapertoppt.exe gui` works
+without a separate Python install.
 **UI ships in all 14 languages** (English, 繁體中文, 简体中文,
 日本語, Español, Français, Deutsch, 한국어, Português, Русский,
 Italiano, Tiếng Việt, हिन्दी, Bahasa Indonesia) — first run picks

autopapertoppt/mcp/server.py

@@ -14,12 +14,13 @@
   LLM agent can drive the PDF retrieval step before authoring rich
   summaries.
 - export(papers, keywords, formats, out_dir, filename_stem, include_abstract,
-         language, max_slides_per_paper) -> {written: {fmt: path}}
+         language, max_slides_per_paper, dark_mode) -> {written: {fmt: path}}
   formats may be any of: pptx, xlsx, md, bib, json
   papers[*].summary may include rich fields (pain_points, research_question,
   headline_metrics, technique_table, literature_table, method_sections,
   research_questions, rq_results, …) — when present, the PPT switches to
-  thesis-style layout.
+  thesis-style layout. ``dark_mode`` defaults to True (project default);
+  pass False for the light/printable variant.
 - pptx_inspect(path) -> {slides: [...]}
 - pptx_update_slide(path, slide_index, title?, body?, meta?, shape_updates?) -> {path}
 - pptx_delete_slide(path, slide_index) -> {path}
@@ -284,19 +285,27 @@ def export(
         include_abstract: bool = True,
         language: str = "en",
         max_slides_per_paper: int | None = 25,
+        dark_mode: bool = True,
     ) -> dict[str, Any]:
         """Export a list of papers (from search / fetch_paper) to disk.
 
         Each paper dict may carry a ``summary`` field — when populated with
         the rich-tier shape (pain_points, research_question, headline_metrics,
         technique_table, literature_table, method_sections, research_questions,
         rq_results, …), the PPT exporter switches to thesis-style layout.
-        ``language`` accepts en / zh-tw / zh-cn / ja.
+        ``language`` accepts en / zh-tw / zh-cn / ja / es / fr / de / ko /
+        pt / ru / it / vi / hi / id.
 
         ``max_slides_per_paper`` caps the per-paper slide count after the
         priority-based trim (cover/references/contributions are kept,
         Q&A/figure slides drop first). Default 25; pass ``0`` (or
         ``None``) for unlimited.
+
+        ``dark_mode`` defaults to True — the post-build pass swaps the
+        brand palette to dark slide background (#12151B) + near-white
+        text (#E5E7EB) so OLED projectors and low-light venues don't
+        glare. Pass False for the light/printable variant (white slide
+        background + navy text).
         """
         if not papers:
             raise AutoPaperToPPTError("export requires at least one paper")
@@ -318,6 +327,7 @@ def export(
             include_abstract=include_abstract,
             language=language,
             max_slides_per_paper=slide_cap,
+            dark_mode=dark_mode,
         )
         written = export_collection(collection, options)
         return {

docs/architecture.md

@@ -375,6 +375,38 @@ how much info each paper carries:
 All three tiers share the same shape-naming convention so
 `pptx_edit.update_slide(..., title=...)` looks up shapes by name.
 
+### Post-build visual-identity passes
+
+After the chosen tier builds the deck on the light palette, three
+non-invasive walk-and-rewrite passes run before the file is saved:
+
+1. **Typography** (`_apply_typography(prs, language)`) — walks every
+   text run, writes `<a:latin typeface=…>` AND `<a:ea typeface=…>` on
+   the run's XML based on `_FONT_FAMILIES[language]`. Setting only
+   `run.font.name` (the Latin slot) leaves CJK glyphs in PowerPoint's
+   default East-Asian font; both slots matter.
+2. **Accent geometry** (`_decorate_with_accents(prs)`) — adds the
+   `accent_top` bar to every content slide and an `accent_left` band
+   to the cover. Both are full-width / full-height navy rectangles
+   the user never sees as separate shapes but instantly reads as
+   "this deck has an identity".
+3. **Dark-mode recolour** (`_apply_dark_mode(prs)`, runs when
+   `ExportOptions.dark_mode=True`, which is the default) — walks every
+   slide / shape / run / table cell and swaps light-palette RGBs to
+   their dark equivalents via `_LIGHT_TO_DARK_TEXT` + `_LIGHT_TO_DARK_FILL`
+   dicts. The slide background switches to `#12151B`; body text goes
+   to `#E5E7EB`; the teal accent (`#0E7490`) goes to a brighter
+   `#2DD4BF`. The pass is intentionally non-invasive: it doesn't
+   refactor the 100+ direct `_BRAND_*` constant references in the
+   builders, it just rewrites RGBs after the fact.
+
+The three passes ship with regression tests in
+`tests/test_exporters.py`: `test_pptx_default_is_dark_mode`,
+`test_pptx_dark_mode_has_no_invisible_runs` (no run is `rgb=None` or
+black), `test_pptx_dark_mode_no_light_text_on_light_fill` (no
+near-white text inside a near-white-filled callout), and
+`test_pptx_no_red_text_runs` (red `#C0392B` is banned for text).
+
 ## Why the design choices
 
 | Choice | Reason |

docs/en/index.rst

@@ -318,8 +318,59 @@ to ``en``.
 
 Every text box on every slide carries a semantic ``name`` (``title``,
 ``meta``, ``body``, ``subhead``, ``footer``, ``page_number``, ``kpi``,
-``rq_box``) so the editing tools can target shapes without depending
-on visual position. See :doc:`/pptx_editing`.
+``rq_box``, ``rq_question``, ``figure``, ``accent_top``,
+``accent_left``) so the editing tools can target shapes without
+depending on visual position. See :doc:`/pptx_editing`.
+
+Designed-deck visual identity
+^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+
+The exporter applies three non-invasive post-build passes so a deck
+doesn't look like generic ``add_slide()`` output:
+
+* **Per-language typography** — every text run gets both an
+  ``<a:latin>`` and ``<a:ea>`` typeface override. Latin defaults to
+  Inter; East-Asian to Microsoft JhengHei UI (zh-tw), YaHei UI
+  (zh-cn), Yu Gothic UI (ja), Malgun Gothic (ko); Nirmala UI handles
+  Devanagari (hi). Setting only the Latin slot leaves CJK glyphs in
+  PowerPoint's default East-Asian font — both slots matter.
+* **Accent geometry** — every content slide gets a thin navy top bar
+  (``accent_top``, full width); the cover slide also carries a left
+  vertical band (``accent_left``).
+* **Academic-style tables** — the default PowerPoint heavy black grid
+  is stripped; the header row gets a solid navy fill + 1.5pt navy
+  bottom rule; data rows alternate light blue / white with a 0.5pt
+  soft inter-row divider; vertical alignment is middle; the first
+  column is bold so row labels read as headers.
+
+Dark mode
+^^^^^^^^^
+
+Dark mode is the **default** render path. The post-build pass swaps
+the light palette to a dark deck (slide bg ``#12151B``, body text
+``#E5E7EB``, brighter teal accent ``#2DD4BF``) — designed for OLED
+projectors and low-light venues. Opt out per render with
+``--light-mode`` on the CLI, the **Light mode** checkbox on the
+GUI's Deck tab, or ``ExportOptions(dark_mode=False)`` in Python.
+Over MCP, pass ``dark_mode: false`` to the ``export`` tool.
+
+Red is **banned as a text colour** in both modes. The sanctioned
+emphasis colour is teal ``#0E7490`` (bold + teal for KPI values and
+RQ question callouts); grey is for captions / placeholders. Regression
+tests in ``tests/test_exporters.py`` pin every contract — no
+``rgb=None`` runs, no near-white text on near-white callouts, no
+``#C0392B`` text runs.
+
+zh-tw / zh-cn vocabulary guard
+^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+
+A 244-pattern regex catalogue in
+``tests/test_i18n.py::test_zh_tw_files_use_traditional_chinese_vocabulary``
+catches Simplified-Chinese loan words rendered in Traditional hanzi
+— e.g. ``內存`` (should be ``記憶體``), ``魯棒性`` (``穩健性``),
+``軟件`` (``軟體``), ``緩存`` (``快取``). The same guard runs in
+reverse for zh-cn strings. Full rule + the regex catalogue live in
+``.claude/agents/language-vocabulary-check.md``.
 
 ----
 
@@ -380,8 +431,10 @@ Tools at a glance:
    * - ``export``
      - Papers list + formats → writes ``.pptx/.xlsx/.md/.bib/.json``.
        Accepts a ``summary`` field per paper that can carry the
-       full thesis-style schema; accepts ``language`` for i18n and
-       ``max_slides_per_paper`` (default 25; pass ``0`` for unlimited).
+       full thesis-style schema; accepts ``language`` for i18n,
+       ``max_slides_per_paper`` (default 25; pass ``0`` for unlimited),
+       and ``dark_mode`` (default ``true`` — dark deck; pass ``false``
+       for the printable light variant).
    * - ``pptx_inspect``
      - Read slide / shape structure of an existing deck.
    * - ``pptx_update_slide``

docs/gui.md

@@ -36,9 +36,7 @@ install is needed.
 
 ## Tabs
 
-The window has four tabs. The first two are functional in this
-release; the other two are placeholders that will land in a
-follow-up.
+The window has four tabs. **All four are functional in this release.**
 
 ### Search
 
@@ -90,18 +88,49 @@ Empty values clear the corresponding env var. **Restart the app**
 to refresh fetcher singletons that cached the env value at
 construction time.
 
-### Enrich (coming soon)
-
-Per-paper PDF + LLM enrichment will land here. For now: set
-`ANTHROPIC_API_KEY` in **Settings**, run a search, then export —
-the CLI side auto-enriches each paper that has a downloadable PDF.
-
-### Deck (coming soon)
-
-PPTX inspector / editor. The MCP server already exposes
-`pptx_inspect` / `pptx_update_slide` / `pptx_reorder_slides` /
-`pptx_delete_slide` / `pptx_add_slide`; this tab will wire them
-behind a Qt list view + form panel.
+### Enrich
+
+Drives the per-paper PDF + LLM enrichment step. The Search tab emits
+a `collection_ready` signal as soon as the results table populates;
+the Enrich tab catches it and renders a per-paper row showing:
+
+- The paper's BibTeX key + title (clickable → opens the URL).
+- A **PDF** column — green tick when a `pdf_url` is on file (Unpaywall
+  / S2 / arXiv / CORE.ac.uk lifted it), grey dash otherwise.
+- An **Enrich** action that runs either the Python pipeline
+  (Anthropic API, needs `ANTHROPIC_API_KEY` in Settings) or surfaces
+  the LLM-as-agent path for runs without an API key.
+- A progress strip across the bottom — the worker reports the current
+  paper, elapsed seconds, and a per-paper success/failure tally.
+
+Authored summaries are merged back onto the in-memory `PaperCollection`,
+so a subsequent **Export** picks up the rich-tier layout automatically.
+
+### Deck
+
+The pre-export deck-shaping controls. Wires `ExportOptions` fields
+behind Qt widgets:
+
+- **Light mode** checkbox — unchecked (default) ships the dark deck
+  (slide bg `#12151B`, body text `#E5E7EB`, brighter teal accent
+  `#2DD4BF`). Tick it for the printable / well-lit-room variant
+  (white bg + navy text). Mirrors the CLI `--light-mode` flag.
+- **Max slides per paper** — integer spinner. Defaults to 25;
+  `0` means unlimited.
+- **Max figures per paper** — controls how many `figures=` entries the
+  rich-tier layout actually renders. The priority trim keeps the
+  cover / references / contributions slides intact and drops Q&A /
+  figure / paper-table slides first when the cap kicks in.
+- **Include abstract** checkbox — when off, the deck skips the
+  abstract slide entirely (useful for an "executive overview only"
+  variant).
+
+The PPTX inspector / editor surface (drives `pptx_inspect` /
+`pptx_update_slide` / `pptx_reorder_slides` / `pptx_delete_slide` /
+`pptx_add_slide`) is wired to the same shape names the exporter
+produces — `title`, `meta`, `body`, `kpi`, `rq_question`, `figure`,
+`accent_top`, `accent_left` — so the editor list view labels each
+slide by its semantic role rather than by raw shape index.
 
 ## i18n
 

docs/mcp.md

@@ -266,7 +266,8 @@ Render a papers list to any combination of `.pptx`, `.xlsx`, `.md`,
   "filename_stem": "attention",
   "include_abstract": true,
   "language": "zh-tw",
-  "max_slides_per_paper": 25
+  "max_slides_per_paper": 25,
+  "dark_mode": true
 }
 ```
 
@@ -275,6 +276,15 @@ after the priority-based trim — cover / references / contributions are
 kept first; Q&A / figure / paper-table slides drop first. Pass `0`
 (or omit the field) for unlimited.
 
+`dark_mode` (default `true`) toggles the post-build recolour pass.
+On: dark slide background (`#12151B`) + near-white body text (`#E5E7EB`)
++ darker table-row stripe — designed for OLED projectors and low-light
+venues. Off: the light/printable variant (white background + navy text
+`#1F3A66`). Both modes share the same builder pipeline — the dark pass
+runs over the rendered tree, so the agent doesn't pick layouts up-front.
+The teal accent (`#0E7490` → `#2DD4BF` in dark) marks KPI values and RQ
+question callouts; red is banned for text in both modes.
+
 Returns:
 
 ```json
@@ -303,7 +313,11 @@ sentence-bucketing on `paper.abstract`.
 `fr`, `de`, `ko`, `pt`, `ru`, `it`, `vi`, `hi`, `id`. Drives both the
 template strings (Agenda / References / "Paper N of M" / footer) and
 the suggested LLM output language when the agent fills in the summary.
-Anything outside that set falls back to `en` silently.
+Anything outside that set falls back to `en` silently. Each locale
+also drives the per-language typography pass (Inter for Latin, plus
+Microsoft JhengHei UI for zh-tw / YaHei UI for zh-cn / Yu Gothic UI
+for ja / Malgun Gothic for ko / Nirmala UI for hi) — replaces the
+PowerPoint Calibri default, which is the biggest "AI-generated" tell.
 
 ### `pptx_inspect`