Persan -> francais : le document traduit restait en lecture de droite a
gauche, heritee de la source. Trois causes :
- PDF : regression du renommage precedent — le choix d'alignement
testait la fonction is_rtl (toujours vraie) au lieu du parametre :
tout texte traduit s'alignait a droite, quelle que soit la cible.
Corrige + test epinglant l'alignement gauche pour une cible latine.
- Word : les marques RTL heritees (bidi, rtl, bidiVisual, notes et
commentaires compris) sont desormais retirees quand la cible est
latine ; retrait fait sur une liste figee (l'iterateur lxml sautait
des elements pendant la suppression).
- Excel : les feuilles heritees d'un affichage droite-a-gauche sont
remises en lecture gauche-a-droite pour une cible latine.
- PowerPoint : les attributs rtl herites sont retires pour une cible
latine (alignements visuels conserves).
3 tests de bout en bout nouveaux : document RTL traduit vers le
francais ressort en lecture gauche-a-droite dans les trois formats.
Le defaut le plus sournois du pipeline etait silencieux : de l'arabe
livre pour une cible persane (meme ecriture, mauvaise langue) passe
inapercu et part chez le lecteur. Desormais chaque lot traduit est
verifie par le detecteur d'ecritures, et chaque segment fautif est
redemande une fois au moteur avec une consigne renforcee (nom de la
langue + lettres specifiques, ex. persan پ چ ژ گ). La seconde tentative
ne remplace la premiere que si elle passe le meme controle.
- services/quality/script_detector.py : extraction d'un controle
script_issue() reutilisable ; detect_arabic_variant signale des
ormais un long texte en ecriture arabe sans aucune lettre specifique
de la langue cible (arabe pur livre pour du persan/ourdou/pachto)
- translators/segments.py : retry_wrong_script() + construction de la
consigne renforcee, sans jamais faire echouer le travail
- Word, Excel, PDF : branchement apres la memoire de traduction et les
validations humaines ; le texte inchange (chiffres, noms propres) ne
declenche jamais de retentative
- 14 tests nouveaux (tests/test_translators/test_script_retry.py)
- source unique RTL_LANGUAGES/is_rtl dans core/languages.py (fin des 3 copies)
- Word: bidi partout (corps, tableaux bidiVisual, notes/fin/commentaires,
zones de texte, 6 zones d'en-tetes/pieds), insertion OOXML ordonnee,
polices cs elargies aux 11 langues
- PowerPoint: alignements explicites preserves, notes du presentateur,
indice de police <a:cs> insert a sa place
- Excel: feuilles affichees de droite a gauche, feuilles graphiques ignorees
- PDF: faconnage bidi (arabic-reshaper + python-bidi), polices par ecriture
(arabe/hebreu), TTF enregistree pour le PDF recompose
- 46 tests nouveaux (tests/test_translators/test_rtl_layout.py), 253 au total
Foundations:
- TranslationSegment model + migration f7e8d9c0b1a2 (segments, workspaces,
workspace_members, glossaries.workspace_id)
- SegmentRecorder injected into all 4 translators: unique (source,
translation) pairs captured per job and persisted (best-effort)
- set_segment_overrides: human-reviewed translations applied verbatim on
rebuild — top priority over TM and provider, zero API calls
Review API (routes/review_routes.py):
- GET /translations/{id}/segments (owner or job token)
- PATCH /segments/{id} edit/approve — feeds the per-user TM so approved
translations are reused in later jobs
- POST /translations/{id}/rebuild — rebuild document with reviewed text
- GET/POST /translations/{id}/xliff — XLIFF 1.2 export/import (edited
segments export their reviewed text)
Review editor (frontend /dashboard/reviews/[jobId]):
- side-by-side source/translation table, inline edit, approve (single or
all), rebuild & download (auth blob), XLIFF export/import, 13 locales
- 'Relire et corriger' link on the translation-complete screen
Team workspaces (routes/workspace_routes.py + /dashboard/teams):
- Workspace/WorkspaceMember models, roles owner/admin/member
- create (Business plan), list with seat usage, invite by email with
seat-limit enforcement (Business=5, Enterprise unlimited), removal
- shared glossaries: workspace members can use a glossary shared to their
workspace (access check extended)
Tests: 1184 passed / 0 failed (11 new: recorder, overrides, docx
capture->rebuild e2e, XLIFF structure/escaping, seats, workspace CRUD,
shared glossary access)
Chart <a:t> elements (title, axis labels, series names) were collected and
sent to the LLM, but the apply loop never invoked their setters. Reason:
the apply loop iterated only text_elements[:sheet_name_offset], which
excluded the chart-text positions (sheet names were inserted in the
middle, pushing chart texts past sheet_name_offset). The LLM correctly
returned French for "Revenue by Product" / "Quantity Trend" / "Order #"
but the result was thrown away.
Fix: include the chart-text slice [sheet_name_offset + N .. total_texts]
in the apply loop. Setters for sheet names are still None, so they are
naturally skipped.
Add test_excel_chart_text_applied.py (2 end-to-end tests) using a fixed
provider that pre-translates every known chart text; the test asserts
all chart <a:t> values in the output .xlsx come from the FR table, not
the original English source.
Multiple translation bugs in Word/Excel/PPTX that caused chart elements
to be left untranslated or charts to render as empty series.
Backend
-------
* providers (deepseek/openai/minimax): tighten system prompt so the LLM
actually translates chart titles/axis labels/legend/category labels,
month abbreviations, and clarifies what counts as a 'real' proper noun
(people/place/company/product names) vs. technical labels. Old rule
'keep proper nouns unchanged' was being read too broadly by the model
and caused chart text to be skipped.
* excel_translator.py:
- Sheet reference rewrite: when a sheet is renamed, the chart XML's
c:f refs (e.g. 'Sales 2024'!$D$2:$D$61) are now rewritten to the
new name with proper apostrophe escaping (Chiffre d'affaires ->
'Chiffre d''affaires') and auto-quoting when the new name contains
spaces or special chars. Without this, the chart points at a sheet
that no longer exists and renders 0/empty series.
- Sheet name offset bug: sheet_name_offset was computed after chart
text was appended to text_elements, causing sheet names to receive
chart text translations. Now captured BEFORE sheet names are added.
* New tests:
- test_excel_chart_sheet_refs.py (10 unit tests, synthetic inputs)
- test_chart_translation_prompt.py (3 contract tests on the prompt)
Frontend
--------
* DashboardSidebar / translate/page: hide the Memento promo section
for paying users (tier != 'free').
* constants.ts: temporarily comment out the 'CLES API' nav item.
Update constants.test.ts to match the new state.
All fixes are generic - no file-specific hardcoding, edge cases covered
(empty mapping, missing bang, apostrophe escaping, partial renames,
multi-series).
Word fixes:
W1 — Fix hyperlink double-collect: a run inside <w:hyperlink> was
previously collected twice (once via paragraph.runs, once via
the manual hyperlink iter). Now uses a dedup set of element
ids to collect each run exactly once.
NB: python-docx 1.x's paragraph.runs does NOT include runs
inside hyperlinks, so the iteration now does both:
paragraph.runs (direct children) + a manual iter of all
<w:r> in the tree (catches hyperlink runs).
W2 — Fix footnotes import: used document.part.package.part_related_by
which doesn't exist in python-docx 1.x, so footnotes were never
collected. Now uses document.part.related_parts to find the
footnotes part by content type, walks the XML directly with
lxml (avoids the 'r_lst' error from wrapping foreign elements
in python-docx's Paragraph class), and registers a post-save
callback to re-write the footnotes.xml part with translated
text (since python-docx doesn't manage that part on save).
Same fix applied to endnotes.
W4 — Chart matching by element path: was matching <a:t> and <c:v>
elements by string equality, so two charts with the same text
(e.g. two 'Revenue' series) would only have the first one
translated. Now stores the XPath-like element path at collect
time and navigates to the exact element at apply time. Falls
back to string matching for legacy entries without a path.
Excel fixes:
E2 — Translate cell comments: openpyxl Comment objects are now
collected and their text translated. The Comment object is
replaced in place after translation.
E3 — Translate cell hyperlink display labels: cell.hyperlink.display
(or .target if no display) is collected and translated. The
URL itself is never sent for translation, so it remains
intact. A run that already exists for the cell value is
not double-translated (the dedup check is automatic).
E4 — Chart matching by element path: same fix as W4 but for
Excel. Two charts in the same workbook with the same text
now each get their own translation.
Tests:
Added tests/test_translators/test_b1_format_fixes.py with 11 tests
covering all the fixes. All 11 pass. Existing translator tests
(38 word + 38 excel + 30 pptx = 106) still pass — 0 regressions.
Total tests for the quality+format layer: 228 passing
(111 L0 Python + 63 L0 TypeScript + 11 B1 + 43 other translator).
All fixes are surgical: existing translation flow is preserved.
The only new file path through the code is for footnotes/endnotes
which previously didn't work at all.
Two critical fixes:
1. Provider "google" (default classic mode) now checks for a Google Cloud
API key (GOOGLE_CLOUD_API_KEY in env or admin settings). If present,
uses GoogleCloudTranslationProvider (official API). Previously it
always fell through to deep_translator (free scraper) which gets
blocked in production, silently returning untranslated text.
2. Added translation verification: each translator now tracks how many
texts were attempted vs actually changed. If 0 texts were translated,
the job is marked as FAILED with a clear error message instead of
returning the original file as "completed".
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
- Restructured docker-compose for Nginx Proxy Manager (no custom nginx)
- Added domain wordly.art configuration
- Added Prometheus + Grafana monitoring stack with pre-configured dashboards
- Added PostgreSQL backup script to NAS (daily/weekly/monthly rotation)
- Added alert rules for backend, system, and Docker metrics
- Updated deployment guide for NPM + IONOS DNS homelab setup
- Added marketing plan document
- PDF translator and watermark support
- Enhanced middleware, routes, and translator modules
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Major changes across backend, frontend, infrastructure:
- Provider system with model selection (Google, DeepL, OpenAI, Ollama, Google Cloud)
- Admin panel: user management, pricing, settings
- Glossary system with CSV import/export
- Subscription and tier quota management
- Security hardening (rate limiting, API key auth, path traversal fixes)
- Docker compose for dev, prod, and IONOS deployment
- Alembic migrations for new tables
- Frontend: dashboard, pricing page, landing page, i18n (en/fr)
- Test suite and verification scripts
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
