fix(translate): prevent content loss in long-form translation

CHANGELOG.md

@@ -0,0 +1,28 @@
+# Changelog
+
+All notable changes to this project will be documented in this file.
+
+The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/).
+
+## [Unreleased]
+
+### Added
+- **Targeted Notion Fetching:** Added the ability to fetch data for a single Notion page.
+- **Pre-Release Safety:** Added validation checks to ensure all translations (locales) are complete.
+
+### Changed
+- **Simplified Data Fetching:** Cleaned up and simplified the logic for fetching all pages from Notion.
+- **Docker Tests:** Updated the Docker integration tests to work correctly with the newly added fetch-job types.
+
+### Removed
+- **Code Cleanup:** Removed redundant code from the API schemas to make the codebase cleaner.
+
+### Fixed
+- **Translation Completeness:** Fixed several issues with how the system measures if a page is fully translated.
+- **Long-form Content Translation:** Prevented issues where content could be lost when translating very long pages.
+- **Language Switcher (Locale Dropdown):** 
+  - Fixed a bug where the language switcher would sometimes point to the wrong page.
+  - Corrected an issue that caused "double" language codes in URLs.
+  - Fixed navigation issues when switching languages on category index pages.
+  - Fixed a display issue where the language dropdown might be hidden behind other menu items.
+- **Build Scripts:** Resolved bugs in the TypeScript compilation and Markdown parsing scripts.

scripts/constants.ts

@@ -182,10 +182,22 @@ export const ENGLISH_DIR_SAVE_ERROR =
 // Translation retry configuration
 export const TRANSLATION_MAX_RETRIES = 3;
 export const TRANSLATION_RETRY_BASE_DELAY_MS = 750;
-/** Max characters per translation chunk.
- *  Targets ~143K tokens (500K chars / 3.5 chars per token).
- *  Leaves generous buffer within OpenAI's 272K structured-output limit. */
-export const TRANSLATION_CHUNK_MAX_CHARS = 500_000;
+/**
+ * Reliability-oriented cap for proactive markdown translation chunking.
+ * This keeps long-form docs away from the model's theoretical context ceiling,
+ * even when the model advertises a much larger maximum context window.
+ */
+export const TRANSLATION_CHUNK_MAX_CHARS = 120_000;
+/** Smallest total-budget chunk size used when retrying incomplete translations. */
+export const TRANSLATION_MIN_CHUNK_MAX_CHARS = 8_000;
+/**
+ * Maximum times to retry with smaller chunks after completeness checks fail.
+ * Each retry halves the chunk limit. Starting from 120 K chars:
+ *   120k → 60k → 30k → 15k → 8k (floor)
+ * Four halvings are needed to descend from the default cap to the 8k floor,
+ * so this must be at least 4.
+ */
+export const TRANSLATION_COMPLETENESS_MAX_RETRIES = 4;
 
 // URL handling
 export const INVALID_URL_PLACEHOLDER =