Changed the domain root to a fixed page that reads the README introduction so that the latest‑date archive does not appear in search results like the site’s main page.
Organized the site‑name structured data, canonical tag, language‑specific hreflang, and favicon at the domain root to make the site’s source and identity clearer in search results.
Excluded date archives from the search index and sitemap, and modified individual posts so that search metadata is automatically generated from the title, category, and summary read from the document.
Added a search‑metadata audit script that checks all individual posts and the sitemap for nine languages without manually registering document paths.
Separated the Vercel production deployment from the self‑hosted Mac translation work and configured it to run in GitHub‑hosted Actions.
Configured main with code‑only changes to be deployed immediately, and when README or docs change, the multilingual translation pipeline builds and deploys the latest main automatically after finishing.
Re‑verified the SHA‑256 of the committed SQLite and the 1,570 original Markdown files, the three tables and FTS5 row counts, and inspected actual JSONL search using automatically generated queries from record contents.
All dynamic samples—60 docs_chunks, 60 api_mapping, and 2 label_prototypes—were re‑searched with BM25 rank 1; the single api_mapping entry with an incorrect SHA‑256 was excluded from search candidates.
Imported the 2d/hexagonal_map project from the MIT‑licensed godotengine/godot-demo-projects repository commit cae8dc567a56d3e7936f171bcb85f0ccb9634ad0, split its six actual files into 134 chunks, and tested SQLite JSONL search in the Source Flow Debugger GUI.
Verified that the “Manually editing project.godot” chunk in project settings, the 2D movement and CharacterBody2D chunk in movement, and the TileMapLayer chunk in scenes appear in the top results.
Improved the structure table to return records only when there is a matching source_api, target_api, or prototype reference, excluding records that only share negative patterns or decomposed common words.
The Qwen endpoint was temporarily unavailable, so this run only examined SQLite search results; re‑testing Qwen version detection will be done after the endpoint is restored.
Since record_embeddings currently has 0 entries, we verified SQLite lexical/BM25 search; the full F‑strategy end‑to‑end test that connects embedding, reranker, and Qwen validator will be performed after the actual index is created.