Foreign-language items arrive already translated and summarized — original preserved.
How this works
This pipeline runs at ingest, not on demand: any observation whose lang isn’t English is routed through a translate transform (DeepL + MEMRI / Arabic dictionaries for Arabic terms), then summarized by an LLM into a neutral 2–3 sentence abstract. The entities the summarizer extracts are written back onto the observation, closing the loop with The Wall (01), One Query (02), and Web (04). The original text is stored alongside the rendering and is never overwritten — see the tabs below.