Digest sources and editorial checks

Sources, topic boundaries and verification for the China and world economy, technology and research briefing.

The briefing covers the Chinese economy, the world economy, and technology and research. Each edition contains 5–7 stories: 2–3 economic stories and at least 3 technology or research stories. Chinese and English recordings each provide a complete edition. The list below defines the candidate sources; each edition links to the releases and reporting it actually uses.

Economic coverage prioritizes statistical agencies and central banks for original data, with reporting from Xinhua and BBC for company, consumption and industry developments. Technical sources include maintainer release notes, research blogs and developer articles. Community posts hosted by Hugging Face represent their authors, not platform certification or independent testing. Vendor performance claims are attributed to the vendor.

Sources checked: 2026-11-06. A configured source may have no eligible stories on a given day. Availability, freshness and topic checks still apply.

China and world economy sources

Source Type and use Latest check
国家统计局 Official statistical releases; reference periods and revisions checked Index reachable; 0 initially eligible
新华社财经 News-agency economic reporting; geography checked against the actual event Index reachable; 10 initially eligible
BBC News Business International public-service business reporting; full article required Index reachable; 2 initially eligible
U.S. Bureau of Economic Analysis Official statistical releases; reference periods and revisions checked Index reachable; 0 initially eligible
European Central Bank Central-bank banking/economic data; political and policy debates excluded Index reachable; 1 initially eligible

Technology and research sources

Source Type and use Latest check
Hugging Face Researcher/developer articles; not platform certification or independent testing Index reachable; 4 initially eligible
Google Developers Maintainer/vendor primary publication; performance claims attributed Index reachable; 0 initially eligible
GitHub Engineering Maintainer/vendor primary publication; performance claims attributed Index reachable; 0 initially eligible
Google Research Maintainer/vendor primary publication; performance claims attributed Index reachable; 1 initially eligible
Google AI Maintainer/vendor primary publication; performance claims attributed Index reachable; 1 initially eligible
OpenAI Maintainer/vendor primary publication; performance claims attributed Index reachable; 0 initially eligible
Ruff releases Maintainer/vendor primary publication; performance claims attributed Index reachable; 1 initially eligible
Zed releases Maintainer/vendor primary publication; performance claims attributed Index reachable; 1 initially eligible
uv releases Maintainer/vendor primary publication; performance claims attributed Index reachable; 0 initially eligible
Dify releases Maintainer/vendor primary publication; performance claims attributed Index reachable; 0 initially eligible
ComfyUI releases Maintainer/vendor primary publication; performance claims attributed Index reachable; 0 initially eligible
JupyterLab releases Maintainer/vendor primary publication; performance claims attributed Index reachable; 0 initially eligible
pandas releases Maintainer/vendor primary publication; performance claims attributed Index reachable; 1 initially eligible
scikit-learn releases Maintainer/vendor primary publication; performance claims attributed Index reachable; 0 initially eligible
Ollama releases Maintainer/vendor primary publication; performance claims attributed Index reachable; 0 initially eligible
vLLM releases Maintainer/vendor primary publication; performance claims attributed Index reachable; 0 initially eligible
Transformers releases Maintainer/vendor primary publication; performance claims attributed Index reachable; 0 initially eligible
Quarto releases Maintainer/vendor primary publication; performance claims attributed Unavailable in this check; not used

Reuters, the Financial Times and the IMF are not current automated sources: tested endpoints were unavailable, access-restricted or could not supply reliable article text. Paywalls are not bypassed, and headlines alone are not expanded into stories.

Selection and verification

  1. Source and freshness checks. Stories require an allowlisted domain, a publication timestamp and substantive evidence, normally from the preceding 48 hours. Economic stories require the article body. Headlines, short snippets, paywalls or failed downloads cannot substitute for evidence. Both domestic and international economic coverage must qualify; an overseas story published by a Chinese outlet does not count as domestic news.
  2. Whole-story exclusions. Titles, source text and both language editions are screened for politics, elections, conflict, sanctions, military affairs and territorial or religious disputes. A flagged or uncertain story is omitted entirely, rather than repackaged after removing sensitive sentences. An authoritative publisher does not automatically make a topic eligible.
  3. Facts and figures. Supporting quotations must match the original evidence. Economic checks cover reference periods, currencies, units, annual versus monthly changes, preliminary versus revised data and Chinese/English numerical magnitude conversions. A market movement does not establish a causal explanation.
  4. Factual and bilingual review. After drafting, a separate MiniMax request checks each story against its evidence for factual support, scope and translation equivalence. Every item must pass. A failed draft may receive one evidence-based correction, followed by another review.
  5. A separate publication review. Another MiniMax request then reads the sources and the entire bilingual edition, including titles, summaries, opening and closing. It checks sensitive context, misleading economic claims and programme framing, returning approve, hold or reject for every story and for the programme. It receives no earlier review verdict. All decisions must approve before audio synthesis and page installation. Uncertainty, missing checks, malformed responses or API failures hold publication. Changes to copy, evidence, policy, reviewer configuration or the corresponding review prompt invalidate previous approval records.
  6. Traceable records. Evidence, exclusion counts, drafts and both review reports are retained locally. A publication hold or rejection for unchanged material cannot be overwritten by automatic repeated requests; revised material or an explicit re-audit is required. Published editions link to the original releases or reporting and provide dates and transcripts. Economic coverage excludes stock recommendations, return promises and price forecasts.

These automated checks cannot guarantee error-free or risk-free coverage. Drafting, factual review and publication review currently use separate requests to the same MiniMax model; they are neither independent model providers nor a human editorial review. Errors can be checked against the linked evidence and corrected. Insufficient sources or failed checks delay publication; old news is not relabeled as fresh.

Back to the digest