Effective 14 September 2026. Responsible operator: Tristan McInnis.
What we publish from
Records carry source links so you can inspect the evidence. A link does not establish that a publisher's claim is true. Coverage depends on the sources available to the collectors and can contain gaps. The collection has four main inputs:
- First-party official releases.Central ministries and regulators, the People's Bank of China, the National Bureau of Statistics, provincial governments, and the governments of Hong Kong, Macao and Taiwan, read from their own publication pages.
- Economic and market data. Direct feeds from institutions including ChinaBond, CFETS, HKEX, SGX and PBOC, alongside series obtained through EastMoney and Yahoo Finance. The provider, observation date and collection date matter when assessing a number.
- Institutional press. Party organs, domestic financial and industry media, English-facing state outlets, and Western press with a China desk. Some headlines are collected through Google News RSS. A press report is distinct from a first-party release.
- Platform attention. The Baidu, Weibo and WeChat hot lists, supplied by TianAPI. These are sampled platform rankings, not measures of total reach or representative public opinion.
How to read the evidence
- A collected headline records what a source published or what appeared on a hot list. It is not independent verification of that headline. Check the original before repeating its claims.
- Summaries, extracted text and translations are labelled separately from the publisher's original document. Public access to an extraction does not grant permission to republish it.
- Publication dates, collection times and a number's reference period can differ. Repeated archive observations are not necessarily new events or new data releases.
- Comparisons describe the observed record. An unusual move or two events occurring together does not establish a cause or a forecast. The site provides information, not professional advice.
How automated editions are produced and checked
Collection is scheduled five times a day on Beijing time. Code ranks stories using source appearances, recency and significance rules. A language model drafts the daily reading from the ranked records and available market or baseline data. Its instructions require cited evidence and distinguish observation from interpretation.
Code checks look for problems including missing or invalid citations, unsupported numbers, repeated sentences and generic wording. A model review also assesses the draft. Failed story annotations or narrative passages can be withheld or replaced with a summary of source facts. These checks reduce some errors; they do not verify every sentence or replace a person reading the sources.
Original research and worked examples
The research pages use collected records to answer a stated question. Each example explains the selection, calculations and coverage limits, cites its sources, and provides the records used in its analysis. Counts describe our observed sample; they should not be read as the size of a whole market or population.
These pages are prepared with AI assistance and identify their review status. A reproducible calculation supports the reported result; it does not by itself validate the interpretation. A page awaiting human editorial review stays ad-free.
Human accountability
Tristan McInnis is responsible for the site's research approach and corrections. This responsibility does not mean he personally reads and approves every automated edition or research page before publication. We identify a human review only when it has been recorded for that work. Automated code or model checks do not count as human review. The publisher is CHINA SNAPSHOT LIMITED, Hong Kong.
AI disclosure, in plain words
DeepSeek, operated from the People's Republic of China, produces machine translations and drafts briefing readings, “why it matters” lines, dossier prose and Ask answers from retrieved evidence. Public research examples also use AI assistance, as disclosed on those pages. Code calculates rankings, baselines and forecast resolution. Source numbers can still contain errors, and model-written explanations can misread them. Check material claims against the linked source.
Corrections
If you find an error, write to support@china-snapshot.com with the page address and what you believe is wrong. We check the claim against the cited source. If the source does not support the claim, we correct it or remove it. A correction is noted on the page it applies to, with the date, so a reader who saw the earlier version can see what changed. Archived editions are never silently rewritten; a correction to a frozen edition is added as a note. We reply on working days, normally within five.
Independence and advertising
The site uses Google AdSense integration on eligible public pages. Integration does not establish Google's approval or mean that an ad is being served. The dataset browser and new research pages awaiting human editorial review carry no ads. Word count or an automated quality check alone is not evidence of original editorial value.
Any displayed ad must be labelled as an advertisement. Advertisers cannot buy editorial coverage or influence story rankings. Professional members do not see advertising. Consulting enquiries are welcome through the contact page.
Questions about these standards go to contact@china-snapshot.com. See also the method page, the terms of use and the privacy notice.