About PropagandaScope

Data as of 8 November 2026

What these figures are

Every figure here comes from 27 Chinese newspapers, over the days on which each was collected; the data are as of 8 November 2026. These notes say what a figure counts and where it cannot see. A figure should always be given with its unit, its papers and its window.

What is counted

• Articles. The main unit is the article. A figure says how many articles contain a term: its matching articles. An article matches when the term stands in its title or its text, exactly as typed: no variants, no word splitting, no translation.
• Shares per 1,000. A share is given per 1,000 articles of the same papers over the same days, with both numbers beside it.
• Occurrences are a second figure under their own name: how many times the term occurs in titles and texts. Article counts and occurrence counts are labelled separately.
• Each article counts once (see Duplicates). 1,328,674 articles of the 27 papers count; 20,994 of them have no stored text and can match on their title only. Papers added since 13 October 2026 are kept apart (see the last section).

Which papers, over which days

Each paper was collected over its own period. A paper's figures cover only its period, and only the days in it that were collected.

• People's Daily (人民日报) is counted from 1 December 2024. Before that it is never included, and no comparison with it exists there: such figures say "no baseline".
• 河南日报 and 海南日报 have a few days of 2012 stored, apart from their period, which starts on 2 October 2020.
• 宁夏日报 has delivered nothing since 26 June 2026: its figures stop on the day before.
• 云南日报 has delivered nothing since 27 May 2026: its figures stop on the day before.
• Days not collected inside the papers' periods: 276 paper-days. Stretches of 7 days or more:
• 云南日报: 2025-07-31 to 2025-08-27 (28 days)
• 云南日报: 2025-07-11 to 2025-07-29 (19 days)
• 陕西日报: 2026-10-21 to 2026-11-08 (19 days)
• 江西日报: 2026-10-23 to 2026-11-08 (17 days)
• 云南日报: 2025-08-29 to 2025-09-13 (16 days)
• 云南日报: 2025-09-30 to 2025-10-09 (10 days)
• 云南日报: 2025-09-18 to 2025-09-26 (9 days)

A day that was not collected is not a zero

Every paper has a state for every day. Only a covered day can say that a term was not found:

• covered: readable articles were stored; for a term, found or covered but not found;
• no edition: the paper did not publish (河北经济日报 at weekends, the Spring Festival days);
• collection failed: the day is inside the paper's period but nothing readable was stored;
• unusable: only material that is not an article was stored;
• not yet collected: inside the collection's last seven days, still to come;
• never included: before the paper's period.

A count over days that are not covered is unknown, not zero. A comparison uses only the days on which every paper compared is covered. A covered day without any stored text supports statements about titles only: 黑龙江日报 on 2023-06-17.

Articles still cut at 5,000 characters

Until 12 October 2026 the collection kept at most the first 5,000 characters of an article's text. 24,614 stored articles have exactly 5,000 characters: cut, unless the article happened to be exactly that long. Since 13 October 2026, 20,769 have been collected again in full from the papers' sites (the latest replacement on 8 November 2026), 19 turned out to be whole, and 653 cannot be recovered: their page is gone or shows another text, on two tries at least a week apart. 3,826 are still cut (3,564 of the articles that count, 0.3 %). Each of them is marked: cut = 1 in the members' table articles, and a note on every figure that includes one.

• A term that stands only after the cut is not seen. A count that includes cut articles is a confirmed count, with a note giving the number of cut articles that could hide a match.
• A cut text is replaced only by a longer text of the same article: its first 5,000 characters must be the stored ones (or the same without spacing, or, in the two garbled windows, the decoded ones). The cut text is kept.
• Articles collected from 13 October 2026 on are stored whole.
• Collecting the cut articles again runs every night, gently, and figures rise as it proceeds. What changed is recorded in the change notes.

Titles, texts and page numbers

• Site text removed. The papers' websites wrap titles and texts in their own words: the site's name, the date and the page in the title (内蒙古日报, 南方日报, 新疆日报, 海南日报, 湖南日报, 福建日报, 陕西日报); page headers, navigation and lists of the page's other titles around the text (南方日报, 四川日报, 大众日报, 重庆日报), and some texts stored twice in a row. These words are removed before anything is counted: 238,935 titles and 124,310 texts of the articles that count, 27,541 of them stored twice. The stored rows keep them; the rules are in the data layer's description.
• Page numbers. The page is given as the paper's site gives it, and its meaning differs between papers. It is empty for every article before 25 March 2026, except in 吉林日报 (14 October 2025), 大众日报 (14 October 2025), 天津日报 (25 September 2025), 海南日报 (13 April 2023), 湖北日报 (14 October 2025); in 吉林日报, 河南日报, 重庆日报, 陕西日报 nearly every article is on page 1; 浙江日报 has no page 1. A page number is therefore not a front-page mark for every paper or period.

Duplicates

A paper's site often shows the same article under several addresses (for instance with and without a tracking parameter), and some articles were stored twice. Within one paper and one day's issue, the rows that are the same article are merged and the article counts once. Rows are the same article when they have:

• the same title and the same text (20,816 rows);
• the same text of at least 200 characters under another title, unless that text is standard page furniture repeated under three or more titles (226 rows);
• an empty text and the title of an article with text that day (17,116 rows).

Generic titles (such as 图片新闻, 广告, or any title of three characters or fewer) never merge rows on the title alone. 37,005 stored rows were merged into 27,932 articles; another 1,153 repeat a row that is not an article (below) and are left out with it. The row with the most text is kept (a row collected again in full comes first); the others go with it, with their addresses. Duplicates are judged on the text as first collected, so collecting a cut text again never splits or joins them.

• Never merged: the same text in two papers (each paper printed it: two articles), and the same text in one paper on two days (two printings).
• Left out entirely, not articles: 4,550 HTTP error pages stored as articles, 98 spam pages (西藏日报, 25 March 2024) and 12,090 page placeholders (新华日报, 海南日报).
• In the members' tables each article is one row (articles), with the number of its stored rows and their addresses; rows that are not articles are listed apart (left_out). Counting rows of articles counts each article once.

Pen names that have namesakes

A pen name is counted only where it stands as the byline: at the start of the stored text, at the end of a People's Daily title, or as a People's Daily page-one author in the layout data. Where a pen name is also a person's name or an ordinary word, that rule is what keeps them apart:

• 任平 is also the name of a professor at 苏州大学 who writes in 新华日报's theory pages: 3 of the 5 bylines found in the stored texts were his. 任平 is therefore counted only from People's Daily's page-one layout data, never from a byline in the text.
• 杨明 and 何振华 are named as pen names in other sources, but every byline with these names in this corpus is a person of that name: they are listed as not found in this corpus.
• 钟声, 和音 and 宣言 are also ordinary words (a bell's sound, harmony, a declaration): they are counted only as bylines, never as words in a text.
• A pen name that cannot be found reliably is listed as not found in this corpus, not as zero.

Windows and thresholds

• The latest day is one publication date for all papers: the latest day on which at least 50 % of the delivering papers are covered or published no edition. Papers not yet arrived are listed with their state.
• Rising and declining terms: the latest 7 days against the 28 days before, on the papers covered in both. The reference is each paper's earlier share times its recent articles, summed. A rising term needs at least 10 recent matching articles, at least 10 above the reference and at least 2 times it; a reference below 5 is listed apart as new or low baseline. "No mention" is none against a reference of at least 5; "below the recent baseline" is at most half the reference and at least 10 fewer. These lists state counts, nothing more.
• Comparison with People's Daily (amplification): over the days on which both papers are covered, the paper's share of matching articles divided by People's Daily's share, shown with both shares per 1,000 and their difference. There is none before 1 December 2024, none when People's Daily has fewer than 20 matching articles on those days, and none when its share is below 0.05 %.
• Rankings and maps across papers use one window, the 180 days ending on the data's date; a paper is ranked only when it is covered on at least 90 % of the window's days on which it published.
• First observation is the earliest day a term is seen in this corpus, not where it began. A first match with fewer than 30 covered days in the paper's year before it is marked "insufficient history"; a gap of more than 60 days before a second paper is reported as a gap.
• A title found in two papers is shown as "same title found"; no figure says where a term began or why.

The two garbled windows

人民日报 from 10 April 2025 to 4 June 2025 (4,735 articles) and 内蒙古日报 from 9 May 2025 to 28 October 2025 (3,871 articles) were stored with the wrong character set. They are read decoded: about 99.7 % of the characters come back. A character lost at collection shows as U+FFFD, and a term that spans one is not found, so counts in these windows are lower bounds.

Papers kept apart

Since 13 October 2026 the collection also stores papers in other languages and women's papers. They are kept apart: no figure here includes them, and they are not in the members' tables, until the owner decides how they are used.

• With text, collected every night: 西藏日报（藏文版） (Tibetan), 人民日报（藏文版） (Tibetan), 拉萨日报（藏文版） (Tibetan), 昌都报（藏文版） (Tibetan).
• On trial: 西藏法制报 (Chinese).
• Published only as page images or PDF, stored as files without text (no term can be counted in them): 锡林郭勒日报（蒙古文版） (Mongolian, minority-language), 今日女报 (Chinese, women).