The archive changes what you can ask
A current hot list answers a useful, narrow question: which titles are on the board when you look? The week’s archive lets you ask which titles recur, how their ranks differ between observations, and how many different titles mention the same subject.
For 7–13 September, the archived Weibo feed contains 34 collections of 50 rows, giving 1,700 observations. After removing the collector’s rank prefix and bracketed badges, there are 1,410 distinct titles. Of those, 1,169 occur in one sampled board, 208 in two, and 33 in three or more.
A single sampled appearance does not establish a story’s actual lifespan. Collection is intermittent, and a story can continue under different wording. It means an exact-title watchlist will see many isolated appearances even during a week with recurring subjects.
Most titles occur in one sampled board
Distinct normalized titles, grouped by number of archived appearances.
| Sampled appearances | Distinct titles | Share |
|---|---|---|
| One | 1,169 | 82.9% |
| Two | 208 | 14.8% |
| Three or more | 33 | 2.3% |
Source: China Snapshot’s frozen Weibo hot-search observations, supplied by TianAPI. Shares describe title recurrence in this sample, not audience size or time spent trending.
A brand query needs room for changing wording
Search the same export for the case-insensitive string iphone. It returns 65 distinct titles across 94 observations, present in 24 of the 34 sampled boards. One exact title, iPhone18Pro价格, appears six times. Tracking that title alone would leave most of the matching records outside the brief.
For a consumer-tech team, the useful next step is to read those matching titles together and separate product features, prices, availability and purchase questions. Keep the original title beside any assigned category so another researcher can challenge the grouping. The export supplies the material for that work; the keyword match itself does not establish what people believe or buy.
This query is easy to reproduce. It misses Chinese-only references and may include unrelated uses of the word. Add aliases when the question requires them, and record the additions so a wider search is not mistaken for an increase in attention.
The most frequently observed titles
Exact normalized wording; ties are sorted by title. Times are collection times in Beijing.
| Original title / search link | Appearances | Best rank | First observed | Last observed |
|---|---|---|---|---|
| 早春晴朗 | 9 | 2 | 7 Sep 10:34 | 13 Sep 14:02 |
| iPhone18Pro价格 | 6 | 1 | 10 Sep 06:01 | 11 Sep 06:01 |
| 兰香如故 | 5 | 4 | 11 Sep 22:03 | 13 Sep 06:00 |
| 建议大家把内裤袜子丢洗衣机洗 | 5 | 4 | 10 Sep 18:31 | 11 Sep 14:03 |
| 教师节 | 5 | 6 | 10 Sep 06:01 | 11 Sep 06:01 |
| 法考 | 4 | 2 | 12 Sep 10:32 | 13 Sep 18:31 |
| A股 | 3 | 5 | 8 Sep 18:33 | 11 Sep 10:32 |
| iPhone Duo | 3 | 24 | 9 Sep 18:31 | 10 Sep 06:01 |
These links open current topic searches. They are not original posts, preserved screenshots or proof of the claims in a title. “First observed” means first in this seven-day sample.
Method and coverage
The source is the Weibo hot-search feed in China Snapshot’s retained Git archive, frozen at revision bc8929f. The window starts at 00:00 Beijing time on 7 September and ends just before 00:00 on 14 September. The usual schedule has five collections a day. This archive has 34 of the expected 35; the 12 September 18:30 slot is absent. That gap is left empty.
The calculation strips only initial numbering and the final square-bracket badge, then counts exact strings. It removes duplicate titles within an identical collection timestamp, keeps repeated observations across collections, and does not merge synonyms or infer continuous presence between samples. Neither source publication times nor total views are available. No heat scores, demographic estimates or sentiment labels are used.
To reproduce the counts, group the CSV by topic_normalized and count distinct observed_at values. For the worked query, filter topic_normalized to strings containing iphone, ignoring case. The calculation script records the transformations; JSON records the frozen input revision and SHA-256 hash.
Use the evidence
All 1,700 observations behind this report, with original titles, normalized titles, sampled ranks, collection times and topic-search links.
JSON includes coverage notes and field meanings. Downloads contain metadata and numeric observations, with no full articles or post text.
Continue with the attention data browser, or discuss a monitoring brief for a brand or business question.