The most unstable part of a ranking is often the part people screenshot: number one. During Q1 2026, the top item changed between consecutive observed hours at very different rates across TrendGoing's saved feeds. Weibo changed leaders in 49.1% of valid hour pairs; Hacker News Best changed in 7%.
This is not a comparison of audience fickleness. It is a comparison of recorded first positions after normalizing the archive to one leader per hour. The platforms use different inputs and ranking systems, and the collector does not see every update between two captures.
Share of consecutive observed hours with a different #1 item
Why normalize by hour?
A raw capture-to-capture calculation would mostly measure our scheduler. A feed saved every few minutes gets more opportunities to register a change than one saved every half hour. We therefore kept the first non-empty observation in each hour, sorted the hourly sequence, and compared only pairs separated by one or two hours. Longer gaps were not bridged.
| Source | Hourly #1 change rate | Distinct hourly leaders |
|---|---|---|
| 49.1% | 915 | |
| Hacker News Top | 39.8% | 669 |
| Toutiao | 37.8% | 710 |
| Baidu | 33.8% | 704 |
| Zhihu | 25.4% | 293 |
| Tieba | 19.3% | 383 |
| V2EX | 13.7% | 257 |
| Hacker News Best | 7% | 113 |
The distinct-leaders column adds scale. Two feeds can have the same change rate but a different number of items occupying first place. One might alternate between a small set; another might continuously introduce new leaders. We report both because “volatile” should not hide the shape of the movement.
A screenshot is a timestamp, not a verdict
When a publisher writes “the internet's top topic,” the claim usually outlives the evidence. A screenshot documents the first position at one moment on one feed. If that feed changes leaders frequently, the wording should include the observation time and avoid implying a day-long consensus.
For researchers, the practical response is to save rank and timestamp together. For editors, it is to revisit the list before publication. For readers, it is to treat number one as a location inside a moving system—not a direct measure of importance.
Ranking rules explain part of the difference
Baidu publicly describes an index combining search and information signals and says its hot list is updated frequently. Weibo describes a composite of search, posting, interaction, and reading signals. Hacker News exposes ordered lists of story IDs through its API. These are distinct instruments. The chart should be read as evidence about their captured output, not as a universal measure of attention.
Our next step would be to add semantic clustering so that reworded versions of one event can be treated as a single topic. This article does not do that. If a source assigns a new internal item to a revised headline, the method counts a new leader. Publishing that limitation is more useful than pretending the number is perfectly clean.