Open a hot list at breakfast, then return in the afternoon. How much of the first screen is still there? We tested that six-hour question across a complete quarter instead of relying on memory.
For each source, we kept the first non-empty snapshot in an observed clock hour and compared its first 20 item IDs with the first 20 exactly six hours later. The metric is deliberately plain: the number of earlier items still present, divided by 20. The median result ranged from 5% on Weibo to 85% on Hacker News Best.
Median share of the top 20 retained six hours later
What retention measures—and what it misses
Retention captures list memory. A high value means much of the visible set survived a six-hour interval. It does not mean the conversation stayed equally active, that ranks stayed fixed inside the interval, or that the same real-world event was not reworded under a new item ID. A low value means identifiers turned over; it is not proof that users have shorter attention spans.
| Source | Median retained share | Interpretation |
|---|---|---|
| Hacker News Best | 85% | More carry-over |
| Tieba | 75% | More carry-over |
| V2EX | 75% | More carry-over |
| Zhihu | 60% | More carry-over |
| Hacker News Top | 40% | Mixed |
| Toutiao | 20% | Faster turnover |
| Baidu | 15% | Faster turnover |
| 5% | Faster turnover |
The hourly sample is a control for collection cadence. Without it, a source captured more often would contribute many more comparisons and could dominate the result. We use one observation per hour and require an exact six-hour partner. An unobserved partner is excluded rather than guessed.
Why the result is useful to editors
Fast-turnover lists reward quick verification. If most of a list disappears during a working day, a newsroom should save the source page, capture primary evidence, and avoid assuming the public will remember the original context. Slower lists permit a second pass: the editor can look for updated answers, corrections, and practical follow-up instead of racing the first spike.
Retention also changes how a daily recap should be built. On a sticky feed, counting every capture can overstate a few long-running items. On a fast feed, selecting only the final snapshot can erase most of the day's movement. The better method records both arrival and persistence.
One metric should not become a personality test
It is tempting to turn the chart into a story about entire communities: serious versus frivolous, patient versus distracted. The data cannot support that. Ranking formulas, list depth, moderation, editorial choices, event calendars, and our sampling window all affect retention. The honest result is narrower: these saved top-20 lists had measurably different six-hour carry-over in Q1 2026.
That narrower claim is still valuable. It tells readers how much confidence to place in a single screenshot and how often an archive needs to sample a source to preserve meaningful movement.