Hacker News publishes both topstories and beststories. The names sound close enough to treat as substitutes. Our archive says that would discard a substantial part of each list.
We matched the snapshot nearest noon from both feeds on 89 Q1 days and compared the first 50 story IDs. The median daily overlap was 36%. The lowest observed day was 22%; the highest was 54%.
Average daily overlap between HN Top and HN Best, first 50 IDs
Two products, not two timestamps
The official Hacker News API exposes Top and Best as separate endpoints. It does not describe Best as a delayed copy of Top, and our result is consistent with that distinction: many IDs are shared, but neither list reliably reconstructs the other. Someone studying discovery should keep the endpoint name alongside every observation.
| Date | Shared IDs | Overlap |
|---|---|---|
| 2026-01-31 | 11 | 22% |
| 2026-01-15 | 12 | 24% |
| 2026-02-16 | 26 | 52% |
| 2026-01-05 | 27 | 54% |
The four dates in the table are not anecdotes selected from screenshots. They are the two measured extremes at each end of the daily distribution. Showing them makes the monthly averages less abstract and reminds us that overlap varies from day to day.
What overlap can tell us
A shared ID proves that the same Hacker News item appeared in both captured top-50 lists near the sampling time. It does not prove equal rank, equal exposure, or the reason the item was selected. An unshared ID may have appeared in the other feed earlier or later that day. This is a cross-section, not a complete event history.
For a weekly technology briefing, the two feeds serve different research roles. Top is useful for the current front-page set. Best can recover items favored by a different ranking path. Combining them and removing duplicate story IDs gives broader discovery; keeping them separate preserves the evidence needed to explain where an item was found.
Why we used IDs instead of headlines
Headlines can be edited, and two stories can link to the same underlying event. Hacker News item IDs are stable identifiers in the official data model, so they are the cleanest unit for this particular question. We did not fetch scores or comments for the article, and we do not infer quality from appearing in either list.
The next useful analysis would follow each shared item across rank positions and time. That requires a longitudinal model rather than one sample per day. We are keeping that as a separate question instead of squeezing a stronger claim out of a simple overlap table.