How the list is made, and how it stays current
Everything on this site follows from four rules and one score. This page states them exactly, so you can judge the list rather than trust it.
The four rules
| Rule | Why |
|---|---|
| A comment counts only if it has at least 100 characters of its own text besides the link. | Removes drive-by links and bots. |
| A video needs at least three different people linking it. | One person's enthusiasm is not a signal. |
| Those links must fall in at least three different months. | One viral thread does not qualify a video; coming back to it does. |
| Each comment is labeled by what the commenter was doing. | "Here is the talk everyone should watch" and "here is the clip that got debunked" are not the same link. |
The five labels
- Strongly recommends: enthusiastic, specific, tells you to watch it.
- Recommends: points people to it as worth their time.
- Cites it: uses it as a source or example, neutrally.
- Pushes back: criticizes or debunks it.
- Makes a joke: the link is the punchline.
Only the first two count as recommendations. About a quarter of the videos in the list were only ever cited, never recommended; they are kept because they are part of what HN talks about, and their pages say so plainly.
The score
score = number of distinct months with at least one recommendation + 0.5 × number of months with a strong recommendation
That is the whole formula. It is why a 2012 talk recommended in eighteen different months outranks a 2025 video with more total comments, and why the year strip on every page is the best single summary of a video's standing.
Labels were written by an AI model
Claude Sonnet 5.5 read each of the 39,834 comments with the title of the HN story it was under, and produced the label, a one-line "what you get" and a verbatim quote. The video summaries and the 150-topic taxonomy are also model-written. The author's own estimate is "a few percent" of errors. If a label on a video page looks wrong, the original comment is one click away, so you can check.
Keeping it current
The original HackerTube is a snapshot exported on 2026-10-31. This edition is built from that export plus a small pipeline that collects new HN comments linking YouTube videos after the export date (using the public Hacker News search API), applies the same four rules, and labels new mentions with the same five-way scheme. New videos appear in recent favorites and in their topics; nothing in the original list is removed. The refresh cadence and the exact process are documented in the project repository.
What is deliberately missing
- View counts and likes are not part of the ranking, by design. Where YouTube metadata is shown at all (duration, upload year), it is fetched separately and never redistributed.
- There are no accounts, so there is nothing to optimize for your engagement. "Because you liked" on the home page is computed in your browser from the public similar-video graph.
- Videos that only mattered around one news event (407 of them) are hidden from lists; their pages still exist.
The data and the license are described on the data page; the original site is hackertube.org.