Where our numbers come from, and where they do not apply

Every page on this site rests on the same three measurements, described in what value density means. This page is the other half of that one: the places where those measurements stop applying.

It exists because the failures of a measurement are more useful than its successes, and because writing them down once is more honest than conceding them one argument at a time.

1. The library is not a sample of the platform

This is the limit that undercuts the most headlines, including some we have written.

The 3,486 videos we have measured are English-language explanatory and instructional material across 15 subjects, assembled to build a recommendation library. It was not drawn at random from YouTube and could not have been. It skews towards videos somebody thought were worth measuring, towards subjects a person might deliberately learn, and away from most of what the platform actually consists of.

So no figure here is a statement about YouTube. When a report says the median measured video carries a certain padding share, “measured” is carrying the whole sentence. A different library would produce a different number, and we have no way to tell you how different.

2. We do not check whether a video is true

The limit worth repeating loudest, and the one people most often assume we have crossed.

We measure how much a video says per minute. We do not measure whether what it says is correct. A well-structured, confidently argued, information-dense video about a claim that does not survive contact with the evidence will score well on every number we publish, and nothing in the product will flag it.

Density and correctness are different questions. Only one of them is one we answer.

3. Density is not quality, and low density is not failure

Information rate is the thing we measure, and information rate is not the same as worth.

A meditation video is not dense. A film essay whose purpose is atmosphere is not dense. A video that takes its time because the person watching needs it to is not dense. None of those is worse for it — they are simply not the kind of video this measurement was built for, which is why we do not publish shelves for the categories built on them. Printing a number that reads as a verdict, for a format where the number means nothing, is worse than printing nothing.

4. Padding is not waste

A 45% padding share does not mean 45% of the video was wasted. It means the same material could have been delivered in substantially less time.

Some of that restating is doing real work: a cue that genuinely needs saying twice, a recap that lets somebody who looked away rejoin, a definition repeated because the next section depends on it. Padding becomes a problem at the point where it stops being for you, and our number cannot see who you are.

5. The starting point is not a verdict on the setup

The second at which a video stops setting up and starts delivering is the most practical thing we publish and the easiest to misuse.

It does not mean everything before it is worthless. Context is often necessary and occasionally the best part of a video. It means a viewer who already knows why they clicked can start there and lose nothing — and for a video whose structure is its argument, skipping to it gets you the conclusion without the reason to believe it.

6. Not every reading covers every video

Some measurements are only defined for some material, and the sets do not line up.

Padding is only computed on explanatory and instructional video. Depth, novelty and shelf life are read on most but not all of the library. A curve across a video’s runtime needs a runtime the platform has confirmed. Every report on this site states the set its figures were computed over, and where that set is smaller than the library, the report says so and says why. If a figure does not name its denominator, treat it as unfinished.

7. Small groups are illustrations, not estimates

Several subjects in the library rest on fewer than a hundred videos, and inside a subject a runtime band can hold a few dozen.

A bar built on forty videos moves when four of them change. In the reports we mark those and say to read the ends of a chart rather than the middle, but the general rule is worth stating on its own: any single cell in any table here that rests on a small group is a picture of that group, not a measurement of the world. The most quotable line in our first padding census rested on ten videos, and when the group grew to over two hundred the figure fell by twenty points. The direction held. The number did not.

8. The numbers move, on purpose

Scores are recalculated as the measurement behind them improves. A figure you read last month may have moved.

That is why every live page carries the date its figures were last recalculated — currently 2 September 2026 — and why the dated reports do the opposite and freeze their numbers at the day they were taken. A report people cite should not change under them; a product page that contradicts the product is worse than a stale one. The two rules pull in opposite directions and each page says which one it is following.

9. It is a measurement of a video, not of a person

We publish measurements of individual videos and aggregates over a shelf or a subject.

We do not publish rankings of the worst channels, a verdict on anybody’s body of work, or a judgement of the people who make videos. A subject sitting high on a padding chart is a statement about the subject. A channel with one thin video is a channel that made one thin video, and we have no measurement that would support saying more than that.

How to disagree with us

Every ranking on this site names the measurement it used and the date it used it. That is deliberate, and it is the point of publishing the method at all: it lets you disagree on grounds other than taste.

If a video we called thin was the one that finally made something click for you, our number was wrong about you and right about the runtime, and those are compatible. The measurement is a claim about how much a video says per minute. It was never a claim about what it was worth to you.