Two tools can score the same video 23× and 1.7×. The difference is not the maths, it is which videos you compare it with and whether you use the median or the average.
VideoCore’s multiple is a video’s views divided by the median views of the 10 uploads nearest it in time on the same channel: the five before it and the five after, sliding to the 10 nearest at either end of the channel’s history. Shorts are compared with Shorts and long-form with long-form, and a video under 48 hours old gets no number yet.
Some tools use a different rule: the average of the channel’s last five videos. It is quicker to compute, and it behaves very differently as soon as a channel has had a hit.
The video was published on 14 October 2026 on one long-form channel we track. It had 352,447 views when we read it on 23 October 2026. Here are the videos each rule compares it with.
| Rule | What it divides by | Result |
|---|---|---|
| Median of the 10 nearest uploads | 15,266 views (the middle value) | 23.1× |
| Average of the last five uploads | 206,999 views | 1.7× |
The last five uploads before it had 937,448, 18,160, 9,585, 19,441 and 50,359 views. One of them, a 937,448-view hit five weeks earlier, drags the average up to 206,999. Four of those five videos got under 51,000 views, so the average describes none of them. The median of the 10 nearest uploads is 15,266, which is what a normal video on this channel looked like around then (for the newest video on a channel there is nothing after it, so the window is its 10 previous uploads). Against that, 352,447 views is about 23 times normal.
Both numbers are arithmetically correct. Only one answers the question people ask: did this video beat what this channel normally gets? It did, by a lot.
That 937,448-view video was published on 8 September 2026. Here is how the four uploads after it read under each rule.
| Published | Views | Median of 10 nearest | Average of last five |
|---|---|---|---|
| 11 Sep | 18,160 | 1.1× | 0.09× |
| 12 Sep | 9,585 | 0.5× | 0.05× |
| 16 Sep | 19,441 | 1.3× | 0.10× |
| 4 Oct | 50,359 | 3.3× | 0.25× |
Under the last-five average, all four look like failures for as long as the hit stays inside the window, even the one that was 3.3× the channel’s normal. A trailing window only looks backwards, so one hit distorts the next five scores. That is how a mean lets one past hit hide the next outlier.
One viral video moves an average a long way and moves a median almost not at all. The outlier how-to shows the small-numbers version: 1k, 2k, 2k, 3k and 500k have an average of about 102,000 and a median of 2,000.
A channel’s level changes. Comparing a video with the uploads right beside it measures it against the channel as it was then, not against the channel’s lifetime or only its past. At the newest and oldest ends there are not five uploads on both sides, so the window slides to the 10 nearest it can find.
They are served in different feeds to different audiences, and a mixed list makes both kinds of score meaningless.
A day-old video is still being distributed. A number then compares an unfinished count with finished ones.
VideoCore Lens does this on every thumbnail while you browse YouTube. It is free and needs no account to see the scores.
VideoCore Lens scores each video against its own channel’s nearest uploads while you browse YouTube normally.
Start free with your channel Get VideoCore Lens — free Or use the web Outlier FinderIt divides a video's views by the median views of the 10 uploads nearest it in time on the same channel, five before and five after, sliding to the 10 nearest at the ends of the channel's history. Shorts are compared with Shorts, long-form with long-form, and a video under 48 hours old gets no number yet.
A median ignores one very large value; an average is ruled by it. In one real example, a video with 352,447 views scored 23.1× against the median of the 10 nearest uploads (15,266 views) but 1.7× against the average of the last five (206,999 views), because one earlier video had 937,448 views.
A trailing window only looks at earlier videos, so one hit stays inside it for the next five uploads. Four videos published after a 937,448-view hit read between 0.05× and 0.25× against the average of the last five, although they were 0.5× to 3.3× against the median of the 10 nearest uploads.
It measures the video against the channel's level at that time rather than only its past. At the newest and oldest ends of a channel the window slides to the 10 nearest uploads available.
They are distributed in different feeds to different audiences, so mixing them makes the baseline meaningless for both.
Under 48 hours a video is still being distributed, so comparing its count with videos that have had months to collect views is not a fair test.
No. It is a choice that is wide enough to ignore one hit and narrow enough to follow the channel's level. A different size changes the numbers a little. A channel with fewer than three videos in the comparison set gets no number.