A GTA VI video with 400,000 views sounds like a hit. Whether it actually is depends entirely on whose channel it is on.
If that channel normally does 2 million, the video is running at 0.2× its own baseline — the subject underperformed, and the views came from the audience the channel already had. If it normally does 40,000, the same video is running at 10×, and something about that subject reached well beyond the usual audience.
Only the second is a signal. The first is just a big channel being big.
Why raw view counts mislead
Ranking topics by total views ranks them by who covered them. Large channels dominate every list, and the topics that surface are the ones already saturated by people with the most reach — which is exactly the opposite of useful, because those are the topics a smaller creator has the least chance of winning.
An outlier multiple removes channel size from the measurement. It asks a different question: did this subject perform beyond what this specific channel normally achieves? That question has the same meaning for a channel with 5,000 subscribers as for one with five million.
How a baseline is built
A channel's baseline is the median views its recent uploads had at a matching age. Age matching is the part that makes it honest — comparing a six-hour-old video against a six-month-old one measures nothing but time.
So baselines are computed per channel, per format (long-form and Shorts behave differently), and per age bucket: 1, 3, 6, 12, 24, 48, 72 and 168 hours. A video at six hours old is compared against what that channel's other videos had at six hours old.
Three qualifying uploads is the minimum. Below that we report insufficient_baseline
rather than a multiple, because two videos is not a baseline, it is an anecdote.
The consequence nobody likes
Baselines cannot be backfilled. YouTube exposes no historical view counts, so a video published last week can never tell us what it had at six hours old. We can only know that by having watched it at the time.
This means a channel we started tracking yesterday has no usable baseline for uploads published last month, and no amount of computation invents one. Baselines accumulate over days of observation and then keep improving.
The visible effect is that a young instance reports mostly UNSCORED, and that is the
system working correctly. The alternative — estimating a baseline from channel
subscriber counts, or from a global average — would produce a number that looks
exactly like a measured one and is not.
What a multiple is not
An outlier multiple is not a prediction. It says a subject outperformed a channel's norm; it does not promise it will outperform yours. Audience overlap, thumbnail, timing and format all sit between the two.
What it does reliably tell you is where attention went that cannot be explained by reach alone — and that is a far better starting point than a leaderboard of whoever happens to be biggest.