Outlier score thresholds: 2x, 3x or 5x?
2x the channel median is the most common working threshold for outlier detection: it is high enough to sit outside ordinary variance on most channels and low enough to surface several usable videos per week, while 3x and 5x are the precision settings analysts reach for on large or highly volatile channels.
- Across the industry, 2x the channel median is the most common working default; it clears ordinary variance without emptying the queue.
- Analysts commonly raise it to 3x-5x on large, frequent or volatile channels. If 2x flags more than a fifth of uploads, the bar is usually too low for that channel.
- Standard practice is to score Shorts and long-form against separate medians. A shared median produces false flags in both directions.
- The 2x-5x band tends to hold the most reproducible lessons; extreme multiples are usually luck or external circulation.
- These are conventions creators and analysts use by hand — not a description of any particular tool's configuration, Outlieo included.
This is general guidance on how creators and analysts set outlier thresholds, not a description of how Outlieo is configured. A threshold is a trade between two errors. Set it too low and normal week-to-week variance floods the queue with videos that prove nothing. Set it too high and you get a clean but nearly empty list, and you starve your research loop. The right number depends on how volatile the channel is, not on how impressive the multiple sounds.
Published July 29, 2026 · Updated July 29, 2026
How creators set a threshold by hand
- Pull the last 20 to 30 uploads of the channel, split by format.
- Take the median view count for each format separately.
- Score every upload against its own format median over a consistent view window, so early videos are not compared to settled ones.
- Start at the common 2x default and count how many uploads flag.
- If more than about a fifth flag, step up to 3x, then 5x, until the flag rate lands in the 5% to 20% band most analysts aim for.
- Re-check the median quarterly; channels drift, and a stale median silently miscalibrates every score.
A threshold is a false-positive budget
Every threshold sets an implicit budget for how many videos you are willing to study that turn out to teach nothing. Making the number explicit is what turns detection from a feeling into a process.
Judge the setting by output rather than by the multiple: a healthy threshold flags somewhere between one in twenty and one in five recent uploads on the channels you track.
- Too low: queue fills with normal weeks, and the pattern you extract is imaginary.
- Too high: queue empties, and the loop stalls for lack of input.
- Right: a handful of genuine over-performers per week, each with a visible packaging difference.
Volatility, not size, decides the number
Two channels of identical size can need different thresholds. What matters is the spread of their view counts: a channel whose uploads land between 0.8x and 1.3x of the median needs a lower bar than one that routinely swings between 0.4x and 2.5x.
If you track a small set of channels closely, set the threshold per channel rather than globally. The cost is a few minutes; the benefit is a queue that is actually informative.
Why the window matters as much as the multiple
A score computed on a three-day-old video is provisional, because distribution has not finished. Scoring on a fixed 7-to-14 day window makes videos comparable to each other and stops fast-starting videos from crowding out steadier over-performers.
Worked examples
Tuning a threshold on a volatile channel
A channel with a 60,000-view median publishes twice a week. At 2x, 11 of the last 30 uploads flag, which is over a third and clearly too loose. At 3x, four flag. Reviewing those four shows three share a single title construction, which is exactly the kind of repeatable pattern the queue exists to surface.
Using the threshold in a weekly loop
Run detection once a week against your tracked channels, review only what clears the threshold, and reverse-engineer the top three by packaging difference rather than by raw multiple.
Log the threshold you used alongside each flagged video. When a pattern later fails to reproduce, the log tells you whether the problem was the lesson or the calibration.
Frequently asked questions
Related reading inside Outlieo
Baselines, windows and the maths behind the flag.
The channel-relative definition the threshold is applied to.
Three measurements, three denominators, three different conclusions.
Ship a week of content, not a to-do list.
Track the outliers, script in your voice, and repurpose to LinkedIn. Scanning is free, no account needed.