Every number
needs some context.
An outlier is a video that performs unusually well compared with its own channel. A million views alone doesn’t tell the whole story.
620,000 ÷ 25,000 = 24.8×
A median, not an average
A single previous hit can shift an average dramatically. The median represents the middle of the sample and is less sensitive to extremes. The target video is excluded from its own baseline.
The same age
A three-day-old video should be compared with other videos at day three. Real comparisons require historical snapshots. Without that history, an age-matched score will be unavailable.
Comparable videos
The live baseline requires at least 10 comparable prior uploads, using up to 50 from the same channel and known format. Shorts and long-form are separate. Missing format or history means no public outlier. Demo fixtures illustrate the interface and use fictional metrics.
Velocity and freshness
Views per hour is the observed increase between two snapshots, divided by elapsed hours. The live policy selects a 1–48 hour window closest to 24 hours. Missing or decreasing counts produce no velocity. Cards require observations no older than 48 hours.
Only confirmed outliers
The initial live policy requires at least 3× the comparable channel median, 1,000 video views and a baseline of at least 100 views. Targets are 24 hours to 90 days old and no longer than 30 minutes. Age matching allows a difference of up to 20%, capped at 48 hours. These are provisional research thresholds, not a statistical significance test.
We retain ordinary videos privately to calculate baselines. Newly monitored channels need reliable format evidence and repeated observations before they can produce public results; present-day counters cannot reconstruct missing history.
One video can belong to more than one niche
Browse 30 areas, 124 groups and 551 niches. Classified videos have one primary niche and up to two additional niches for substantial subjects. Areas include every niche beneath them; selecting several can match any selection or require all of them.
Niches describe the video's subject. Video tags are separate metadata keywords, while Shorts and long-form describe its format. Tags come from YouTube metadata. A separate classification service assigns niches from available evidence; unprocessed or uncertain videos can remain unclassified.
Tag filters match any selected tag and combine with the other filters. A tag with multiple words is matched as a complete keyword. Case and whitespace are normalized for matching. Videos with missing or empty tags remain available when no tag filter is selected.
The niche catalog is an editorial candidate, version 1.0.0-rc.2. Empty niches mean the current catalog has no classified matches; they do not measure the size of a real YouTube audience.
Collected YouTube data, with clear limits
Video titles, descriptions, metadata tags and counters come from YouTube Data API. Thumbnails are stored in our European S3 bucket. Each card shows its collection time; this is a collected catalog, not a complete or real-time index of YouTube.
Search, filters, shareable URLs and saved collections work. Sign in to keep saved videos and your watchlist across devices. Our initial 100 monitored channels are scheduled every eight hours, with collection spread throughout the window. This is an English-heavy starting sample, not a representative industry index. Notifications and detailed content analysis are in development. Watching a channel adds it to your reference list, without notifications.
Public cards and basic metrics remain open. Transcript requests are available to subscribers when audio can be processed. Detailed growth history and advanced content analysis are in development; read about research tools and MCP.
A multiplier measures relative performance. It does not prove why a video grew or guarantee that a similar idea will succeed. A sufficient sample is not a statistical probability of success.
Back to the ideas