Algorithm
Watchdog.
A daily-refreshed read on YouTube's Trending tab. Which video lengths are winning, when creators are posting, which categories are dominating, and which channels keep landing on the list. The algorithm is the most-discussed force in the creator economy. This page reads its public signal.
Refreshes every morning at 7am ET. Thirty days of rolling data. The posting-time heatmap is the visual centerpiece — it's the chart that tells you when to publish.
Loading…
What you're reading above is computed, not written. The cron pulls every morning, runs a handful of aggregations on the last seven days of data, and picks one of four templates based on which pattern is most newsworthy that week. A quiet week reads quiet. A category swing or a concentration spike rewrites the lede.
Histogram of trending video durations over the last 7 days. Below, how the median has shifted across the 30-day window — the answer to the "are short videos back?" question.
Day of week × hour of day (ET). Heatmap color is how often that slot shows up in trending across the 30-day window. The hot diagonal stripes are when creator agencies tell their talent to post.
Stacked bars showing the category breakdown of each day's Trending top 50, last 30 days. Sudden vertical bands tell you when one category took over.
Channels that appeared on Trending most often in the last 30 days. Concentration is a creator-economy bellwether — a few channels dominating means the algorithm is rewarding consistency over discovery.
| Channel | Appearances | Total views (est.) | Avg rank | Top category |
|---|
notebooks/watchdog_lab.py ↗ runs every morning via GitHub Actions ↗. It calls the YouTube Data API for the top-50 US Trending videos, appends a row, trims to the 30-day window, computes the lede aggregations, and commits trending.json back to the repo.
Four templates: standard, quiet, category shift, concentration spike. The cron picks the one whose threshold is tripped: any category share Δ > 12pp wins; otherwise concentration spikes (top-3 channels > 30% share); otherwise quiet (no movement); otherwise standard. No LLM in the loop, just deterministic rules over the data.
"Trending" is YouTube-curated, not pure algorithmic. It filters for advertiser-friendliness and applies undisclosed editorial weighting. The dashboard reflects what YouTube promotes, not necessarily what gets the most views. 30 days is short for trend confidence; some patterns visible here are weekly noise. Region=US only in v1. Shorts are on a separate Trending tab and aren't included.