A local web app that displays your YouTube subscription videos sorted by a nowcast ranking algorithm tuned for ad hoc runs.

Videos are ranked by a core score designed for ad hoc runs and then filtered by the selected facet window (2 days, 1 week, 2 weeks, 1 month). This means each facet shows the strongest videos within that window, not a globally recency-biased list.

Core score components:

1. Nowcast vs Expected (55% weight)

   • Compares current views to age-adjusted expected views from each channel's baseline_48h.

2. Velocity Shock (20% weight)

   • Compares current views/hour to expected slope at the video's current age.

3. Subscriber Reach (15% weight)

   • Views relative to subscriber count with diminishing returns.

4. Duration Prior (5% weight)

   • Lightweight bias for durations that tend to produce stable performance.

Score modifiers:

• Confidence multiplier (0.75-1.05) lowers rank impact for weak parse confidence/stale baselines.
• Early breakout boost (up to +0.12) helps very new videos that are simultaneously strong in nowcast and velocity.

Notes:

• Facet sorting uses facet score keys (day, week, twoweeks, month) backed by the core score.
• Freshness diagnostics are still recorded in details for transparency, but recency is not applied as a hard rank penalty across wider windows.

A tool to track YouTube subscriptions and surface high-performing videos. It consists of:

1. A channel stats scraper that collects subscriber counts and a channel baseline_48h proxy
2. A video scraper that collects new videos from subscribed channels
3. Observation tracking per scrape run for better point-in-time ranking
4. A static page generator that creates a feed of videos sorted by performance

You can run the tool directly with uvx — no cloning and no manual installs:

uvx ytsubs@latest scrape-channels
uvx ytsubs@latest scrape-videos
uvx ytsubs@latest open

1. Requirements:

   • Python 3.12+
   • Google Chrome browser

2. (Optional) Install tool versions with mise:

mise install

3. Install dependencies with uv:

uv sync

If local schema/data gets out of sync, reset your local DB:

rm -f ~/.local/state/ytsubs/youtube.db

Run these commands in order:

uv run ytsubs scrape-channels   # When Chrome opens, log in to YouTube
uv run ytsubs scrape-videos     # Collect recent videos and generate the feed
uv run ytsubs open              # Open the feed in your browser

Your YouTube login is saved in ~/.local/state/ytsubs/chrome_profile (or $XDG_STATE_HOME/ytsubs/chrome_profile), so you'll only need to log in once. Subsequent runs will reuse this profile.

1. Collect video data:

uv run ytsubs scrape-videos  # Run daily to get new videos

2. Update channel statistics (subscriber counts and baseline views):

uv run ytsubs scrape-channels  # Run occasionally (e.g., monthly)

3. Open the feed:

uv run ytsubs open  # Opens the latest feed

The feed is written to ~/.local/state/ytsubs/ytsubs_feed.html (or $XDG_STATE_HOME/ytsubs/ytsubs_feed.html).

• Chrome profile: ~/.local/state/ytsubs/chrome_profile (or $XDG_STATE_HOME/ytsubs/chrome_profile)
• SQLite DB: ~/.local/state/ytsubs/youtube.db (or $XDG_STATE_HOME/ytsubs/youtube.db)
• Feed output: ~/.local/state/ytsubs/ytsubs_feed.html (or $XDG_STATE_HOME/ytsubs/ytsubs_feed.html)

uv run ytsubs debug-scrape --scrolls 4 --filter "gymkhana"

The project uses:

• SQLite for data storage
• Playwright for web scraping
• Preact for the frontend (served statically)

The project includes several helpful make commands:

• make clean: Remove Python cache files
• make reset-db: Reset database to empty tables
• make reset-videos: Clear only the videos table
• make reset-channels: Clear only the channels table

• Python 3.12+
• Google Chrome