Lohnt sich dein eigenes 1of10?

Build a local YouTube outlier tracker: a self-hosted dashboard that flags videos massively outperforming their own channel's baseline.

Stack, no substitutions:
- Python 3.11, FastAPI, Jinja2 templates, plain CSS, no JS framework
- SQLite via sqlite3 stdlib, file at ./data/outliers.db
- httpx for YouTube Data API v3 calls
- API key from .env (YOUTUBE_API_KEY), python-dotenv, ship .env.example, never commit .env

Data model:
- channels(id, handle, title, added_at)
- videos(id, channel_id, title, published_at, thumbnail_url, duration_seconds, is_short, view_count, fetched_at)
- snapshots(video_id, view_count, taken_at) for view velocity

Ingest (CLI: python -m app.sync):
- For each tracked channel, resolve the uploads playlist, page playlistItems for up to the 200 most recent videos, then batch videos.list (50 ids per call) for statistics and contentDetails
- Treat anything under 180 seconds as a Short and baseline Shorts separately from long-form
- Baseline = median view_count of that channel's previous 10 videos of the same format that are at least 30 days old; outlier_score = views / baseline
- Log estimated quota units used per run and stop cleanly before 9500

Web UI on localhost:8000:
- Table sorted by outlier_score: thumbnail, title, channel, published date, views, baseline, score, link to the video
- Filters: format (short/long), minimum score, published within N days, channel
- A form to add or remove a tracked channel by handle or URL
- A page per channel showing its videos and baseline curve as a simple inline SVG

In scope: sync CLI, dashboard, filters, CSV export of the current view.
Out of scope: user accounts, auth, hosted deployment, telemetry, payments, cross-channel keyword search, thumbnail downloads, AI title suggestions.

Include a README with setup, how to get an API key, quota math, and a cron line for a daily sync. Handle API errors and deleted videos without crashing the run.