Filter non-watch YouTube Takeout activity

[AndrewSB]

What changed

• recognize standard YouTube watch URLs across bare, www, and mobile hosts
• recognize YouTube Shorts as watch-history entries
• retain deleted/private watch records whose v parameter is present but blank
• require the Takeout activity title to explicitly identify a watch, so shares, reactions, advertiser visits, subscriptions, and account events are not emitted as Watched

Root cause

The Takeout normalizer treated nearly every activity attributed to YouTube as watch history, then validated it with a literal ://www.youtube.com/watch?v= substring. Current Takeout data includes Shorts and non-watch YouTube activities, so valid Shorts raised RuntimeError while unrelated activity was either raised as an error or silently counted as a watch.

Impact

Consumers get a cleaner watch-history stream, including Shorts, without losing deleted/private-video placeholders.

Checks

• uv run --group testing --with google-takeout-parser pytest -o addopts='' -q src/my/tests/youtube.py — 2 passed
• uv run --group testing ruff check src/my/youtube/takeout.py src/my/tests/youtube.py — passed
• uv run --group testing ruff format --check src/my/youtube/takeout.py src/my/tests/youtube.py — passed

Full private Takeout audit using the parser fix from purarue/google_takeout_parser#95:

• 350,849 parsed events; 0 parser errors
• before: 72,286 watches; 15 YouTube RuntimeErrors
• after: 72,278 watches; 0 YouTube errors
• one valid Short recovered
• 14 erroring non-watch activities skipped
• 9 additional non-watch interactions no longer silently counted as watches
• all 7,142 deleted/private-video placeholders retained

The aggregate proof report contains no personal event content. SHA-256: 74f560123787827dcb96e10bda2d1508d17656d85bbe47bbbf77c96583f47048.

[AndrewSB]

hey @karlicoss! this was done by AI, but I've validated the change after it's successfully parsed the youtube section of my takeout

happy to make any edits you'd like

[karlicoss]

Hey, thank you! Looks good :)