Where it works
YouTube, Netflix, Twitch and Crunchyroll, on both live streams and VODs where available. Playback starts the timer; muting or pausing stops it. Twitch is here because it's where I actually watch: Japanese streamers playing games I already play is the easiest immersion I've found. Anything else stays labelled Beta or Experimental in the extension until I've checked it against that site's current player, because there's no point counting time I can't count correctly.
Every language, no switching
It reads the language off each session and files the time under it. Japanese, then something in English, then Swedish, in one evening. All three are recorded and kept apart, and you're never asked to set a language first. The dashboard shows one language at a time today; a side-by-side view of how your time splits between them is the next thing I'm building.
Active and passive
Focused foreground viewing counts as active. Audible playback that keeps running while you cook or walk counts as passive. Both are kept, both are shown, and they're never added together into one flattering number. Active is always green, passive is always amber, everywhere in the product.
When, not just how much my favourite bit
Time of day, day of week, session length, streaks, a year heatmap and a weekly recap. This is the part I actually use: whether my evenings are focused or just background noise, and which day of the week quietly does the work.
YouTube Music, honestly undecided
Music plays through a video element, so the extension can see it and does count it: active if the tab is in front, passive if it's behind. I've deliberately not folded it in with everything else. Eight hours of background music is not eight hours of immersion, and a number that says otherwise is worse than no number. So it sits in its own source category where you can see it, judge it, and turn it off if you'd rather it didn't count. If you have a strong view on how this should work, tell me. I'm genuinely undecided.
Everything that's not video
A persistent timer and backdated entries cover reading, tutoring and speaking practice. Manual time counts towards your own totals, but it's excluded from any shared leaderboard, so nobody can pad a number.