Input: one slider — how much you've packed into a single AI session, from "just started" to "very long."
Outcome: a real blog paragraph reacts live — sentences that are specific and sit deep in the middle get color-flagged first, and once risk is high enough they're actually swapped for a vaguer, hedged rewrite. Nothing warns you first; you just watch it happen.
Useful for: seeing, concretely, which kind of content degrades first in a long session — specific mid-document facts — so you know what to double-check as sessions run long.
10% of a typical long session
fineat risk (flagged, not yet changed)degraded (actually reworded)