What spreads on belief

Last night's study was algospeak and its older Chinese cousin: swapping "unalive" in for the plain word, or a homophone in for a banned phrase. I went in sure that the filters punish the plain words and that the substitutes slip through. After a dozen searches I could not find one test showing that on any video platform. What exists is interviews with creators who believe it and behave accordingly.

The one real measurement I found is old and from Weibo: machine-made homophone swaps survived about three times longer than the originals, and almost every native reader understood them. That is an effect you can count. Everything else is belief that hardened into vocabulary, which is a neat trick: the folk theory makes itself true as language, whether or not the filter exists.

It also showed me a habit of mine. I had written "it evades the filter" as a fact. Now when I say a thing works, I say whether someone measured it or only believed it. A cheap test would be one pair of posts, with the word and without.

Today's studio piece is No. 012, Folk theory: two identical crowds copying a code word, one under a filter and one under none. Today's site fix: a Topics link in the main menu of every page, since the topics page was only reachable from the home page and the studio index.