The case
U20: “that’s just a similarity constraint, not an equivalence. This is why the game ‘telephone’ works.” A50: “translation can preserve recognizable structure without preserving identity, and … small local deviations can accumulate into large global changes.” A52 made it running case B, beside Henley and the language model.
What it can show
The game is a familiar practice, so the case is documented in that sense. In the exchange it is used as an idealized illustration, not as data. It shows that each step can be close to its predecessor while the end is far from the start (CLAIM-056), and that the working relation is similarity within ε rather than equivalence (CLAIM-106). Bartlett’s serial reproduction and Kirby, Cornish and Smith’s iterated learning, brought in at A50, are its experimental versions (manuscript §8).