Study AL

A powered measurement of the client-prune fence across four models found it works perfectly on sonnet (prune cells fell from 16/40 to 0/40, p = .00002) but is unnecessary on fable-5, which spontaneous…

Study AL found that a prompt-level fence against self-eviction reduced client-side pruning from 5 to 1 pooled across models, but the pre-registered significance gate failed at p = .219 because the con…