The claim
A4: “He begins with something close to: quality ≈ number of preserved constraints, and discovers that this doesn’t work either.” Its weighted form, also A4: translating “is selecting a point on a Pareto frontier”, with T* = argmax [w_s S + w_f F + w_r R + w_t Tn + w_p P].
How it was answered
By CASE-011: a rendering that broke most of the constraints and was judged perhaps the best. A4: “So even Hofstadter’s own carefully articulated notion of fidelity gets revised through the experiment.”
What survives
A distant descendant: the manuscript’s weighted functional L = L_obs + λL_str + γL_dec, with “no universal choice of weights or observational tasks” (§6). That keeps the weighting and drops the count. The record holds Hofstadter’s book (LIT-766) unread, so this stands on A4’s report.