# Argument selection for Báez-Duarte B_N squared errors

The only free parameter is the positive integer N, ordered increasingly.
Category: discrete_index. Báez-Duarte, math/0011254v1 §1.2 equation (1.14),
defines B_N by truncating the Möbius sum at N and correcting its final dilation
by -N g_N rho_N, where g_N=sum_{n<=N} mu(n)/n. There is no coupling,
probability, shape, normalization, or continuous argument to sample.
The named quantity is ||chi+B_N||^2 in L^2((0,infinity),dx), for every N>=1.

Compute every 1<=N<=30, including nonsquarefree endpoints. This is a
noncanonical initial prefix: its purpose is comparison with all currently
stored optimal errors T641 (N=1,...,30), the Selberg errors T642 (N=2,...,30),
and the untapered baseline T706. Those definitions and live metadata were
refreshed before claiming. All pairwise ratios fall within the documented
T640 kernel range; fresh enclosed calculations, not its decimal entries,
provide the kernel. The whole prefix exposes moving-endpoint corrections,
including changes where mu(N)=0. Pruning to prime/squarefree/famous indices
would conceal this behavior. There are no additional argument families for
this specified one-index construction; the other named constructions are
separate tables outside this assignment. A longer prefix is a future extension,
not an intrinsic mathematical boundary. Thirty 100-digit scalars are readable;
computed block sizes and worst radii are retained in the numerical evidence.

Default enumeration, Programs and completeness must all say N=1,...,30.
The parameter domain remains N>=1 and complete is no. N=1 is exactly 1
because B_1=0. No quantiles, percentiles or statistical critical values occur.

Instructions: skill-manifest.json copies the run's archived matching bundle,
SHA-256 d6661c2f21613524889fa3a57d3c95f8330344ca62e5a856211c08a282b7dbab.
This is the builder's justification, not the independent argument review.