What fixes the zero of network energy?
What a shift leaves invariant
Let a network assign a real value to each configuration. Consider the records that depend on differences of those values, their extrema or the normalized probabilities in a fixed finite Gibbs ensemble. Add the same constant to every configuration. Every difference, minimizer and normalized probability is unchanged, while the value at a minimum changes by exactly the constant. In particular two models disagree on the alleged vacuum value while agreeing on every record in this menu. The finite ensemble statement follows by cancelling the same positive exponential in numerator and denominator.
What would remove the ambiguity?
Let histories have a measured additive volume. An action term proportional to that volume need not cancel between histories of different volumes. If the observable is sensitive to an action contrast between two histories and their volumes differ by a known nonzero amount, that contrast can identify the coefficient provided the other contributions and units have been fixed independently. If all accessible histories have the same volume, no contrast in that menu separates any two coefficients. These are conditions on what the network can measure; neither a spectrum nor a dimensionless cancellation supplies the volume response.
What a numerical prediction requires
The geometric cosmological constant has units of inverse length squared. A vacuum energy density has different units; converting it to curvature requires a dynamical coupling and calibration. A cancellation of a signed spectral sum, equal sector counts, or multiplication of chosen attenuation factors supplies none of these. A predictive model must specify an absolute volume-coupled observable, its calibration, competitors and an independently testable value or interval. No numerical value of the physical cosmological constant follows from the present network observations.
See cosmology and physical correspondence.