In noisy physical reservoirs, the classical information-processing capacity $C_{ip}$ quantifies how well a linear readout can realize tasks measurable from the input history, yet $C_{ip}$ can be far smaller than the observed rank of the readout covariance. We explain this "missing capacity" by introducing the innovation capacity $C_i$, the total capacity allocated to readout components orthogonal to the input filtration (Doob innovations, including input-noise mixing). Using a basis-free Hilbert-space formulation of the predictable/innovation decomposition, we prove the conservation law $C_{ip}+C_i=rank({\Sigma }_{XX})\le d$, so predictable and innovation capacities exactly partition the rank of the observable readout covariance ${\Sigma }_{XX}\in {ℝ}^{d\times d}$. In linear-Gaussian Johnson-Nyquist regimes, ${\Sigma }_{XX}(T)=S+T{N}_0$, the split becomes a generalized-eigenvalue shrinkage rule and gives an explicit monotone tradeoff between temperature and predictable capacity. Geometrically, in whitened coordinates the predictable and innovation components correspond to complementary covariance ellipsoids, making $C_i$ a trace-controlled innovation budget. A large $C_i$ forces a high-dimensional innovation subspace with a variance floor, and under mild mixing and anti-concentration assumptions this yields extensive innovation-block differential entropy and exponentially many distinguishable histories. Finally, we give an information-theoretic lower bound showing that learning the induced innovation-block law in total variation requires a number of samples that scales with the effective innovation dimension, supporting the generative utility of noisy physical reservoirs.