Covariance of the unit-intensity noise increment across the axis named by correlated_over — the second-order structure of the driving process, stated as mathematics rather than as a factorisation. Must be square in that axis, symmetric and positive semi-definite; its diagonal carries per-element variances and need NOT be unity, so unequal variances and cross-element correlation are expressed together. The realised increment is the declared sigma times a draw from this covariance, so parameters: {sigma: ...} continues to mean what it means for uncorrelated noise. Absent (the default) means the increment is independent across that axis. Being a Parameter, it carries the full provenance triple: a literal value:, existing bytes via source:+measure:, or — the usual case for a derived operator — producer:, whose FunctionCall records how the structure was obtained (projecting a spatially white field into a modal basis, spreading independent per-region drives through a connectome). The generative account therefore lives in the provenance, and the matrix itself stays a statement about the process. How a backend samples it (Cholesky, eigendecomposition, whitening) is a backend concern and is never stated here.
name: covariancedescription:'Covariance of the unit-intensity noise increment across the axis named by `correlated_over` — the second-order structure of the driving process, stated as mathematics rather than as a factorisation. Must be square in that axis, symmetric and positive semi-definite; its diagonal carries per-element variances and need NOT be unity, so unequal variances and cross-element correlation are expressed together. The realised increment is the declared sigma times a draw from this covariance, so `parameters: {sigma: ...}` continues to mean what it means for uncorrelated noise. Absent (the default) means the increment is independent across that axis. Being a `Parameter`, it carries the full provenance triple: a literal `value:`, existing bytes via `source:`+`measure:`, or — the usual case for a derived operator — `producer:`, whose FunctionCall records how the structure was obtained (projecting a spatially white field into a modal basis, spreading independent per-region drives through a connectome). The generative account therefore lives in the provenance, and the matrix itself stays a statement about the process. How a backend samples it (Cholesky, eigendecomposition, whitening) is a backend concern and is never stated here.'from_schema: https://w3id.org/tvborank:1000owner: Noisedomain_of:- Noiserange: Parameterrequired:falseinlined:true