Evaluates the normalized cubic Bernstein basis at the supplied local parameter.
Checks finite Cartesian coordinates without repairing or clamping values.
Lower componentwise physical control bound in registered millimetres.
Admitted ceiling on sparse stencil entries after affine substitution.
Fixed affine contributions after subtracting the admitted origin and dividing by the common scale.
Physical millimetre origin subtracted from coefficient positions before normalization.
Admitted acyclic affine substitution, preserving fixed values and independent control identities.
One sparse millimetre expression per original control, in input ID order.
Unowned original IDs, in ascending order; these are the free variables.
Largest declared relation-edge depth, including zero-weight dependencies.
Largest final independent term count, excluding folded fixed sources.
Largest absolute deviation of pre-folding source weight sum from one.
Least retained flattened source weight; zero for an empty expression.
Positive common millimetre scale used to normalize the physical coefficient box.
Original control index to independent solver slot; fixed or dependent entries retain the owner's sentinel.
Sum of positive observation weights used to normalize the assembled objective.
Number of independent vector controls after fixed and affine dependent controls are eliminated.
Upper componentwise physical control bound in registered millimetres.
Smallest accepted over largest data-QR diagonal, or null when unavailable; not a singular-value certificate.
Rank of column-scaled twice-reorthogonalized data rows, or null when its work budget is exceeded.
Observed admission failures and qualification limits in production order; an empty array adds no extra certificate.
Hessian diagonal in independent-control slot order, used for congruence scaling.
Smallest accepted over largest Cholesky pivot, or null when unavailable; not a condition number.
Sparse symmetric Hessian rows; each entry stores a column index and its numeric coefficient.
Constant term of the assembled normalized least-squares objective.
Largest assembled sparse observation stencil, including exact-zero retained terms.
Numerically observed Gram rank, or null when the explicit dense diagnostic budget is exceeded.
Actual number of signed physical-derivative rows included in the objective.
Relative binary64 rank-admission threshold; this is not an exact rank predicate.
Assembled normalized right-hand-side values in the same unknown-slot order as the Hessian.
Number of unknown controls with no positive observation-basis support.
Assemble the same sparse weighted system and its bounded rank diagnostics.