Compressed sparse column offsets for A, length variable-count plus one; final offset equals the coefficient count.
Diagonal P coefficients of the convex objective 1/2xᵀPx + qᵀx; the public adapter admits finite nonnegative coefficients.
OptionalinitialOptional compatibility vector; the kernel deliberately ignores it and always starts a cold homogeneous solve.
Finite linear objective vector q, with the same variable count as diagonal.
Lower bound for each original interval row; negative infinity represents an absent lower side.
Zero-based constraint row ordinal for each CSC coefficient, aligned with values.
Upper bound for each original interval row; positive infinity represents an absent upper side, and equal finite bounds form an equality.
Finite CSC coefficients of A in the caller's numerical scaling, aligned with rowIndices.
Solve one admitted interval CSC problem entirely in engine-owned TypeScript. The public adapter owns semantic admission and original-row residual checks. This boundary owns equivalent conic expansion, reversible equilibration and the cold homogeneous interior-point solve. The compatibility initial vector remains ignored, with no retained factors or preceding-call warm state.
Status 1 is Solved, 2/3 are primal/dual infeasibility, 4 is AlmostSolved; 5/6 are reduced-accuracy infeasibility, 7 is the iteration limit, 9 is numerical refusal and 10 is insufficient progress. Only status 1 means full convergence. Zero variables return negative API status with unavailable diagnostics as NaN. Unsuccessful vectors must not be accepted as geometry. Historical native transport and bytes remain in repository history, outside shipped product paths.