[Rafael-SOWNet]

EquationSystem.Solve threw WrongNumberOfArgumentsException for a system with more or fewer equations than unknowns, unless the triangularisation took it or it was short and linear with rational coefficients. The exception says the call was wrong, and it was not. On #212, @Happypig375 said accepting new forms and not throwing is fine, which answers the question my earlier comment there left open.

Such a system is now eliminated one unknown at a time:

• An unknown that nothing is left to constrain becomes a parameter, t_1, as in the linear case.
• With no unknown left, an equation that is a number away from zero drops its branch. One that cannot be decided, such as (a + b)/2 - c, means that equation gives no answer, and the next equation is tried.
• Every family is substituted back into every equation and kept only where each one vanishes. Otherwise the system is declined as null, which Solve documents as no solution found.

A square system does not reach this path, so no square answer changes.

	2.6.0	this branch
x^2 + y^2 - 1 in x, y	WrongNumberOfArgumentsException	[[t_1, sqrt(-(t_1 ^ 2 + -1))], [t_1, -sqrt(-(t_1 ^ 2 + -1))]]
a*x + y - 1 in x, y	WrongNumberOfArgumentsException	[[t_1, -(a * t_1 + -1)]]
sin(x), x in x	WrongNumberOfArgumentsException	[0]
sin(x) - y, cos(x) - y, x - pi/4 in x, y	WrongNumberOfArgumentsException	[[pi / 4, 1/2 * sqrt(2)]]
sqrt(y) - x in x, y	WrongNumberOfArgumentsException	null
x + y - a, x - y - b, x - c in x, y	WrongNumberOfArgumentsException	null

Why the substitution back is needed. Two of the declines would otherwise be wrong answers:

• sqrt(y) = x is solved for y as x^2, which is no solution for a negative x. sqrt(t^2) - t does not simplify to 0, so the family (t, t^2) is not returned.
• A coefficient that is zero without being written so. For x^2 - c y with c = (a + 1)^2 - a^2 - 2a - 1, the elimination divides by c, and Simplify calls the residual 0 provided not c = 0. A provided condition counts only where it holds with the symbols pinned, so that family is declined too.

NonSquareSystemTest checks the families numerically at negative and positive values of their parameters, instead of the way the solver checks them. It also checks the overdetermined answers exactly, and the three declines. ParametricSystemTest.WhatIsNotLinearInTheUnknownsIsRefused pinned the exception for four of these shapes and is removed. Entry in BREAKING-CHANGES.md; Exceptions.md and the Solve doc comment no longer say the counts must match.

Not in this PR:

• The square path has the same missing check, and it answers wrongly. sqrt(y) - x, x + 1 in x, y gives [[-1, 1]], while sqrt(1) - (-1) is 2. That is a changed answer for square systems, so it goes in its own PR, for A system of equations is answered with rows that do not satisfy it #2123.
• A family with an integer parameter is declined. sin(x) - 1/2 in x, y gives null, because sin(pi/6 + 2 pi n_1) - 1/2 does not simplify to 0 without knowing that n_1 is an integer.

Part of #212

🤖 Generated with Claude Code

https://claude.ai/code/session_012sonx8iAspMiwRwokT1Ura