Nishi FamilyCompareComputational › Run-and-Compare Bench

Run-and-Compare Bench

Exact rationals + polynomials + matrices + bignums — Nishi is RUN on each input; its output is compared to the mathematically-correct value.

This is the deepest form of “mechanically test and bench”: not symbol-presence, but actual EXECUTION. The real nx_rational, nx_poly, nx_matrix and nx_bigint_lib organs are run on each input and the computed value is compared exactly to the correct answer (identical to what Wolfram or any exact CAS returns). The NEG-CONTROL rows are checked against deliberately-wrong answers — they MUST fail, proving the comparison discriminates.
CapabilityInputNishi computedCorrect (Wolfram-class)Verdict
Exact rational addition1/6 + 1/31/21/2PASS
Fraction normalization (gcd)2/41/21/2PASS
Exact rational multiply(1/3) * (3/1)1/11/1PASS
Signed normalization-2/6-1/3-1/3PASS
Common-denominator addition1/2 + 1/35/65/6PASS
Euclid gcdgcd(48, 36)12/112/1PASS
Polynomial multiply(x+1)(x-1) -> constant coeff-1/1-1/1PASS
Symbolic derivatived/dx(3x^2+2x+1) -> x coeff6/16/1PASS
Polynomial evaluation (Horner)3x^2+2x+1 at x=217/117/1PASS
Polynomial addition(x^2+2x+3)+(4x+5) -> x coeff6/16/1PASS
Power sum (Gauss)1+2+...+1005050/15050/1PASS
Matrix multiply 2x2[[1,2],[3,4]] x [[5,6],[7,8]] -> (0,0)19/119/1PASS
Determinant 2x2det [[1,2],[3,4]]-2/1-2/1PASS
Determinant 3x3det [[1,2,3],[4,5,6],[7,8,10]]-3/1-3/1PASS
Matrix transposetranspose [[1,2],[3,4]] -> (0,1)3/13/1PASS
Bignum carry across limb4294967295 + 14294967296/14294967296/1PASS
Bignum multiply65536 x 655364294967296/14294967296/1PASS
Bignum byte decodebytes DE AD BE EF3735928559/13735928559/1PASS
NEG-CONTROL: 1/6+1/3 vs wrong 1/31/6 + 1/3 =? 1/31/21/3correctly rejected
NEG-CONTROL: poly const vs wrong +1(x+1)(x-1) const =? 1-1/11/1correctly rejected
NEG-CONTROL: det vs wrong +2det [[1,2],[3,4]] =? 2-2/12/1correctly rejected
NEG-CONTROL: carry vs wrong4294967295+1 =? 42949672954294967296/14294967295/1correctly rejected

Real KATs passed 18/18 · neg-control correctly rejected 4/4

Generated by nx_swcompare_bench — imports the real nx_rational organ, executes it, compares to the exact correct value. This upgrades the computational comparison from capability-presence to executed-and-correct. Zero JS, zero trackers.