a test_throws should generally speaking not be this permissive -- you need to check it was actually the error you expected, not some completely unrelated error (thus hiding an actual bug, making the test actively harmful)

Thanks for the review @Krastanov! I've pushed an update addressing both points, and would like to flag one design question that came up while working through the dispatch verification.

1. Tests now explicitly verify dispatch routing

I've added a new testset "noisify dispatches NoiseModel fields correctly" that exercises each dispatch path with visually distinguishable noise per field: different probabilities (0.01–0.05) and different Pauli biases (X-only / Y-only / Z-only / depolarizing), and asserts that each NoiseOp points back to the correct NoiseModel field, on the correct qubits.

Here's the noisified output for a 3-qubit, 5-operation circuit (5 ops → 15 ops):

 [1]  NoiseOp(PauliNoise(0.05, 0.0, 0.0), (2, 3))   ← idle on q2, q3
 [2]  NoiseOp(PauliNoise(0.01, 0.0, 0.0), (1,))     ← single_qubit on q1
 [3]  sHadamard(1)
 [4]  NoiseOp(PauliNoise(0.05, 0.0, 0.0), (3,))     ← idle on q3
 [5]  NoiseOp(PauliNoise(0.0, 0.02, 0.0), (1, 2))   ← two_qubit on q1, q2
 [6]  sCNOT(1, 2)
 [7]  NoiseOp(PauliNoise(0.05, 0.0, 0.0), (1, 3))   ← idle on q1, q3
 [8]  NoiseOp(PauliNoise(0.01, 0.0, 0.0), (2,))     ← single_qubit on q2
 [9]  sX(2)
[10]  NoiseOp(PauliNoise(0.05, 0.0, 0.0), (2, 3))   ← idle on q2, q3
[11]  NoiseOp(PauliNoise(0.0, 0.0, 0.03), (1,))     ← measurement on q1
[12]  sMZ(1)
[13]  NoiseOp(PauliNoise(0.05, 0.0, 0.0), (1, 2))   ← idle on q1, q2
[14]  NoiseOp(PauliNoise(0.04, 0.04, 0.04), (3,))   ← reset on q3
[15]  sMRZ(3)

Every gate is preceded by (a) idle noise on the non-affected qubits and (b) the op-specific noise from the matching NoiseModel field. The testset asserts both noise and indices for each NoiseOp and that the original gate is preserved at the expected position. Because each field uses a distinct PauliNoise parameter, any misrouted dispatch would fail the assertion.

2. @test_throws now matches the specific error message

Changed @test_throws ErrorException ... to @test_throws "nqubits must be provided" ..., so the test only passes on the intended error path rather than on any unrelated ErrorException that might be thrown.

3. Design question: should noise_model.reset apply to plain Reset?

I noticed that Reset (without measurement) currently falls through to the no-op fallback, it picks up noise_model.idle on non-affected qubits but not noise_model.reset:

julia> noisify([Reset(S"Z", [1])], model; nqubits=3)
2-element Vector{Any}:
 NoiseOp(PauliNoise(0.05, 0.0, 0.0), (2, 3))   # idle on q2, q3
 Reset(...)                                     # no reset noise on q1

The reset dispatch is currently noisify(g::AbstractResetMeasurement, ...), which covers sMRZ and friends but not Reset itself (which is AbstractOperation but not a subtype of AbstractResetMeasurement).

Is the intended semantics that noise_model.reset applies only to measure-and-reset operations (current behavior), or should it also apply to plain Reset?