Describe the issue:

When updating from numpy 2.2.6 to 2.3.0, I encountered a new type check error that I believe is incorrect.

When you iterate over a matrix, the elements should be the same shape as the original matrix, but with one dimension dropped. Because the NDArray type has ambiguous shape, elements could either be an array OR a scalar.

However, in numpy 2.3.0, when I iterate over a matrix of type NDArray[np.float64], numpy typing says each element has type np.float64. It is assuming that the matrix is 1-dimensional and each element is a scalar, which is incorrect.

arr: NDArray[np.float64] = np.zeros((5,3))
for element in arr:
  # Each `element` is an `NDArray[np.float64]` of shape (3)
  # But instead, numpy typing says it is a `float64`

In numpy 2.2.6, each element had type Any - vague, but at least not incorrect.

From the release notes, I see that improvements were made to numpy's static typing in 2.3.0. I also see mypy_plugin was deprecated, but I don't think that's the issue because I didn't have the plugin enabled in the first place.

So is this an actual bug, or am I misunderstanding how to use static typing with numpy? I feel a little shaky on the subject. If I am doing this wrong, please show me the correct way 🙏

Reproduce the code example:

"""A simple example that should throw a type check error."""

import numpy as np
from numpy.typing import NDArray


def example(pixels: NDArray[np.float64]) -> None:
	"""`pixels` is an NDArray of shape (length, 3)."""
	# Convert NDArray to a list of tuples
	tuples: list[tuple] = [tuple(pixel) for pixel in pixels]

Error message:

error: Argument 1 to "tuple" has incompatible type "float64"; expected "Iterable[Any]"  [arg-type]

Python and NumPy Versions:

2.3.0
3.13.5 (main, Jun 12 2025, 22:47:08) [GCC 12.2.0]

Type-checker version and settings:

mypy version: 1.16.1
mypy command: mypy .

Additional typing packages.

No response