Stricter `DataFrame.at` typing introduced in #1313 caused unexpected CI failures

[yechielvi]

Hi,

After PR #1313, our CI started failing due to stricter typing of
DataFrame.at. This wasn’t mentioned in the PR title/description.

What changed

• Previously, .at was loosely typed so mypy didn’t flag invalid usage.
• After Add check assert_type to tests for frame.py and add __new__ method for DatetimeIndex #1313, .at enforces stricter signatures (correctly).
• This surfaced real issues on our side where we assumed int indexing without explicit narrowing.

Impact

• CI failed immediately because our unpinned install pulled the latest version.
• The failures were surprising since the PR only mentioned adding assert_type tests and a __new__ for DatetimeIndex.
• Users may perceive this as a “silent breaking change,” even if it’s a stubs bugfix.

Suggestion

• Please call out such changes explicitly in PR descriptions and release notes, as they can break downstream CI.
• Consider labeling them as breaking typing changes.

Refs: #1313

[Dr-Irv]

This is a real burden for us to be able to document "breaking typing changes". We have very few people working on this project. We are continuously improving the stubs to provide better support for type checking. In some cases, an "old" version may have had weak checking for a particular method, so we tightened it up. In other cases, an "old" version may have had too strict type checking, so we loosened it up.

You can assume that any new release of pandas-stubs can have this kind of effect.