The spec says that the last successful position option is cleared in the following cases:

• when the element is not absolutely positioned anymore
• when the computed value of position-try-fallbacks for the element changes
• when any @position-try rules referenced by the element's position-try-fallbacks changes

But from testing WebKit's implementation against WPT, it seems that some other cases could also trigger invalidation:

1. when the computed value of position-try-order changes (WPT position-try-order tests seem to expect this)
2. when the "base style" of the element changes (e.g by adding/removing a class from an element, see https://github.com/web-platform-tests/wpt/blob/719745a09c32182d0b0785a7ac8c393cc44d8c51/css/css-anchor-position/base-style-invalidation.html)
3. when a position-try fallback rule doesn't refer to an anchor, and the base style changes (???) (see https://github.com/web-platform-tests/wpt/blob/719745a09c32182d0b0785a7ac8c393cc44d8c51/css/css-anchor-position/anchor-position-non-anchored-fallback.html)

(1) makes sense to invalidate, but (2) and (3) seems like a bug (or at least, (un)intentional side effects in Chrome?) So I'm looking for clarification on if the tests are correct or not, and if they are, we should include those cases in the spec.