Lambda that uses INDEX with array arguments behaves inconsistently when saved in the Name Manager

Hello,

 I have encountered what appears to be inconsistent behavior when a particular type of lambda is saved in the Name Manager. The issue seems to occurs when the LAMBDA uses INDEX with either the row_num or column_num argument supplied as an array. The following is a minimal reproducible example: The formula "=LAMBDA(arr,INDEX(arr,1,{1,2,4}))({1,2,3,4})" correctly evaluates to {1,2,4}. Now, save the lambda as TEST (or any other name) in the name manager. The formula "=TEST({1,2,3,4})" also spills the expected array {1,2,4}. However, when the result is passed to another function, the behavior changes. For example, "=SUM(TEST({1,2,3,4}))" and "=COLUMNS(TEST({1,2,3,4}))" both evaluate to 1. In contrast, using the lambda inline instead of the defined name produces the expected results: "=SUM(LAMBDA(arr,INDEX(arr,1,{1,2,4}))({1,2,3,4}))" returns 7, and "=COLUMNS(LAMBDA(arr,INDEX(arr,1,{1,2,4}))({1,2,3,4}))" returns 3.

Unfortunately, I am unable to attach a workbook to this post. Additionally, the issue is not reproducible on every machine I have tested, although it is consistently reproducible in Excel for the Web. I am currently using Excel version 2607 on the Current Channel. Does this appear to be expected behavior, or is it a bug?

In the meantime, what would be the best way to mitigate this issue? I have found two potential workarounds. The first is to prepend the result of the named lambda with a unary + (e.g "=SUM(+TEST({1,2,3,4}))"), which appears to force excel to treat the result as an array. However, when using a shared lambda library (as is the case for most of my team), users generally do not know the implementation details of each lambda, so it is difficult to determine when this workaround is necessary. The second approach is to avoid passing an array to the row_num or column_num argument of Index by using MAP.  For example, the TEST lambda defined above can be rewritten as =LAMBDA(arr,MAP({1,2,4},LAMBDA(idx,INDEX(arr,1,idx)))), which causes it to behave as expected. However, I am concerned about potential performance implications. Intuitively I would expect that a single INDEX call with an array argument would be more efficient than multiple INDEX calls wrapped in MAP, but I do not know whether this is the case.

Has anyone else encountered this behavior before?