Quick note re timing: It would be good to have this instrumentation deployed by the time the new types are rolled out if we want to collect data on initial usage.

Per our measurement plan and the PRD for the types work, this instrumentation will be used to monitor progress towards the following success metrics

• "New functions created with these types are used at least 1000 times "
• The community will have created 100 new functions using the new types"

May I chip in with a question?

How will we cope with custom-typed objects in an untyped context? I’m thinking mainly of Z1-lists functioning as informal objects. One might have subject, verb, object in a list representing a “sentence” (supposing these to be custom types), but the function’s argument might be Z1-list and its return type might just be monolingual string, or just string, and inflecting the verb could be a simple string function but one that depends on an interpretation of some object as a particular kind of verb, based on its type or its keys (but that might be just a call to Z803/4 with some indeterminate (Z1) object, so we’d need to spot the returned value(s) to infer that some subsequent string argument might be functioning (implicitly) as an instance of a custom type (when I might have been checking that it was *not* a string of that type, or some other type…)

I’m assuming that “explicit user request using the function evaluation widget” will include function calls while editing a test case or implementation. That is the context in which the majority of my function calls have been made (mostly Try this implementation, I would guess, but then it’s hard to distinguish between de-bugging, testing and actual “use”).