For deterministic computation, we have wide range of approaches, to name few:

• unit-testing
• property based testing
• fuzzing
• symbolic execution
• formal verification

(just to name few...and each of those has many flavors and approaches)

Those are traditionally focused/applied on ensuring that deterministic computation "behaves" :).

Here, I wonder, in terms of tooling, frameworks, and how they would look like in code for actual programmer,
if there is some universal way, insights, how to map those approaches into LLM programs (LMPs) / programs using LMPs /LLMs ?

Or if alternative approaches required , how to make them ergonomic so actual software engineers feel like implementing them? (so implementing testing does not become either 100x harder then actual code, not flaky or unreliable to a point of not providing actual value...)