Verifying callback-to-async conversion
An AI agent migrated a 2,000-line legacy module from callbacks to async/await across a Node service, asserting 'pure mechanical conversion, no behavior change.' It's too large to read carefully and the test suite is thin. You can't ship it on faith but you also can't rewrite the tests for 2,000 lines first. How do you get enough confidence to verify behavior preservation, and what's the async-specific regression you're hunting for?
compare_effect_traces(old_trace: list[str], new_trace: list[str]) → str[["a","b","c"],["a","b","c"]]out"identical"[["db_write","email_send"],["email_send","db_write"]]out"reordered at 0: expected db_write got email_send"[["load","save","notify"],["load","save"]]out"missing at 2: notify"Treat the AI’s output as a draft to verify, not an answer to trust. Name the specific flaw and the input that triggers it, say how you’d catch it (tests, edge cases, reading critically), and how you’d re-prompt or decompose to get it right.
Vibe & agentic: describe the solution in plain language (or narrate it) and the coach grades your approach.