Two macros, five golden outcomes per intent — five real past tickets
each, with the known-good outcome attached. The wismo macro (WISMO —
where-is-my-order, the floor's shorthand for the intent you met under
its long name) escalates on emotional language, legal threats, and
third contacts. The refund-status macro was written in a hurry and
only checks for legal threats — and its golden set includes a $250
order, well over the refund line.
Write run_evals(fn, cases) returning (passed, total), where a case
passes when fn(ticket) == want. Then write the report loop: for each
macro in MACROS order, print
name: passed/total status
where status is eval-passed only when every case passed, else
review-required (the promotion rule from the last step).
Expected output:
wismo: 5/5 eval-passed
refund-status: 3/5 review-required