Exercise 3: An Off-by-One Bug the Original Suite Misses — Possible Solution
====================================================================

THE BUG
------------------------------
    def calculate_shipping_offbyone_bug(weight, zone, express):
        ...
        if weight >= 50: base *= 0.9  # BUG: >= instead of >

RESULTS AGAINST THE ORIGINAL 6 CHARACTERIZATION TESTS
------------------------------
    (10, 'A', False):  expected 5.0,  got 5.0:  PASS
    (10, 'B', False):  expected 6.0,  got 6.0:  PASS
    (10, 'C', True):   expected 17.5, got 17.5: PASS
    (-5, 'A', False):  expected 0.0,  got 0.0:  PASS
    (10, 'Z', False):  expected 10.0, got 10.0: PASS
    (60, 'A', False):  expected 27.0, got 27.0: PASS

    Any of the 6 original tests caught the bug: False

RESULT AGAINST EXERCISE 2'S OWN BOUNDARY TEST
------------------------------
    weight=50: buggy version gives 22.5, characterization expected 25.0
    Exercise 2's own test result against this bug: CAUGHT THE BUG

WHY NONE OF THE ORIGINAL 6 CATCH IT
------------------------------
None of the original 6 inputs uses weight=50 exactly - the only test
anywhere near the threshold is weight=60, which satisfies BOTH
"weight > 50" and "weight >= 50" identically, so changing the operator
produces no observable difference for that specific input. The bug is
real and would affect any weight=50 order in production, but it's
invisible to a test suite that never happens to probe that exact
value.

WHY THIS WORKS AS AN ANSWER
------------------------------
This is the chapter's own central caution made concrete: a
characterization test suite is only as good as the inputs it was built
from. A suite that passes completely can still miss a real regression,
if the regression happens to only manifest on an input nobody thought
to characterize. This is exactly why Exercise 2's own boundary-value
test - discovered by deliberately probing the edge of a conditional,
not by chance - matters: boundary values are disproportionately likely
to expose exactly this class of bug, which is why deliberately testing
them (rather than only "typical" values) is a core characterization
testing skill, not an optional extra.