A strict < between a rounded result and an unrounded one passes or fails by luck
A wiring test asserted plan.baseline_after.electricity_kwh.expected < run_baseline(...).electricity_kwh.expected. The plan's figures are rounded to 2 decimals for output; the direct baseline call is not. When the selected measures do not change the building load at all (layout plus rooftop PV, where PV is netted elsewhere), after equals before, and the test outcome depends only on the rounding direction: 294391.31 < 294391.313261 passed, but after a small fixture change 294568.76 < 294568.757764 failed. Lessons: (1) when comparing an output-rounded value with a raw one, round both or use a tolerance; (2) a strict inequality encodes a claim (the plan reduces load) that only holds if the fixture selects a load-reducing measure, so either assert <= or build the fixture to make the claim true; (3) when a regenerated fixture flips a test like this, check whether the old pass was real before blaming the new data.