Bug
calculate_monthly_from_annual() raises KeyError: Period('1997-12', 'M') — reproduced on current master (2026-06-06) and confirmed pre-existing (the same crash reproduces on the code before the ordering normalization; the function has never been covered by tests).
Root cause
The backfill loop does a hard label lookup on the annual series:
s_monthly[date] = (s_annual.shift(11)[date] + 1.0) / prod_monthly - 1.0
fetch_series() drops observations whose value is empty, so gaps in the NBS annual CPI series (leaf A01010201, 1987–2015) become missing labels — and s_annual.shift(11)[date] raises KeyError instead of yielding NaN. 1997-12 is such a gap in the data NBS currently serves.
Suggested fix
Tolerate missing labels: s_annual.shift(11).get(date) → None/NaN → the affected month gets NaN instead of crashing the whole backfill (Series.prod() skips NaN by default, so neighbouring months still compute). Optionally warn listing the months that could not be derived.
Test
tests/test_annual_and_monthly_inflation.py::test_calculate_monthly_from_annual_is_ascending is marked xfail(raises=KeyError, strict=False) referencing this issue — fixing the crash flips it to pass (it then asserts the result is ascending and starts at 1987-01).
🤖 Generated with Claude Code
Bug
calculate_monthly_from_annual()raisesKeyError: Period('1997-12', 'M')— reproduced on current master (2026-06-06) and confirmed pre-existing (the same crash reproduces on the code before the ordering normalization; the function has never been covered by tests).Root cause
The backfill loop does a hard label lookup on the annual series:
fetch_series()drops observations whosevalueis empty, so gaps in the NBS annual CPI series (leafA01010201, 1987–2015) become missing labels — ands_annual.shift(11)[date]raisesKeyErrorinstead of yielding NaN. 1997-12 is such a gap in the data NBS currently serves.Suggested fix
Tolerate missing labels:
s_annual.shift(11).get(date)→None/NaN → the affected month gets NaN instead of crashing the whole backfill (Series.prod()skips NaN by default, so neighbouring months still compute). Optionally warn listing the months that could not be derived.Test
tests/test_annual_and_monthly_inflation.py::test_calculate_monthly_from_annual_is_ascendingis markedxfail(raises=KeyError, strict=False)referencing this issue — fixing the crash flips it to pass (it then asserts the result is ascending and starts at 1987-01).🤖 Generated with Claude Code