Problem
The embedded CPI-U data in recipe.yaml jumps in large multi-year gaps (e.g. 1938, 1940, 1945, 1950, 1955...). Questions asking for real-dollar values in unlisted years (e.g. FY1943, FY1963) have no reference point. The model will silently interpolate or skip the conversion entirely — both produce wrong answers with no error signal.
Fix
Either fill in every year from 1938–2020 in the CPI table, or add an explicit instruction to the prompt:
If the exact CPI year is missing from the table, use linear interpolation between the two nearest values and note the interpolation in your working.
Impact
High — any real-dollar question touching an unlisted year scores incorrectly with no visible failure.
Problem
The embedded CPI-U data in
recipe.yamljumps in large multi-year gaps (e.g. 1938, 1940, 1945, 1950, 1955...). Questions asking for real-dollar values in unlisted years (e.g. FY1943, FY1963) have no reference point. The model will silently interpolate or skip the conversion entirely — both produce wrong answers with no error signal.Fix
Either fill in every year from 1938–2020 in the CPI table, or add an explicit instruction to the prompt:
Impact
High — any real-dollar question touching an unlisted year scores incorrectly with no visible failure.