Data revisions & vintages
Economic statistics are not fixed at publication: payrolls are revised for months, GDP gets three official estimates plus annual rewrites, even unemployment series are restated. The number in the database today is often materially different from the number that was published on that date.
The look-ahead trap
A backtest that asks 'would this signal have fired?' using today's revised data is subtly cheating — it answers the question with information nobody had at the time. Studies have shown indicators can look dramatically more accurate on revised history than they were in real time.
Vintages as the fix
ALFRED (FRED's archival sibling) stores every published version of a series with its release date — the 'vintage'. This monitor keeps vintages for its series, letting episode analysis and backtests replay exactly what was knowable on a given date. It is the difference between 'what we knew then' and 'what we know now'.