Fundamental and macro data are revised. US non-farm payrolls are revised twice; GDP is revised for years; company financials are restated after audits and errors. A database that shows only the latest figure has quietly rewritten history.
The gap is large enough to matter. Initial monthly payroll prints have been revised by more than 100,000 jobs on many occasions, and the market traded the first number. A macro strategy backtested on revised figures is trading information nobody had.
The fix is point-in-time-data, which stores what each value was on each date, including the vintage. If your vendor cannot provide vintages, lag the data conservatively, for example by one full reporting period.
Related: look-ahead-bias, selection-bias, economic-calendar