Fundamental databases are usually stored in their corrected form. If a company reported EPS of 1.12 on 3 February and restated to 0.94 in August, a non-PIT database shows 0.94 for the whole period, and your backtest trades February on information that did not exist until August.
The same applies to macro data. US non-farm payrolls are revised for months; GDP for years. A strategy trading the surprise against consensus must use the first print and the consensus as it stood, not today's revised series.
The practical test: ask your vendor whether the database records the date each value became available, not just the date it refers to. If it does not, treat any fundamentals-driven backtest result as an upper bound and expect live results well below it. See also restatement-bias.
Related: restatement-bias, look-ahead-bias, data-vendor, survivorship-bias