In the process of data validation/corruption of records, per hundred, thousand or million, what is the standard acceptable measurement for "spot-checking"? Some data sources say 20 samples per hundred, others say 20 per million. Does anyone have published/ data on the professionally suggested number or range of records to be considered when spot-checking?

Excluding automation tools and data sorting(high/low), I'm specifically targeting manual record sampling: how can you quantify how much is the right amount to sample per (whatever) amount?