Existing error correction systems clean tables in isolation, requiring separate rules or labels for each table. For large sets of tables, this approach is costly and does not scale. At the same time, tables in a collection may be related, creating both opportunities and risks for sharing correction signals. We present a systematic study of multi-table error correction, characterizing table relationships along two structural dimensions: schema- and record alignment. We also introduce two multi-table correction strategies: post-merge correction, which combines related tables before cleaning, and cross-table correction, which shares cell-level signals across tables. Both are evaluated against isolated correction across different scenarios, error types, and data lakes. The results show that the effectiveness depends strongly on table relationships and error types. Joining with clean keys and induced dependencies can yield gains, as well as unions with overlapping records. We compile these findings into guidelines for choosing when and how to use multi-table correction.