A healthcare analytics team wants to load customer records containing credit card numbers into BigQuery for fraud analysis. The data scientists must be able to correlate transactions belonging to the same card across different tables (referential integrity), and an authorized security team must occasionally re-identify specific cards during confirmed fraud investigations. The solution should keep the tokenized values in the same numeric format as the original card numbers so downstream schemas remain unchanged. Which Sensitive Data Protection (DLP) de-identification technique should you configure?