# What is referential integrity in data? > What is referential integrity in data? The rule that every link points at a real record. Why Spirit's data sale and its union objection turned on it. By Matt Burns, Vaultfish. Canonical: https://vaultfish.com/answers/what-is-referential-integrity-in-data Published: 2026-10-02 Updated: 2026-10-14 ## The short answer Referential integrity is the rule that every link between records points at a real record. In AI data deals, Spirit Airlines kept links while de-identifying, and the union said links raise re-identification risk. The UK ICO says pseudonymised data stays personal data for anyone with the key. Two groups use the term for opposite reasons. Database engineers want the links kept. Privacy lawyers want to know who can follow them. ## What does referential integrity mean in a database? A foreign key constraint says the values in one column must match values in a row of another table. [PostgreSQL's documentation](https://www.postgresql.org/docs/current/ddl-constraints.html) calls this maintaining referential integrity between two related tables. Its example has an orders table that points at a products table. Once the constraint exists, no order can name a missing product. ## Why do AI companies want the links kept? A trajectory only shows up when records connect. [Fortune](https://fortune.com/2026/09/14/micro1-tries-to-beat-googles-bid-to-buy-bankrupt-spirit-airlines-data-for-12-5-million-highlighting-value-of-data-to-ai-companies/) described the value in Spirit's archive this way: an IT ticket linked to a Teams discussion, a code change and an operational result. That works only if the links can be rebuilt. Our [trajectory entry](/answers/what-is-a-trajectory-in-ai) explains the term. micro1's David Revland, who works on human data, said in an undated recorded forum talk that micro1 swaps each name for a consistent stand-in across every platform. The same person gets the same stand-in in Slack and in Google Workspace. micro1 staff said in the same talk that companies cannot pre-scrub. Redacting first destroys the cross-platform trail it needs. ## Why does it worry privacy lawyers? A stand-in that stays the same across tools is a persistent identifier. The [ICO](https://ico.org.uk/for-organisations/uk-gdpr-guidance-and-resources/data-sharing/anonymisation/pseudonymisation/) says pseudonymised data is personal data in the hands of anyone who holds the additional information. That extra information must be kept apart and secure. The [FTC wrote](https://www.ftc.gov/policy/advocacy-research/tech-at-ftc/2024/07/no-hashing-still-doesnt-make-your-data-anonymous) that hashes are not anonymous and can still be used to identify users. Both points ask one question: can a link be followed back to a person? ## What did the Spirit union say? The union's objection (Doc 1489) quoted section 3(c), the de-identification clause in the sale agreement. It said the clause keeps the joins among timekeeping, pairing, training, payroll and communications data by design. It made no technical claim that a particular record can be re-identified. [Time](https://time.com/article/2026/08/25/google-spirit-airlines-ai-data-RL/) reported that the union warned even anonymized information about members could be reconstructed. Google told Time the data would be rigorously scrubbed by a third party. Our page on the [union objection](/answers/spirit-airlines-data-union-objection) lists what it asked the court to do. ## What should a company ask about stand-ins? Ask who holds any key that maps stand-ins to real people. Ask whether stand-ins repeat across systems. Ask what the AI company promises about re-association. [Pseudonymization versus anonymization](/answers/pseudonymization-vs-anonymization) and [re-identification risk](/answers/re-identification-risk-emails) cover the legal test. Run the [rights check](/rights-check) before you list anything. We give no legal advice, so talk to a privacy or commercial lawyer in your state before you sign. micro1 may pay us a referral fee if a company we introduce goes ahead with it. [How we are paid](https://vaultfish.com/disclosure) Figure: steps Real name found, swapped -> Same stand-in in every tool -> Cross-tool trail survives. Revland (micro1) said in an undated recorded forum talk that one consistent stand-in per person keeps the trail across tools intact. [Apply to micro1's data partnership (paid link)](https://req.micro1.ai/post/74520e8b-4673-4da5-b855-53609aaf6710?referralCode=f82e519e-5d64-4e0e-9574-760c84371cf2&utm_source=referral&utm_medium=share&utm_campaign=data_referral) ยท [micro1's own data-partnerships page](https://www.micro1.ai/data-partnerships) _Paid link. [How we are paid](https://vaultfish.com/disclosure)_ _Vaultfish helps you run the process. We are not lawyers and this is not legal advice. If you need counsel who knows this area, we can point you to one._ ## Sources - [PostgreSQL's documentation](https://www.postgresql.org/docs/current/ddl-constraints.html) - [Fortune](https://fortune.com/2026/09/14/micro1-tries-to-beat-googles-bid-to-buy-bankrupt-spirit-airlines-data-for-12-5-million-highlighting-value-of-data-to-ai-companies/) - [ICO](https://ico.org.uk/for-organisations/uk-gdpr-guidance-and-resources/data-sharing/anonymisation/pseudonymisation/) - [FTC wrote](https://www.ftc.gov/policy/advocacy-research/tech-at-ftc/2024/07/no-hashing-still-doesnt-make-your-data-anonymous) - [Time](https://time.com/article/2026/08/25/google-spirit-airlines-ai-data-RL/)