Sanderson — The Rise of Data Contracts

The article widely credited with defining the modern data-contract movement. Agent-fetched 2026-06-26 to ground Data Contract.

Summary

Chad Sanderson (who built the first at-scale data-contract implementation at Convoy) defines data contracts as "API-like agreements between Software Engineers who own services and Data Consumers" to produce well-modeled, trusted data. The problem: production databases are treated as "nonconsensual APIs" — engineers never agreed to expose operational tables for analytics, so a producer can drop a column with no warning and silently break downstream (his stark example: "the pricing algorithm that generates 80% of your company's revenue is broken after the column is dropped in a production table upstream"). The fix is to set expectations at the producer–consumer boundary via strongly-typed schemas and versioning, so producers commit to semantic contracts instead of consumers "stitching the world together retroactively through SQL."

Andrew Jones (GoCardless) is widely credited as a parallel originator of data contracts and wrote the O'Reilly book Driving Data Quality with Data Contracts (2023). He is not named in this particular article (noted from the maintainer's 2026-06-26 web research, not from the source itself).

Key claims

• A data contract is an explicit producer↔consumer interface: schema + semantics + ownership.
• Without it, operational DBs are "nonconsensual APIs" and upstream changes break downstream silently.
• Push the commitment to the boundary rather than reconstructing meaning downstream in SQL.

Connections

• Data Contract — confirms the heuristic; supplies the "nonconsensual API" framing and the boundary-enforcement rationale ("don't re-stitch meaning downstream").
• Pipeline Legibility — a contract is the producer's promise; legibility is the consumer reading the lineage honestly. Two halves of trust.
• Do It Once, at the Right Place — "set it at the boundary, don't re-derive downstream" is the contract face of doing it once.

Source

• Original: Chad Sanderson, Data Products (Substack), 2022-08-22 (source-url above); faithful capture in source-file.