Article illustration 1

Operation‑Based Versioning for Structured Data

A new platform called Baseline proposes an operational differencing technique that treats changes to tables and schemas as high‑level operations. Published on arXiv on 10 Dec 2025, the paper shows how this approach can give developers a version‑control‑like workflow for databases without the overhead of a separate repository.

What is Operational Differencing?

Instead of diffing raw rows, Baseline records operations such as add_column, rename_table, or even complex refactorings. These operations become the first‑class citizens of the history graph. Because the history is append‑only, merging two diverging branches boils down to replaying the operation streams in a consistent order.

branch‑A:  add_column users (age INT)
branch‑B:  rename_table orders → purchases

When the two branches are merged, the system can detect that the add_column operation applies to users while the rename touches a different table, so the merge succeeds without manual conflict resolution.

Solving Schema Evolution

Schema changes are the bane of traditional database migration tools. Baseline’s operational model treats a schema change as just another operation, so queries written against the current schema can be operationalized—turned into a speculative branch that runs in the future. If the schema evolves between the query’s start and its execution, the differencing engine rewrites the query automatically, preserving correctness.

Baseline’s authors write that queries can be operationalized into a sequence of schema and data operations, hinting at a future where dynamic query rewriting is built into the database engine itself.

Practical Implications

  • Fine‑grained diffs: Developers can see exactly which columns were added, removed, or altered, rather than a raw row‑by‑row diff.
  • Branching without a repo: Branching is simply a copy of the current state; no separate Git‑style repository is needed.
  • Collaboration: Branches can be shared like document files, and the system’s diff/merge logic handles concurrent edits across teams.

These features could make data‑centric collaboration as painless as code collaboration, especially for data‑driven products where schema churn is frequent.

Looking Ahead

Baseline addresses four of the eight challenge problems identified in a recent survey of schema‑evolution research. While the prototype is still academic, the concepts point toward a new class of database tools that blend version control semantics with schema‑aware operations. If adopted, it could reduce the friction that currently forces teams to lock schemas in place or rely on brittle migration scripts.

Article illustration 2

Source: arXiv:2512.09762