Version Control and Provenance
Recording exactly what was done, by whom, and when turns a computation from an unrepeatable event into an auditable, recoverable record.
A Time Machine for Work
Version control tracks every change to a body of files over time, who made it, when, and why, and lets any past state be recovered exactly. It is the foundation of collaborative and reproducible work: nothing is silently overwritten, every change is attributed, and any prior result can be reconstructed from its recorded state.
What It Enables
- Recovery: return to any earlier state precisely.
- Attribution: know who changed what and why.
- Parallel work: many contributors on separate lines, merged deliberately.
- Tagging: mark the exact state that produced a specific result.
Provenance
Provenance is the full lineage of a result: the code version, the input data, the parameters, the environment, and the commands that produced it. Version control captures the code side; complete provenance extends it to data and environment. A result with full provenance can be traced back to every ingredient that made it, and therefore checked and regenerated.
Why It Matters for Science
A published number that cannot be traced to a specific code version, input set, and environment is not reproducible, and an error found later cannot be scoped, which results does it taint? Provenance answers that question. It is also the audit trail that lets a claim be defended: this figure came from this exact state of this code.
At Kronos
Frozen physics results are tied to tagged code versions, checksummed inputs, and recorded environments, so a design number carries its full lineage. If a model changes, provenance identifies exactly which prior results are affected, and the re-freeze is deliberate and documented rather than a silent overwrite.