Documentation and Audit Trails
The written record that lets a result be understood, checked, and defended long after the run that produced it.
The Record That Outlives the Run
A computed result is fleeting; the record of how it was produced is what endures. Documentation and audit trails capture the assumptions, methods, inputs, and decisions behind a result, so that months or years later it can still be understood, checked, and defended. Without this record, even the person who ran the calculation cannot reliably reconstruct what was done, and no one else can verify it.
What the Record Contains
- The model and its assumptions, including what physics was included and omitted.
- The inputs and parameters, with sources and versions.
- The verification evidence: convergence studies, order-of-accuracy results, error estimates.
- The validation evidence and comparison metrics, where available.
- The uncertainty analysis and error budget.
- The decisions made and their rationale, including choices of tolerance and mesh.
Audit Trails
An audit trail is the chronological record of what was done and by whom, linking each result to the code version, inputs, and environment that produced it. Combined with version control and provenance capture, it lets a reviewer follow a result from a published figure all the way back to raw inputs. For safety-critical and regulated work, this traceability is mandatory; for all serious work, it is what makes a result auditable rather than merely asserted.
Documentation as Part of V&V
Documentation is not a chore appended after the science; it is part of the credibility case. A verification result that is not documented cannot be relied on later, because no one can confirm it was done or what it showed. An error budget that lives only in someone's memory is worthless to a reviewer. The discipline of recording, alongside the discipline of testing, is what allows a simulation-based claim to be defended and revisited.
A well-kept record also enables honest correction: when a method or input is later found flawed, the documentation shows exactly which results are affected and what must be redone, rather than casting doubt on everything indiscriminately.