Automated Verification of Results
Automated checks re-run analyses and compare outputs to expected values, so a claim and its evidence never silently drift apart.
The problem it solves
Codebases change. A refactor, a library update, or a data fix can quietly alter a result that a paper or a design decision depends on. Automated verification runs the analyses continuously and flags any output that moves outside its expected range, so drift is caught immediately rather than discovered later.
What gets checked
- Key numbers reproduce to a stated tolerance.
- Physical invariants, like conservation laws, still hold.
- Results agree across methods where two methods exist.
- Published figures match the code that generated them.
Two tiers of reproduction
Verification distinguishes exact reproduction, where outputs match to the byte, from tolerance reproduction, where outputs match within a stated numerical band. Both are legitimate; what matters is stating which applies to each result, so agreement is judged against the right standard.
Continuous integration for science
These checks run automatically whenever the code changes, the scientific analog of continuous integration. A failing check blocks the change until the discrepancy is understood, keeping the analysis record trustworthy.
For Kronos claims
The frozen numbers behind the Hyperion breeder and the burner are guarded by these checks, so a published figure like TBR 1.8 or a neutron fraction of 5.44 percent stays tied to the code and inputs that produced it.
Discipline over trust
Automated verification replaces trust in individual care with a system that checks continuously, which is what makes reproducible science durable rather than aspirational.