The Reproducible Analysis Record
Kronos keeps a reproducible record of its analyses — code and data that a third party can re-run to get the same numbers — because a result you cannot reproduce is a rumor.
Reproducibility as honesty
A number is only trustworthy if someone other than its author can regenerate it. Kronos maintains a reproducible analysis record: the code, inputs, and data behind its design points, deposited so an independent party can re-run them and obtain the same values. This turns 'trust us' into 'check us'.
How the record is built
- Analyses are versioned and deposited to permanent archives with DOIs
- Two tiers of reproduction: byte-for-byte where possible, tolerance-based where floating point makes that impossible
- Cold-repro runbooks describe environments so results regenerate on a fresh machine
- Design points quoted on this site trace back to entries in that record
What reproducibility does not do
Reproducing our calculation confirms that our reported number follows from our code and assumptions. It does not confirm that our assumptions match the real machine — that is the hardware gate. We are careful to claim only the first: our published results are reproducible, and they remain design points until measured.
A reproducible record is the minimum a program owes the people it asks to believe it. It is also self-discipline: knowing an analysis must survive an outsider re-running it changes how carefully it is built.