Environment Pinning
Code that runs differently on different machines is not reproducible, so deposits pin their software environment to make a cold rerun match.
The hidden variable in reproduction
The same code can produce different results on different machines when its dependencies differ in version or behavior. That variability is the enemy of reproduction. Kronos pins the software environment of each deposit — recording the exact versions of the tools and libraries the pipeline needs — so an outside reader can recreate the environment and get the same answer rather than a subtly different one.
What pinning records
- The language runtime and its version.
- Every library the pipeline depends on, at a fixed version.
- Any configuration that affects numerical behavior.
- Enough of the environment that a fresh machine can be brought to the same state from the runbook.
Why it enables the two-tier check
Environment pinning is what makes the two-tier check meaningful. If the environment were free to vary, even a correct reproduction might drift outside tolerance, and a real error might hide inside it. Pinning narrows the sources of difference to the ones that matter, so a Tier 1 byte match is achievable where it should be and a Tier 2 tolerance match is trustworthy where it is used.
Pinning plus multiple environments
Pinning does not mean testing in only one place. Kronos exercises deposits across more than one independent environment; the AI and quantum deposit follows a multi-environment cold-reproduction procedure. The combination — pin the environment, then confirm the result holds across several — is how a reproduction is shown to be robust rather than lucky. See the cold-reproduction runbook.
A reproducible result must specify the environment it reproduces in. We pin ours so a stranger can match it.