The Reproduction Environment

The breeder analyses pin their software environment — interpreter, libraries, and versions — so a rerun is not silently changed by whatever a reproducer happens to have installed.

Pinning what the code runs on

A number can change because the code changed, or because the environment underneath it changed — a different library version, a different floating-point path, a different platform. To make reproduction meaningful, the breeder (Hyperion) analyses pin the environment as tightly as the code.

• The interpreter version is fixed.
• Every library dependency is pinned to an exact version.
• Random seeds are recorded for any stochastic step.
• A container or lockfile captures the whole environment for one-command rebuild.

# Environment is captured, not assumed
# - lockfile pins exact dependency versions
# - container image freezes interpreter + libs + OS libs
# - each analysis records the seed it used
# Reproduce with the pinned environment, not your local one.

Why byte-exactness is hard

Even with pinned versions, different CPUs and math libraries can produce tiny floating-point differences. That is why reproduction is defined at two tiers: a Tier-1 byte-exact match where the environment is identical, and a Tier-2 within-tolerance match where it is not. Both are legitimate; the record states which tier a given analysis is expected to meet.

Longevity

Pinned environments age: libraries are deprecated, platforms move on. The container approach mitigates this by freezing the whole stack, and the versioned deposit means an old environment specification remains retrievable alongside the numbers it produced.

This page describes a design-and-simulation study, not a built machine. Construction of the breeder (Hyperion) begins Q2 2027; first-of-a-kind first tritium is targeted near 2030. No hardware net-gain claim is made before FOAK.