You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Ubuntu is returning different results from macos #1209
I've made a private GitHub repo here which runs code on both ubuntu-latest and macos-latest that stores the full entity dataset as artifacts so I can see exactly which variables are different (have so far established the input data is the exact same, not modified by OS behaviour).
Think I've identified the bug as in parameter uprating- for a limited subset of parameters, we're not uprating in ubuntu, but are in macos. Here's two parameter trees in text files as proof.
Closing as resolved. #1254 removed the import-time BASELINE_GROWFACTORS table responsible for the ordering-sensitive uprating. The Linux CI migration in #2115 verified byte-identical processed parameter trees on macOS arm64, Ubuntu x86 and Ubuntu arm64, with matching test outcomes. The diagnosis and verification are recorded in the reproducibility page. #1223 was closed unmerged; it is not the fix being relied on here.
Activity
nikhilwoodruff commentedon Jul 11, 2025
Think I've identified the bug as in parameter uprating- for a limited subset of parameters, we're not uprating in ubuntu, but are in macos. Here's two parameter trees in text files as proof.
Linux_parameters.txt
Darwin_parameters.txt
Document Ubuntu vs macOS reproducibility diagnosis (#1209)
Document Ubuntu vs macOS reproducibility diagnosis (#1209)
Correct the Ubuntu vs macOS cause and CI history
juaristi22 commentedon Oct 24, 2026
Closing as resolved. #1254 removed the import-time
BASELINE_GROWFACTORStable responsible for the ordering-sensitive uprating. The Linux CI migration in #2115 verified byte-identical processed parameter trees on macOS arm64, Ubuntu x86 and Ubuntu arm64, with matching test outcomes. The diagnosis and verification are recorded in the reproducibility page. #1223 was closed unmerged; it is not the fix being relied on here.