Repository navigation
JMOA 2.15 - scoped Patient RAM benchmark
Pre-release
JMOA 2.15 - scoped Patient RAM benchmark
This provisional evidence release reports a direct Patient comparison under Java 26/Leyden: 8.38 MiB lower median stopped-process PSS after the defined workload across 12 paired blocks (24 fresh units). The conditional lower saving bound is 8.19 MiB.
Calibration required a user-approved exception after the original controls failed; performance bounds remain unqualified or failed. Startup and primary cache warmth, whole-stack/cold/production parity, live ML and parity with the uploaded JMOA 2.1 binary are not established.
- All 12 paired RAM results and all 82 cost metrics are disclosed.
- Both arms used the same native heap trimming and allocator limit.
- Exact tested application, cache, core and JRE hashes are recorded.
- All 96 campaign attempts remain counted, including failures; no historical medians were added.
- Original infrastructure was restored. Saved raw evidence was independently reanalysed, which is not external replication.
Release contents
Human- and machine-readable benchmark records, this scope statement, a commit-bound evidence manifest and SHA-256 checksums. No new plugin/runtime JARs or Maven version bump are included. The version 2.15 identifies this evidence publication; the existing stable 2.1.0 binaries and PetClinic acceptance are unchanged. This release is marked provisional/prerelease and does not replace the stable release.