JMOA 2.0.0

JMOA v2.0.0

Stable v2.0.0 closes the V2 launch matrix under service-specific confirmed
runtime policies.

Service	Confirmed policy	Median V1-to-V2 PSS
PetClinic customers-service	NO_CDS_LOW_DIRTY	-6,012 KB
Doctor-service	APPLICATION_CDS	-5,156 KB
Patient-service	JDK_BASE_CDS_LOW_DIRTY	-8,279 KB

The release does not claim one universal runtime policy. Patient no-CDS is also
independently confirmed at -8,903 KB; dynamic Patient application CDS remains
blocked. Stock JDK base CDS and a Patient application archive are distinct
policies.

The tag and source archive are public. Frozen PetClinic profile/admission inputs
are not currently uploaded as GitHub release assets, so the qualified quickstart
has explicit external prerequisites.

Release: https://github.com/AlphaSudo/jmoa/releases/tag/v2.0.0

JMOA 2.0.0-rc2

JMOA 2.0.0-rc2

RC2 is a release-hardening candidate. It publishes the plugin/runtime artifacts
with checksums and a sanitized public evidence archive for the final PetClinic
customers-service comparisons.

Reproducible Claim

Under the exact public EXPLODED_BOOT_APP, NO_CDS_LOW_DIRTY,
MALLOC_ARENA_MAX=1, no-CDS/no-javaagent protocol, final V2 improved on
finalized V1 across 2/3 paired runs:

• median PSS: -6,012 KB
• median Private_Dirty: -5,708 KB
• median memory.current: -8,081,408 B

V2-C returned CONFIRMED_WIN; V2-D identifies heap page-touch reduction as
the primary attribution. This is an incremental, service-specific claim. It is
not a startup, universal Spring Boot, fat-JAR, CDS, or direct B0-to-V2 claim.

Direct Baseline Boundary

The historical B0-to-V2 three-pair record is preserved for provenance, but its
raw capture was unavailable. A fresh exact-image five-pair replication was
valid but mixed (2/5 PSS wins, median PSS +585 KB), so RC2 does not claim a
direct baseline win.

Consumer Qualification

The published RC1 assets were downloaded, checksummed, installed into an
isolated Maven repository, and used for a clean PetClinic quickstart through
exploded-Boot materialization and semantic smoke. RC2 carries the same
qualification path plus the public evidence archive.

JMOA 2.0.0-rc1

JMOA 2.0.0-rc1

JMOA V2 closes the public optimizer workflow from profiling and admission
through artifact materialization, evidence validation, memory attribution, and
an opt-in raw LVT/LVTT reducer.

Confirmed Memory Matrix

• PetClinic customers B0 to V2: median PSS -8,956 KB, Private_Dirty
  -8,620 KB, memory.current -11,247,616 bytes, 2/3 paired wins.
• PetClinic customers V1 to V2: median PSS -4,812 KB, Private_Dirty
  -4,512 KB, memory.current -6,791,168 bytes, 3/3 paired wins.
• Public PetClinic visits raw reducer: median PSS -2,012 KB, scope-bound.
• Private Doctor D2 to D2R: median PSS -5,156 KB, scope-bound.

The primary PetClinic customers comparisons used cache-controlled balanced
three-pair runs. V2-C accepted all runs and V2-D attributed both primary wins
mainly to heap page-touch reduction.

Distribution

The GitHub Release contains plugin/runtime JARs, POMs, source JARs, checksums,
and the three public PetClinic reproduction inputs. Maven Central publication
is not claimed for RC1.

Boundaries

• No universal, startup, CDS, or fat-JAR memory claim is made.
• Generated/proxy mutation remains deferred to V3.
• Broad bytecode reducers remain disabled; V2 ships only the opt-in raw
  LocalVariableTable/LocalVariableTypeTable reducer.