# Primitive JMH lane parked outside the public harness, 2026-10-03

The user asked whether the separate JMH version added value and should be
removed from the public repository. The active 25-adapter microbench uses the
custom Stage 2 harness, including Java. The JMH utility class implements its own
ASCII scan/parse/format/checksum methods and does not call a libhft parser. Its
diagnostic costs cannot supply an engine comparison or converted-field parity.

## Boundary and preservation

The launcher, Java class, original tests and guide moved to corresponding
`internal/` paths. The original project pom is retained beside them, and
[the parking note](../../benchmarks/retained/java/fixbench/README_JMH_PARKED.md) explains replay
from the preserved complete capsule/private source ref. These are historical
snapshots outside active build/test paths, not a newly maintained private project.
The three executable source files are byte-identical to private main
`24ed18574386c56a19597eec3271600d9cc8d1bd`.

The public guide link was removed. The public fixbench pom lost only the unused
`jmh.version`, JMH core and annotation processor declarations. A semantic XML
comparison verifies every other property/dependency/profile/plugin unchanged.
Fullbench's pom had no pending edits in the ownership check; the narrow scope
was communicated in the existing archive coordination note. No manifest edit
was necessary: its existing boundary already excludes `internal/`.

The original source refs and HP functional evidence remain unchanged under
`~/archive/libhft/codex-microbench-jmh-20261003/` on BEELINK2 and HP. Its seven
real utility results retain their diagnostic grade. The source pin and raw
runtime manifests were not rewritten. The existing SCAN field-parser ranking,
all Stage 2 source/contract/preparation/grading bytes, SDKs, full-session runtime,
canonical workload and common helper APIs are unchanged by this patch.

## Checks

An actual complete manifest materialization verifies that the exported tree
contains no parked JMH launcher/class/guide/tests, no JMH imports in active Java
sources, and no JMH dependency/version declarations. The three preserved source
files match their original Git bytes. Actual archive Stage 2 `--list-engines`
still returns 25 adapters and 320 required probes, even with missing unselected
SDK inputs and an invalid unused build-job setting, with no source/output writes.

**59 tests pass, zero skips**: build-output/toolenv, public parking boundary,
layout, output defaults and complete-archive controller/import checks.
The earlier build-output scanner failure from 73e0cb8af is retained privately;
its active offender paths no longer exist. The source scanner passes. The
intermediate comment-only repair also passed 45 checks without changing either
executable AST, before the user chose to park the lane instead.

All new/adjusted private guide and record links resolve. Public readiness at the
parking source selects 737 files, retains 88 scrub findings in 21 files and
85 textual escapes, and has zero dead documentation links. Whole-harness release
admission remains pending. No new build, JMH measurement, SDK preparation or
SCAN/HP CPU job was started for parking.

Verification/source/publication evidence is retained on both hosts under
`~/archive/libhft/codex-microbench-jmh-parked-20261003/`, separately from the
unchanged original functional generation.