# Explicit Java reply-owner comparisons — 2026-10-17

The independent Java/Philadelphia physical reader and cross-rig merger now
accept an explicit `--reply-owner callback` policy. Their default remains
`poll`, including when reading original reports that predate the policy field.
Callback admission requires its own `java-epsilon-callback` producer root,
actual callback owner observations, exact JVM arguments and the distinct
`+reply-cb` runtime variant at every endpoint. The declared policy is saved
in each new report and its actual runtime facts.

The merger requires the same explicit declaration and actual owner on every
rig. A legacy poll report, missing callback declaration, changed runtime
owner, missing JVM property or mixed variant cannot supply callback admission.
The existing source/archive/container/member, full-sample, original producer,
host collector, CPU placement/readback, clock and cold replay gates remain.
Failed rows keep their blockers and no metrics or ratio. Whole comparison
admission still requires both declared physical rigs.

**111 controls pass, zero skips** across the independent reader, archive
identity, merger and original callback comparison suites. New controls cover
both matched owner policies, refusals in both directions, incomplete endpoint
proof, rehashed cross-rig owner substitutions and a blocked callback flow.
The first callback test fixture placed its JVM property in the wrong order;
the existing exact-arguments gate correctly refused it. The fixture now
uses the actual builder's order; that production gate is unchanged.

Both hosts separately pass the actual callback consumer's 56 checks, twelve
TCP pairs and 107 cold refusal controls against their original fresh public
SDK. Both input inventories remain unchanged. The physical runs are queued
(HP high priority). These reader controls and functional qualifications
supply no physical timing result or current Java minimum. Fresh original
callback reports must pass the new explicit reader and archive binding
before any compact comparison can be published.

## Both original physical rigs independently replayed

The original poll and callback configurations now finish on HP as well as
SCAN. On each host, each explicit reader passes 111 controls and independently
replays both original 120,000-sample report collections. Installed product
collectors bind all original JAR containers and complete canonical member
inventories. Each host's original report and archive receipt is downloaded
with exact size/SHA-256 checks; no raw sample/capture download or new physical
producer is involved in combination.

The frozen merger combines the two original per-host decisions, requiring
two distinct accepted physical rigs, the same frozen source, actual runtime,
reply-owner policy, complete compiled members and wrappers. Raw JAR container
hashes remain distinct where ZIP packaging differs. Results:

| Explicit Java reply owner | HP matched flows | SCAN matched flows | Whole two-rig comparison |
|---|---|---|---|
| poll | none | none | blocked |
| callback | NOS/ER, MD, tick to trade | none | blocked |

All failed decisions retain their original blockers and empty metrics. The
three HP callback flow metrics match the independently admitted original
compact cells; they do not admit a Java minimum across both rigs.

The exact combined downloads are
[`java-callback-poll-two-rig-20261017.json`](comparisons/java-callback-poll-two-rig-20261017.json)
(17,090 bytes, SHA-256
`698c7faf59b5bb12cfc71abb1ea99527546ac7a245418eb26620b4faea78790e`)
and
[`java-callback-reply-two-rig-20261017.json`](comparisons/java-callback-reply-two-rig-20261017.json)
(19,144 bytes, SHA-256
`3a6b279b15b7ab090f4f9cab84831309b432bb4c68785e3d0c9da68595e46cfd`).
Both original SCAN-only decisions stay intact. The changed ASCII numeric
consumer remains a separate experiment; it cannot borrow these products,
source identities, original grades or qualification.

## Subsequent numeric-source publication

The independently replayed numeric source now replaces seven passing
compact keys and supplies a separate two-rig decision: HP MD/tick match,
SCAN none, whole comparison blocked. This source's exact old callback
report is retained above unchanged and marked superseded for current web
publication, so its original ratios cannot borrow the replacement cells.
The poll and original SCAN-only diagnostic downloads remain unchanged.
See the [new source, exact decision and publication scope](RECORD_2026-10-17_java_ascii_numeric.md).