# Java callback physical comparison on both rigs — 2026-10-14

Both accepted physical DAC rigs complete the separately registered Java and
Philadelphia callback generations from the exact clean `c176e7512fd8` controller.
The 112 actual functional/physical pairs retain 481,280 primary samples, 481,280
scheduler delays and 1,323,800 original application frames per peer. There are 64
physical pairs, including 48 full report pairs. All complete fixture, full owner,
prepared/callback, administration, namespace, per-thread CPU/runtime, exit and
physical-counter checks pass; payloads run as the normal user.

HP's individual decisions remain as recorded in the
[HP originals](RECORD_2026-10-14_java_callback_hp_physical.md): Java is diagnostic on
all four flows, while Philadelphia passes physical MD and mixed. On SCAN both
engines remain diagnostic on all four flows. Java misses rate everywhere; NOS/MD/
mixed also miss stability, and MD/mixed miss delay. Philadelphia misses rate on
NOS/mixed/tick, MD misses stability, and tick also misses stability. No gate is waived.

The actual **two-rig** comparison independently replays both full generations on
both rigs. Every matched flow is blocked, with empty metric/ratio objects. Full
profiles, physical topology, CPU budgets, actual callback spans and all source/config
bindings are matched; this is a timing-admission refusal, not borrowed clock evidence.

The initial cold call correctly refused the experimental in-progress grading source
because it differed from the producer. That refusal is retained. The corrected loader
verifies the literal published source pin and inventory/archive hashes before selecting
that exact grading policy. It leaves the source-match check intact and changes no
original data. All eight decisions remain blocked, without compilation or new sockets.

The [immutable seal](evidence/java-callback-two-rig-20261014/retention-seal.json) binds
2,203 inputs / 372,976,161 bytes; SHA-256
`2c51404de284af4d5cbfb309ec4a4994d245d140d9c600dd64b46c095a2319db`.
It retains every SCAN original, the combined report and its original refusal/loader,
complete clean producer source and explicit bindings to the unchanged HP seal. Current
Java minimum admission remains incomplete.