# Registered canonical physical generations on HP

The high-priority `isol-no-shared` collection completes on HP's Solarflare DAC
between its two fixed namespaces. Callback CORE and Philadelphia-fast each
complete four tiny and twelve full pairs: 32 actual pairs, 240,160 primary and
scheduler-delay samples, and 660,500 original application frames per peer.
Every body, owner/fold, administration, sample-count, namespace/CPU, normal-user
entry and physical counter/polling proof passes. Both runtime generations and
the frozen 855-file producer remain unchanged; no new compile is performed.

| Engine | NOS/ER | MD | Mixed | Tick-to-trade |
|---|---|---|---|---|
| callback CORE | S5 pass | S5 pass | S5 pass | S5 pass |
| Philadelphia-fast canonical | S5 diag | S5 diag | S5 diag | S5 diag |

Philadelphia retains rate and scheduler-delay failures on all four flows and
full percentile-stability failures on NOS, MD and mixed. The target rate,
warmup, sample counts and grade thresholds are unchanged.

Four original profile cold receipts and four relocated SDK-free receipts
agree exactly, including quantiles, actual maximum extrema and every original
gate. Independent CORE/NexusFIX cold review also passes all four flows on this
rig. Its source, full profile, configuration, CPU masks, original peer manifest
and compatible tick clocks are checked separately. The combined two-rig review
has its own [record](RECORD_2026-10-14_cpp_callback_two_rig_comparison.md).

The physical producer completed with its final marker and queue exit zero.
Its local SSH connection then remained stale and was closed only after verifying
completion. The earlier matched-review waiter's remote process disappeared
before producing any file. After confirming the session-owned review directory
was empty, the cold review was recovered under a high-priority shared lease.
This recovery performs no compile, socket pair or benchmark repeat. Recovery
metadata and original start failures are retained. The shared-slot export
completed before a proposed queue change; a replacement export refused the
existing local evidence without changing it.

The seal retains 1,847 inputs, 372,512,999 bytes; SHA-256
`c2ee99217a39055bc474a376745766b9be43ae9ab84c9abb368c9b514bd75b87`.

[Original physical matrices, profile cold receipts, relocated cold receipts,
matched review, source and recovery records](evidence/registered-generation-hp-physical-20261014/retention-seal.json)

Java matching, other comparator gates and whole-harness release remain open.