# HP framework TX: live background identity and operator preparation

2026-10-12. HP only; turbo remains deferred. The positive inactive-irqbalance
functional proof and all three rejected quiet attempts remain at their original
pins in [the preceding record](RECORD_2026-10-11_hp-framework-tx-irq-paused.md).
This generation changes no benchmark, SDK, source3 input, installed runtime,
host policy, activity threshold or admission rule.

## Independent readiness refusal

An independent 30.02-second readiness check completes at 02:44:06 UTC under an
actual shared/high HP lease. It uses the published future diagnostic SHA256
`edab16435c02e36ae094114b296c5ee63729769d03c5564109a0af9d997fcf6f`, with live
process context. It does not replace the active frozen preflight SHA256
`f91dc21cf715bd76c5e97cacbb16f60575de66be103d18033f393195d5e799fd`.

Load averages start and end at zero; requested/effective CPUs 16/17, observer
affinity 0-7 and kernel isolation pass. Readiness still correctly refuses because
PID 892412 (`codex`) peaks at 31.5% of one CPU. Context retained while the process
is observed at 02:43:47 UTC binds start tick 182386193, parent 2756020, process
group/session 892412 and directory `/home/yann/code/libhft`. No arguments or
environments are collected. The original 5% activity criterion remains intact;
zero native pairs start and no timed matrix or ordinary raw audit is produced.

## Temporary operator control

The [private operator profile](../../bench/llfix_reference/diagnostics/background_pause/PROFILE.md)
describes a scoped pause helper, defaulting to read-only identity inspection.
Its explicit `--run` would pause only the above exact same-user process,
delegate to the unchanged quiet4 controller under an exclusive/high HP lease,
and resume that same process identity on normal completion or interruption.
It uses PID file descriptors, validates the entire target set before signalling,
refuses replaced or already stopped processes and records partial-stop rollback.
Signals are blocked during stop registration and restoration. Hard process or
machine loss remains outside the automatic-resume guarantee.

The operator script is staged at `~/run-llfix-quiet-with-background-pause.sh`.
Actual use requires explicit user approval because that helper may serve other
sessions. No shared helper is signalled during preparation. Seven local tests
using real disposable children pass with zero skips. The same seven controls
pass on HP with zero skips, followed by actual read-only inspection matching
every target identity field. Staging completes at 03:18:12 UTC under shared/high,
after the other owner's exclusive Rust work. The home script has mode 0755 and
SHA256 `dea085deb71c14b82a2f8de8a16d2db6026b527270c03d1990be1bc3f7e4488c`.
Pause source SHA256:
`552250e94ae1110b4b233ba60307fb0c42d57741d18635f65eb12e6567cc1366`;
test SHA256:
`81d85641376bad654ad0aa5d0ebf5ce251c9e7ea229296fcb7d168b39dfbcccd`.
No other session's lease, process or service is altered. Actual pause approval
is pending; quiet4 has not started.

The existing source3 capsule, source commit
`09906965356e1eab39b3a593a35c8d2ec50354ef`, protected llfix 1.0.7 TX/peer payload
and positive functional2 proof remain the only proposed quiet4 inputs. The
older framework ELF scope remains explicit: this cannot qualify the current
framework preallocation, SDK, Stage 2 llfix 1.0.8, full-session benchmark,
engineering target or vendor reproduction.

## Retention operations

The first archive connection ends while waiting for another owner's exclusive
C++ qualification. Its local launcher is absent, the transfer is empty, and HP
has staged no retention source or archive. Only our own pending request
`72353242fae24ac792fa2c01f57d9891` (PID 1932469, this thread) is dequeued.
The empty transport, exact source, queued log, readbacks and dequeue receipt
remain preserved. No failure cause beyond these observations is asserted.

A durable fresh retry stages its source under a shared/high lease, then fails
before sealing because HP lacks the preceding second archive's **local** replica
receipt. The existing second archive and its separately verified local copy are
unchanged. Both the original exception and failed retry source remain preserved.
A fresh continuation supplies the exact already verified local receipts as new
capsule inputs and repeats the same full archive comparison. It changes only
the receipt input path, retains the original retainer/verifier and writes fresh
operation sources/logs. No fallback, altered comparison or timing grade is used.

## Verified evidence and current state

The continuation finishes at 04:06:08 UTC. The original retainer verifies both
preceding complete archives against their exact local receipts, the eleven
source3 inputs and prepared/functional2 proof hashes, plus all six staged
operator input hashes/modes and the actual home script. It verifies that no
quiet4 directory, idle receipt or pause receipt exists before sealing.

The new archive passes full stream verification on HP and locally: 55 inputs,
178,909 compressed bytes, archive SHA256
`95e12de76bb935fd213eeb164ef53a5ea1e2444b1a40d14fc870bf7e982dcc49`,
seal SHA256
`5bece895d5109978ee629de0bd58db2e8131973b3e3ee356973aa65c93b214df`.
Both copies are `~/archive/libhft/codex-llfix-reference-20261012/hp-framework-background-preparation1.tar.gz`.
The full archive retains the readiness refusal, real child-test/inspection
logs, exact operation sources and both preceding archive-operation failures.
The compact [selection index](evidence/hp-framework-tx-quiet-20261012/selected-files.json)
binds 63 exact files / 696,027 bytes, including postseal receipts/logs. No raw
native file is excluded: this generation starts zero native pairs. The verified
successful local transfer wrapper is removed; both original empty failed
transports remain preserved.

The home script and three control sources are published at `3613510cb`, preserved
on `s/codex-llfix-reference-hp-background-control-20261012`; the failure record
is published at `f53c997f8`. Seven local and seven HP child-process checks pass
with zero skips. Quiet4 and any actual pause still require the pending explicit
user approval; elapsed time or other sessions' coordination is not approval.
No exclusive capture is queued. The old active epoch remains for that future
attempt. No timing, engineering-target, vendor, current framework/SDK,
Stage 2/full-session or release qualification is claimed.