# NexusFIX production application capture and prepared tick clock

The full-session adapter records actual SDK session bytes through the public
`SessionCallbacks::on_send` and incoming `ParsedMessage::raw()` APIs. The
one-cycle dump no longer rebuilds messages with placeholder headers. An
optional `--wire-transcript` retains all application frames in each direction;
session administration is outside this capture scope. The initiator installs
its public session callback after successful startup and before its first poll.
The pinned SDK headers remain unchanged.

The tick owner arms its start clock before calling the SDK. Serialization and
optional capture finish before the serialized-send callback records the clock,
immediately before `TcpSocket::send`. Only successful full-span sends count as
prepared ticks. The end clock remains the first action at application callback
entry, before benchmark extraction. Rows declare both clock boundaries,
capture source/scope and last measured run's prepared tick count. Warmup and
earlier measured runs remain excluded from the owning sink.

## Original HP functional qualification

Tested source **`67cc59c32fb1930a114a520ffd699911f6e4c8e3`** is retained on
`s/codex-fullbench-nexusfix-wire-source-20261003`. The separate immutable capsule
contains **722 source files** and **91 pristine SDK headers** at upstream pin
**`ffa8b0a014b5cb4020ec8a99fe22fa3a24bfc494`**. It does not use the microbench
session's frozen SDK or runtime.

HP's status/process/memory check precedes a high-priority `host-cpu-slot wait
--pool isol` lease, allowing concurrent shared work. The outer foreground
controller runs on CPUs 8–17. Clang 21.1.8 builds the driver and cold fixture
with C++23, `-O2 -DNDEBUG -mavx2`, the declared compatibility and harness
headers, and no SDK patch. Both builds, the adapter self-test, **48 cold codec
cycles**, **51 malformed-payload rejections** and **20 real local TCP pairs**
pass, with zero skips.

Each flow covers `(warmup,runs,samples)` values `(0,1,12)`, `(5,2,12)`,
`(3,3,20)` and `(7,1,100)` with capture, plus `(2,2,12)` without capture.
The **16 captured pairs retain 2,220 application frames per peer**. Client and
server transcripts match byte for byte. Every frame passes production header,
BodyLength, checksum, identity and dynamic SendingTime checks; per-sender
session sequence numbers increase. Required values and payload markers follow
the independent common fixture across warmup and measured resets. Price
normalization is restricted to the numeric required-value check; original
body differences remain in the report.

All 20 measured owners match independent sink/fold counts and retain all
**880 primary samples** across runs. Tick owners retain the declared prepared
send count and clock boundaries, including capture-disabled controls. All
source and SDK bytes remain unchanged. The driver SHA256 is
`934118953846e9a389eeddf75864faa174638ad7da08b5bd61d23a59025ed16b`;
the cold fixture SHA256 is
`fc31c761874cd54318a7c0454ec3abdc6b554ea3fc441523b1114138769cbc94`.
Compiled products and original receipts remain under
`~/archive/libhft/nexusfix-wire-hp-20261003-67cc59c32fb1` on HP.

[Original runner, immutable input receipt, report, logs, bilateral wire
transcripts and raw samples](evidence/nexusfix-wire-20261003/hp-67cc59c32fb1/SHA256SUMS)
are sealed without modification; the original HP seal also retains the binary
hashes. The local SDK/scenario/configuration/DAC checks pass **113 checks**,
with zero skips, and the source comprehensiveness gate passes.

## Remaining gates

Production application capture and prepared tick boundaries now have
functional evidence. No cold or production application body matches the entire
minimal model: SDK default extra fields, price spelling and V/8 body order
remain different. The common OrdType/dictionary choice remains pending.
Complete admin observations and scheduler-delay samples, physical DAC rate
and stability, timing and release qualification remain open. The CPU lease
permits shared work and these are tiny functional runs. Every row retains
`fixture_pass=false`; no Stage 5 pass, installed payload pin or historical
benchmark grade changes.

The subsequent [delay/extrema qualification](RECORD_2026-10-03_nexusfix_scheduler_delay.md)
adds paced delay retention and correct maximum reporting with fresh receipts;
the remaining body/admin, physical and release gates are preserved.