# NexusFIX required values and measured owner extraction

The full-session adapter now reads required values through the pinned SDK's
`FieldView` integer, quantity, price and character APIs. It rejects missing,
empty, duplicate, malformed, fractional integral quantities and out-of-range
values. Snapshot extraction requires every ordered bid/ask triple and the
declared entry count. Requests have no fallback symbol/depth; ACKs and tick
reactions must correlate with the owning request before their counters advance.
Callback validation failures become failed diagnostics rather than escaping
the SDK's `noexcept` callback boundary.

Generated payloads use the common `LCOM1` symbol, decimal sequence markers,
quantity/price tables, constant ACK identifiers and fixed depth book. Prices
are constructed as exact SDK fixed point integers instead of truncating a
binary floating point initializer. The measured owner resets its sink and
payload sequence at each measured run; warmup and earlier runs are excluded.
Rows declare the last measured run, fold count and `canonical_spec_v1` fixture,
while retaining `fixture_pass=false` and a diagnostic contract status.

## Original HP qualification

Source **`ccee2b3a53136ba02b0f826491aaef5626fcc13b`** is retained on
`s/codex-fullbench-nexusfix-owner-source-20261003`. The clean managed NexusFIX
reference is **`ffa8b0a014b5cb4020ec8a99fe22fa3a24bfc494`**. Its 91 include files
were archived directly from that Git revision into this session's immutable
capsule. This is a separate full-session build; the microbench session's
unpacked SDK, frozen runtime and timing campaign were untouched.

HP's status/process/memory check preceded a high-priority `host-cpu-slot wait
--pool isol` lease, permitting concurrent shared work. The foreground controller
built the driver and cold fixture with Clang 21.1.8, C++23, `-O2 -DNDEBUG
-mavx2`, the declared repository compatibility include, SDK headers and common
harness. It ran the existing self-test, **48 cold SDK codec cycles**, **51
malformed-payload rejections** and **16 actual local TCP pairs**, with zero skips.
The pairs cover all four flows at `(warmup,runs,samples)` values `(0,1,12)`,
`(5,2,12)`, `(3,3,20)` and `(7,1,100)`. Every measured owner matches the
independent common sink and fold count, all samples are retained, and both
roles exit successfully. All **718 source files** and **91 SDK headers** remain
unchanged after the run. The complete argv, logs, source/header/binary hashes,
unaltered fixture transcript, primary raw samples and original JSON report are
retained in [the evidence bundle](evidence/nexusfix-values-20261003/hp-ccee2b3a5313/SHA256SUMS).
Compiled products also remain in HP's session archive.

The local source/config/DAC fixture checks pass **89 checks**, with zero skips;
the source comprehensiveness gate passes. The existing static guards were
updated to follow the SDK validator/folder and fixed point touch price; the
driver was not changed to reproduce retired function names.

## Remaining gates

Numeric value parity is separate from byte spelling. The untouched SDK strips
trailing price zeros; the cold receipt therefore compares exact extracted
values and retains every original wire-format difference. None of the 120
cold messages passes the minimal model's complete body alignment:

| Type | Remaining body differences |
|---|---|
| D | SDK defaults emit 21, 40 and 59; price spelling differs |
| V | SDK default emits 266; group/body order differs |
| W | Price spelling differs |
| 8 | Body order and price spelling differ |

At this source generation, the dump rebuilds a builder with placeholder
headers and the tick start clock precedes SDK serialization. The subsequent
[production application capture and clock qualification](RECORD_2026-10-03_nexusfix_production_wire.md)
replaces those two paths and retains separate fresh evidence. The common
OrdType/dictionary choice, complete admin and scheduler-delay observations,
physical DAC rate/stability and release qualification remain unmet. These functional runs grant no Stage
5 pass and do not revise historical benchmark grades or installed payload pins.
