# NexusFIX exact full-session order bodies

Recorded 2026-10-04 for [the fullbench plan](../../../../docs/bench/PLAN_FULL_BENCH.md)
and [standalone harness export](../../../../docs/bench/PLAN_BENCHMARK_EXPORT.md).
Source `ce5ae1074a25242cb4197f8e063105e01569d4f0` is preserved on
`s/codex-fullbench-nexusfix-orders-qualified-20261004`.

## Source correction

Generated and causal D messages now use NexusFIX's unchanged public
`MessageAssembler`, as the already qualified V/W/8 builders do. The D body
has the common field order `11,55,54,38,40,44`, limit OrdType `40=2` and exact
five-decimal Price. Required TransactTime remains present. Optional
HandlInst/TimeInForce are omitted. Sequence-dependent payload variation,
parsed-tick causality, quantity clamps and original integer-scale price
values remain unchanged. SDK sessions still supply headers, envelopes,
checksums and transport.

The D extractor rejects missing, empty, malformed or non-limit OrdType.
The ACK extractor retains the common value checks without requiring OrdType
on `35=8`. Existing owner folds and the shared golden reference are unchanged.
No SDK header, Stage 2 adapter, shared build/clock API or installed generation
changes in this pass.

The cold checker retains the original 51-rejection diagnostic scope. A new
full-body mode requires the explicitly declared 57-rejection set, including
six added OrdType cases, and exact original D/V/W/8 wire bodies. Existing
synthetic stock-order and five negative-control outputs keep their original
bytes; they cannot borrow the new full-body admission. Numeric normalization
cannot satisfy the full-body check. All fixture/timing/Stage 5 flags remain false.

## Actual complete-archive verification

The frozen harness contains **755 files**, including templates, with no
development engine tree or Git metadata. The separate SDK input has **91
unpatched headers** at `ffa8b0a014b5cb4020ec8a99fe22fa3a24bfc494`. Its optional
CMake/LICENSE declarations are verified against their recorded upstream
blobs. This is supplied-header identity, not complete upstream or public
release admission.

HP status/process/memory checks preceded an outer foreground high-priority
`isol` host lease. Concurrent shared work was permitted; both peers inherit
the lease's isolated CPU mask. Clang 21.1.8/C++23/O3 builds the actual generic
controller's driver and cold fixture. There is no quiet or disjoint-role
timing grade.

| Check | Result |
|---|---|
| Actual driver self-test | Pass |
| Cold codec/value cycles | 48 pass |
| Exact cold D/V/W/8 bodies | 120 pass |
| Malformed required-value/OrdType cases | 57 rejected |
| Original generic controller comparisons | All four pass their wire audits |
| Actual closed-loop controller pairs | Four pass, 12 measured cycles each |
| Paced two-run pairs | Four pass, 3 warmups and 12 cycles/run at 2,000/s |
| Original primary samples | 144 retained |
| Original scheduler-delay samples | 96 retained |
| Actual application frames per peer | 470, bilateral bytes identical and exact bodies |
| Owner sink/fold reference, prepared tick clock, admin windows and Logon heartbeat | Pass |
| Independent frozen-source wire regrade | All 120 cold and 470 production frames pass |
| Focused source/admission checks | 79 pass, zero skips |

The generic controller now returns **0**. NOS and mixed flow's ACK Price
textually echoes the original D Price, satisfying the unchanged wire auditor.
All four rows remain `contract_status=not-audited`,
`controller_revision=source-archive-unqualified` and `stage5_status=S5 diag`.
Source files and all supplied SDK bytes remain unchanged after execution.
Public release identity never borrows the enclosing empty Git repository.

| Compiled product | SHA-256 |
|---|---|
| Driver | `68cc831af7f0a9790ffa2783e714c8c9b7591db83b306a6b8a538c0fe71398cc` |
| Cold fixture | `cfc0b340da8f83c48dd3ba9ca3f69c2c35549bfdde096db321df5bcc392ead39` |

## Failures and limits

The original setup rejected directory entries in the SDK tar. Its shell then
continued to a separate lease whose SDK verification correctly failed with
no compile or pair checks. That complete original setup/controller outcome is
retained. A separate `setup2` packet contains identical source/SDK/controller
bytes, accepts directories, and stops on extraction errors before heavy work.
It performs fresh successful builds and all eight pairs; no earlier outcome
is overwritten.

[Earlier stock-D spelling failures](../../../../docs/bench/RECORD_2026-10-03_nexusfix_body_alignment.md)
also remain failures under their original pins. The new small localhost
diagnostic does not grant runtime fixture, physical link, report
rate/delay/stability or public driver release admission. Root payloads and
older coverage cells keep their original grades. New adapter bytes need their
own next-generation driver licence review.

## Retention

- [Actual compiled controller and cold/paced result](evidence/nexusfix-orders-20261004/passing-setup2/report.json)
- [Independent frozen-source wire regrade](evidence/nexusfix-orders-20261004/independent-regrade.json)
- [Original failed SDK setup](evidence/nexusfix-orders-20261004/failed-setup/report.json)
- [Six unchanged synthetic negative-control outputs](evidence/nexusfix-orders-20261004/legacy-vector-preservation.json)
- [Byte-preserving import seal](evidence/nexusfix-orders-20261004/sha256-import.json)
- [Both-host archive verification](evidence/nexusfix-orders-20261004/replication-receipt.json)

The private archive retains **94 completed files** plus its named-input seal
at `~/archive/libhft/nexusfix-orders-20261004` on BEELINK2, HP and SCAN.
The seal SHA-256 is
`3d73a0f01ccd167e899016d1503403e85dc5c270b3f3b4a6cb8f2418a0fbe14c`;
both hosts verify every named file's bytes and mode. The **93-file devel
import** excludes only the source/SDK tarballs and two compiled products,
which remain in all three archives. Their identities and original complete
reports are retained. Original failures, successful samples and captures are
copied without normalization or grade changes.