# NexusFIX V/W/8 body alignment through the public SDK

The full-session adapter now builds V/W/8 with NexusFIX's public
`MessageAssembler`, while the SDK session supplies sender/target identities,
sequence numbers, timestamps, envelopes and checksums. V uses the common group
order without stock `AggregatedBook(266)`. W retains bid/ask order and the
declared ten-level or one-level book. ACK 8 uses the common field order,
identifiers and values, including required TransactTime. Integer-scale prices
are written with exactly five decimal places, including zero AvgPx.
The stock D builder and shared canonical workload are unchanged.

The independent cold checker can require exact V/W/8 alignment with
`--require-nonorder-alignment`. It retains the complete D gap and always leaves
`fixture_pass`, timing and Stage 5 admission false. Five negative controls
prove that equal required numeric values cannot hide extra tags, different
group/body order or alternate price spelling.

## Fresh complete-archive proof

Qualified source **`454d61190e6d7588dc6a1cc6bb015f631783a207`** is preserved on
`s/codex-fullbench-nexusfix-body3-source-20261003`. Its complete materialized
draft archive contains **732 files**, including templates. It has no `.git`,
`bench/`, `engine/` or `internal/`; an unrelated enclosing Git repository
does not supply a release identity. The separately supplied NexusFIX SDK has
**91 unpatched headers**, all pinned to `ffa8b0a014b5cb4020ec8a99fe22fa3a24bfc494`.

HP status/process/memory checks preceded an outer high-priority `isol` lease
on CPUs 8–17. Concurrent shared work was permitted: this is functional proof,
with no timing grade. Clang 21.1.8/C++23/O3 builds the actual archive controller
driver and cold fixture. The driver self-test, **48 cold codec cycles** and
**51 malformed-value rejections** pass, with zero skips. All **84 cold V/W/8
bodies** align exactly. D retains extra tags 21/40/59 and alternate tag 44
spelling; its required values and pinned field order still match.

Four actual controller pairs use five warmups and twelve measured cycles.
Four additional paced pairs use three warmups, two measured runs of twelve
cycles and a 2,000/s offered rate. All eight actual owners match the independent
common sink and fold reference. The original **144 primary samples**, **96
scheduler-delay samples** and **470 matching application frames per peer** are
retained. All **329 production V/W/8 bodies** align exactly. FIX envelope,
checksum, sender identity and increasing sender sequence checks pass.

Both peers retain complete admin windows and original Logon heartbeat bytes
showing thirty seconds. All controller observation audits and paced admin
audits pass. Primary/delay maxima and paced per-run delay quantile means match
independent raw recomputation. The prepared server tick boundary and its
measured count remain correct. Bound SDK/binary receipt reuse passes. All 732
source files and all 91 SDK headers remain unchanged after execution.

Driver SHA256:
`07d22ccf0aa17071e06a32cc8f19d37ac7eb105248dbd986e9d6a9c3a1b91748`.
Cold fixture SHA256:
`083272af56ef81799ac8fe719735831f61a7c594de2a7d53d1bf23539bf5825e`.

## Comparison failures remain failures

The actual generic controller exits **1**. NOS and mixed comparison rows fail
the existing wire auditor because ACK 44's five-decimal spelling does not
textually echo stock D 44. The independent auditor's sole deviation in each
flow is retained: `35=8: tag 44 must echo 35=D`. MD and tick controller rows
pass their diagnostic execution. Every row remains `contract_status=not-audited`,
`controller_revision=source-archive-unqualified` and `stage5_status=S5 diag`.
Passing the scoped V/W/8 proof does not change these failed comparisons.

The first source generation `d4bab2dcffea` failed compilation: the new header
helper expected a string view but the public SDK message constant is a char.
Its original compiler/controller/outer failures remain unchanged. Corrected
source `71e3ff3c6fbe` built and completed all four functional controller pairs,
with passing actual owners and admin observations, but the outer checker
expected controller success. Its NOS/mixed wire audit correctly rejected
both tag 44 spelling and a missing ACK TransactTime introduced in this pass.
Those original failures remain sealed. The qualified source restores
TransactTime and performs fresh builds/pairs; its checker expects and verifies
the remaining tag 44 comparison refusal. Neither earlier failed receipt is
rewritten or reclassified.

## Retention and export preparation

The 155 original files, including compiled products, are byte verified on HP
and BEELINK2 under
`~/archive/libhft/nexusfix-body-hp-20261003-454d61190e6d`. Repository evidence
omits only three compiled ELF products and retains the original HP seal and
all omitted hashes. The final and both failed source refs are pushed privately.

Local focused checks pass **110 checks**, with zero skips; comprehensiveness
passes. The original development test-helper failures and passing correction
are retained separately. Historical campaign commentary was moved from five
exported files into `internal/bench/`, preserving all original wording and
verifying unchanged executable tokens in the four source files. The public
source-layout guide now links the microbench owner's portable SDK/artifact
input contract. That inventory selects 729 files, with **134 scrub findings
in 32 files**, 83 textual source-root escapes and zero dead documentation links.
These remain migration counts, without release qualification.

[Sealed source, controller, cold fixture, raw, production wire and admin
evidence](evidence/nexusfix-body-alignment-20261003/SHA256SUMS).

Publication integrates the microbench owner's portable roster and discovery
command without changing qualified full-session inputs. All controller,
driver, fixture auditor, SDK/admin, canonical workload and shared build/clock
inputs remain byte identical. Of the 732 qualified archive files, 721 remain
identical; eleven differences are the verified comment relocation, guides and
Stage 2 roster entrypoints. The integrated inventory selects 732 source files,
with the same 134 scrub findings in 32 files, 83 textual escapes and no dead
links. Integrated 110 checks and comprehensiveness pass, with zero skips. The
[input map](evidence/nexusfix-body-alignment-20261003/integrated-inputs.json)
retains every original and changed hash; it does not extend runtime qualification
to the unrelated changes.

The common D/OrdType dictionary choice, full fixture admission, fresh physical
DAC runs, report rate/delay/full stability and standalone release remain open.
Installed payload generations, queued root jobs, Stage 2 adapters and shared
build/clock APIs remain unchanged.