# Current parity source rule qualification — 2026-10-04

## Scope and source

The qualified complete **750-file** harness is
`8c2897273c52dafda8cae6b1b02c58188886552f`, following the
[portable auditor input qualification](RECORD_2026-10-04_parity-audit-inputs.md).
Three changes correct obsolete source facts without changing a workload:

- Capability enumeration cross-checks the original shell reader against the
  strict owning Python catalogue reader. The active catalogue has 26 pairs,
  104 cells and 84 supported cells; restoring the five parked Rust-native
  pairs remains independently auditable. A parked pair cannot create an
  implemented row. The VMA no-shard check applies to declared pairs.
- The C++ pool header declares `class AcceptorThreadPool`, within its namespace;
  the old pattern required a qualified class spelling absent from its definition.
  Acceptor labels/lookups now use the actual public CoreAcceptor, SocketAcceptor
  and .NET MeasuredSessionRunnerAcceptor names instead of printing empty rows.
- Protocol P9 checks current standards **v4**, its 30-second measurement and
  five-second warmup, 50,000/s aggregate offer and OPEN mode. C++ T1 reads the
  actual `fix::CoreSession` reactor signature and retains the initiator-feed
  requirement. The old standards-v3 and former FixSession literals failed on
  current sources despite the requested implementations being present.

No SDK, benchmark workload, canonical wire/fold, timing boundary, catalogue,
standards file, Stage 2 source or installed payload is changed. Both original
protocol failures and all four original capability gaps remain in the raw control
outputs. Static source checks do not establish executed work or measured traffic.

## Actual whole-archive HP checks

HP preflight checked slots, processes and memory. The outer foreground suite ran
under a **shared/high** host CPU lease. It supplied the previously frozen open
SDK archives at `28f276e2ed6c58b7e99478ebf34b563ce09a9dfc`:
**926 C++**, **844 Java** and **670 .NET** files in three layouts each.
The exact prior eleven-file .NET runtime and its original build/test receipt
remain inputs; native library SHA-256 is
`1d6aea36dc8b1ae02a7f18893e026c2f2b87a7e652b42373bd9643e597039252`.
These sources are reused explicitly; no current SDK rebuild or native load runs.

All **8 original public CLI tests pass with zero skips**. **27 actual recorded
commands** drive original/corrected development/standalone audits and controls:

- Corrected development and standalone protocol stdout, stderr and exit are
  byte-identical: **150 pass, 0 fail, 0 deviation, 0 unknown**, exit 0.
  After removing ANSI colour, the original **148 pass / 2 fail** output differs
  only in the two corrected P9/T1 verdicts and the resulting total. The original
  stdout, stderr and exit remain separate artifacts.
- The corrected capability auditor has **0 gaps**. The complete original output
  differs only in the reviewed active-catalogue/class-name corrections, their
  derived gap total, current labels and declared document path.
- Independent negative mutations of the **actual** standards file catch old
  version, shortened measurement, shortened warmup, changed offered rate and
  closed-loop default. Each retains one P9 failure and exits 1.
- Removing the current reactor or the initiator feed call in the actual C++
  workload retains one T1 failure and exits 1. This suite never compiles or
  executes either intentionally broken source.
- An explicit SDK missing its pool implementation still reports the missing
  C++ pool. Restoring all five old Rust-native catalogue rows keeps 0 gaps;
  falsely permitting VMA sharding for that restored flavour reports 1 gap.
- A deliberately broken shell enumerator reports both its zero total versus
  the strict reader's 104 and its zero supported count versus 84. Dynamic
  catalogue counts therefore do not make the cross-check self-fulfilling.
- An explicit nonlibrary and a declared smoke row each retain the original
  symbol/publication failure. Missing explicit inputs/support refuse before
  audit output. Omitting optional archive evidence reports **147 pass, 0 fail,
  3 unknown**; a zero exit with unknown evidence is not report admission.

Both corrected complete harness layouts and all nine SDK layouts keep their
input bytes and modes. Every mutation is restored before full source verification.
All runtime bytes/modes and private declared input bytes remain unchanged; the
standalone sources have no generated outputs and the unrelated parent campaign
sentinel is unchanged. The disposable qualification root is removed.

## Evidence and admission

Original and corrected outputs, controls, full source maps, scripts and receipt
are retained in [`evidence/parity-rules-20261004`](evidence/parity-rules-20261004).
Whole harness/SDK/runtime source products stay outside website downloads in
`~/archive/libhft/parity-rules-20261004`. The previous preparation failures and
exact original audit bodies remain in the separately retained input archive.

This supplies a passing current **source/grammar/symbol audit** at the stated
pins. It supplies no benchmark timing, current SDK/ABI compatibility, native
load, physical-link proof, fixture/rate/stability admission or public release.
Existing campaigns, raw bytes and grades retain their own source generations.