# QuickFIX/J public parsed-group consumption

2026-10-12. This maintained driver change advances the
[full-bench plan](../../../docs/bench/PLAN_FULL_BENCH.md) without changing
the pinned stock SDK or any earlier runtime/result generation.

## Public getter

The driver previously used `getGroup(index, destination)`, copying the parsed
field map into a separate Group before reading it. The stock public
`getGroup(index, countTag)` returns the SDK's parsed group itself. The retained
stock FieldMap bytecode verifies the overload. The request's two entry-type
groups, symbol group and snapshot's twenty entries now use that public getter.
Extraction reads the returned groups synchronously and never mutates them.
All body checks and typed owner folds remain. This removes an unnecessary
consumer copy; it does not bypass production parsing or group decoding.

The complete 816-file source inventory is
`uncommitted-quickfixj-parsed-groups-331fee036f5e98540751e035791e2d3643e47e6d9e27afada87ebffbf2b8e4b3`.
Both hosts freshly compile the five driver sources against the same six
SHA-bound stock SDK JARs and pass twelve actual pairs with zero/repeated warmup.
Each cohort retains 240 primary/delay samples and 700 original application
frames per peer. Original complete warmup/measured bodies, bilateral Logon and
Logout, owner folds and public prepared-send/receive clocks pass independently.

Both-host cold replay reproduces every original receipt and rejects 75 altered
proofs each. All 25 migration constructor/readiness self-tests pass. The public
source scrub reports no finding.

## Full HP matrix

A high-priority exclusive lease completes twelve full localhost pairs, with
1,000 excluded warmups, 10,000 measured cycles and a 10,000/s offer. Every full
fixture and owner passes. The cohort retains 120,000 primary/delay samples and
330,000 original application frames per peer.

| Flow | Achieved cycles/s across three runs | Failed timing gates |
|---|---:|---|
| NOS/ER | 3,811–3,902 | rate, delay |
| MD | 3,108–3,214 | rate, delay |
| Mixed | 2,028–2,074 | rate, delay |
| Tick | 3,961–4,147 | rate, delay |

There is no full stability failure. The independent default reader reproduces
all twelve receipts, every raw quantile/extremum and the failed decisions
exactly, with original bytes/modes and reader source unchanged. It opens no
sockets and compiles nothing. These results do not establish a general latency
improvement over separately run configurations, and still fail the report-rate
contract. No physical or historical grade is promoted.

## Immutable retention

[Import](evidence/quickfixj-parsed-groups-20261012/import-receipt.json)
and [seal](evidence/quickfixj-parsed-groups-20261012/retention-seal.json)
retain **536 files / 177,991,522 bytes**, seal SHA-256
`cb6b09c2b846096e110477dae7fb86f406288296a6e5aa9b77ab41b85479ebe4`.
Both host replicas verify all originals. The supplement contains 36 actual
pairs, 120,480 primary/delay samples, 331,400 original frames per peer, exact
source/runtime inventories, the qualified source, stock bytecode, original logs
and all independent replay receipts. Earlier copied-group runtimes remain intact.