# Artio complete original comparison fixture

2026-10-12. The new Artio source qualifies on **HP** from a complete stripped
source archive; its matching SCAN continuation also passes. These are
localhost functional checks; **physical, timing and release gates stay open**.

## New generation

The retained v9 frames have extra D fields `21`/`59`, an out-of-order `40`,
and extra V field `265`. Its transport observer retains only twelve cycles.
The new producer emits the shared minimal D/V bodies and streams every
original completed TCP application frame to buffered per-instance files.
Capture memory remains bounded. The producer waits for the complete expected
frame counts, flushes before writing observations and closes both streams on
success or failure after closing its SDK runtime. Write failures invalidate
the transport observation.

The new `canonical_body_v2` declaration supplies no admission alone. The
controller independently checks **every warmup and measured body** for exact
values/order/spelling with no extra fields; bilateral original byte identity,
framing/checksums/increasing sender sequences; thirty-second original Logons
with matching peer bytes; owner sink/folds and measured administration. Both
endpoint declarations, complete capture counters and original JSON files must
agree. The common extraction model remains `canonical_spec_v1`; the shared
fixture/model, clocks, SDK and installed v9 pins are unchanged.

Legacy observations retain the twelve-cycle diagnostic path and its original
result fields. They cannot borrow new admission. Local report blockers derive
fixture alignment only after every actual pair passes; topology, numeric
timing and host fairness remain separate gates. The grading snapshot now
retains all three new checker dependencies.

## Real qualification

| Input | Identity |
|---|---|
| First new source / archive | `d9fcf6e091cd57ff41cb5ce83508522758bdf3d0`, 791 files |
| Corrected private-test boundary | `80aa66ff6316956ae238678ff40ded39fea2b0ad`, 790 files |
| Fresh HP adapter JAR SHA-256 | `66ab83974eaf352027c1dbe6151863180bbec64d456d13480f86cc7a5f7cd337` |
| Unchanged SDK inputs | Entire 24-file preserved v9 runtime inventory, including all original library pins |

All **138 local checks pass without skips**, including the old original
observations and owner evidence. The first real stripped archive exposes
**twelve private-data test failures / 59 passes**, before any compilation or
socket. The historical owner/observation tests now live under `internal/`;
all are preserved and pass. The public suite's **59 portable checks** run
without private data rather than skipping missing campaign records.

HP freshly compiles all seven production sources and the test sources against
the unchanged external libraries. All **16 Java tests and 53 payload/semantic
checks pass, zero skips**. The archive has no engine/internal/Git source and
is extracted under an unrelated Git parent. The shared model resolves from
the archive; the complete source hashes/modes and SDK inventory are checked
before and after. Original compiler/version/argv, JUnit/Hamcrest hashes,
class hashes and deterministic JAR packaging are retained.

The actual source controller completes **sixteen real TCP pairs**: all four
flows with warmup zero, warmup five, and real measured TestRequest/Heartbeat
probes, twenty samples each; then every flow with warmup seventeen and twelve
samples. It retains **288 primary and 288 scheduler-delay samples**, plus
**990 original application frames per peer**. Every actual complete fixture
passes. A separate replay from the same full source archive reproduces all
sixteen original receipts exactly after normalizing the copied artifact
directory path; original bytes/modes stay unchanged. It creates
zero new sockets or compiles.

Heavy work uses cooperative leases after host/process/memory checks. HP waits
at **high priority in shared**. SCAN waits in isolated CPUs with concurrent
shared work allowed, then remains behind another owner's exclusive job.
Only the verified obsolete own SCAN wait PID `1321343` was cancelled: UID,
command, thread identity, absence of children/output and the busy other lease
were checked first, and a PID descriptor prevented targeting a reused PID.
No other owner's job or queue was stopped. Its replacement preserves the
matching corrected archive and remains queued, not qualified.

## Retention and remaining gates

The first HP archive is
`~/archive/libhft/artio-complete-fixture-hp-20261012`, replicated to both hosts
with exact SHA-256/size/mode checks. It seals **455 files / 7,052,783 bytes**,
seal SHA-256
`9118c441584c0c34f29c2b9d439826237630a70b79c3d0eaa0e8d7ed5d37af5c`.
The committed import retains 453 original files plus import and replication
receipts. Two source tarballs and the compiled JAR remain archive-only.
The original private-test failure and queue snapshots are immutable; SCAN's
later completion belongs in a separate supplement.

Matching immutable payload/physical controller admission,
fresh DAC rate/delay/stability and matched libhft evidence, whole-harness
standalone consumption and release remain open. No installed or historical
v9 row is relabelled. The eight-runtime home installer currently covers the
previous qualified generations; it does not claim this Artio generation.

## Matching SCAN continuation

SCAN completes the same 790-file archive, 59 portable checks, sixteen Java
tests and 53 semantic controls, without skips. Its sixteen actual TCP pairs
also retain 288 primary/delay samples and 990 original application frames
per peer. Both independently compiled class inventories and deterministic
JARs match byte for byte. Source bytes/modes, all SDK inputs and test libraries
remain unchanged. The joint read-only replay exactly reproduces all **32
receipts, 576 primary/delay samples and 1,980 frames per peer**, with no new
sockets or compiles. Both qualifications are functional and have no timing
grade.

SCAN originals, product and the joint replay form a separate supplement at
`~/archive/libhft/artio-complete-fixture-scan-followup-20261012`, replicated and
verified on both hosts. It seals **432 files / 1,795,498 bytes**, seal SHA-256
`94661f07a73b7b1dedfc757a807eff45c09f73929d653cdf4a7b3ce0bbfc451c`.
The import retains 432 originals plus import/replication receipts; the compiled
JAR remains archive-only. The first HP seal and queued snapshots stay
immutable. Physical/runtime/report, whole-consumer and release gates remain
open, and every installed/historical v9 pin and grade is unchanged.