# Two-rig Rust canonical comparisons

**Publication superseded 2026-10-17:** the original admission below is
withdrawn because its clock spans omit required work. Original receipts and
metrics are unchanged; see the [clock-boundary erratum](RECORD_2026-10-17_rust_clock_boundaries.md).

**Rust open versus NanoFIX passes all four flows on both accepted DAC rigs.**
This meets this campaign's two-rig minimum comparison gate. Both use complete
`canonical_body_v2`, the same full warmup/sample/run/rate profile, identical
deployment configuration and the unchanged pinned production runtime inputs.
The accepted topologies are HP's Solarflare and SCAN's mlx5 same-host physical
DAC namespaces; neither is called cross-machine.

The separate TrueFix comparison remains **blocked**: HP mixed flow misses rate;
SCAN MD misses rate, and mixed misses rate and scheduler-delay gates. Failed
flows have no ratios. Original failures and historical campaign cells stay intact.

## Actual independent admission

Each host reader replays the Rust baseline and each peer's twelve full original
pairs before matching. Complete original bodies, typed owner/counts, original
administration, primary/delay series, exact raw quantiles and extrema, physical
counter identity/polling and report decisions agree. Every arm retains 120,000
primary and delay samples and 330,000 application frames per peer. This reader
adds no SDK build, socket or physical run.

SCAN writes its final controller marker. HP writes both comparison receipts
with unchanged-source flags, but its transport ends without a final marker.
The remote worker has exited; both receipts are independently inspected and the
complete consumer source inventory has no differences. Only this session's
stale local SSH client is closed. The missing original marker is preserved.

The combination binds every original report manifest to its immutable physical
seal and checks its source/completion/deployment identities against each actual
cold receipt. It requires both distinct accepted rigs and preserves every
per-rig gate decision. Percentiles use medians of the three run percentiles;
**worst uses the maximum of all run extrema**. Original per-rig receipts, whose
earlier aggregate field used a median of run extrema, remain unchanged. The
corrected aggregate is separate and passes 25 source/shape/refusal/extremum
controls with zero skips.

Full-plan completion still requires the other language/anchor gates, TrueFix's
failed flows, remaining comparators and whole-harness release admission. This
new Rust comparison does not promote old campaign result cells or infer a
successful deployment of the website.

[Original receipts, frozen consumer and corrected combination](evidence/rust-matched-canonical-20261014/retention-seal.json)
are retained separately with an immutable byte/mode inventory.