# TrueFix corrected callback clocks: physical collection

## Frozen inputs and first infrastructure failure

The complete 897-file stripped harness is pinned to
`51d67814e6218dbab95ad3441012a12a7ab13bda`, inventory SHA-256
`6b2a5ba1c4aaa1fb0590be7348dadc63fb3f762be728195598a7623f8ea09b88`.
It exactly matches the [qualified consumer's](RECORD_2026-10-17_truefix_callback_preflight.md)
exported files, including both original SDK/compiler/dictionary bindings.
Both physical rigs install the two original qualified products through the
existing immutable normal-user generation helpers. No sudo change is needed.

SCAN's first four tiny pairs pass. Its first report reuses their four ports;
NOS runs 1–3 and MD run 1 fail to bind (`AddrInUse`), leaving their client
logs empty. The remaining report cases retain their actual outcomes, but
complete cold admission fails and no compact cell is written. The script
stops before the current-thread arm. This entire failed original stays under
`~/build/codex-fullbench/libhft/truefix-callback-physical-20261017-v1` on SCAN.

The pinned SDK's `SocketOptions::default` sets `reuse_address=false`; its
acceptor applies that option at bind time. Reusing tiny ports for the next
report collides with the prior connections' TIME_WAIT lifetime. The generic
matrix now allocates disjoint ranges, checked before output creation. Nine
focused port-boundary/lease controls pass without skips. The waiting old HP
TrueFix request is dequeued before execution; the three other waiting
campaigns retain their own immutable inputs and high priority.

## Fresh complete collection

Both rigs prepare fresh output roots while verifying and reusing the exact
installed generations. The new SCAN matrix uses default tiny 61600–61603,
report 61604–61615, and current-thread tiny 61800–61803, report 61804–61815.
No SDK socket option, binary, clock span, workload or grading threshold changes.
All original failures remain visible; a complete new matrix is collected
under the corrected port configuration, with no selection of better repeats.
Operational recovery controller SHA-256:
`3a38e825d0ab7d0a40c91da41742a52e4e62be29cb1e87732be07e4965231341`.

The quiet SCAN matrix completes 32 actual pairs: four tiny and twelve full
report pairs per runtime. Every exit, preflight/body/owner/clock observation,
normal-user/namespace/CPU admission and physical wire-counter gate passes.
Each report retains 120,000 primary and 120,000 scheduler-delay samples and
330,300 application frames per peer. Each independent cold reader reproduces
all raw quantiles, extrema and decisions exactly, with original bytes/modes
and both frozen source inventories unchanged. Capture is only before warmup;
measured application capture is zero. Actual Rust/Cargo versions and SDK/
dictionary pins are retained in each result's provenance.

| Runtime | Generation |
|---|---|
| Default multi-thread | `6384e7af6dee73770b50c2e1fcfcf668e73aa7239866cbc25ca376eace0dc6de` |
| Current-thread | `efe3892cff15ff6336f16d879b57f38521d1127ac6d1189af78472260e3255cb` |

## Actual gates

Every diagnostic row below has all 30,000 primary/delay samples and passes
functional/physical checks; the timing admission gates remain unchanged.

| Runtime | Flow | Achieved /s, min–max | Grade / blockers |
|---|---|---:|---|
| Default | NOS/ER | 6,709–6,889 | S5 diag: rate |
| Default | MD | 6,184–6,745 | S5 diag: rate, percentile stability, delay |
| Default | MD + NOS/ER | 3,066–3,090 | S5 diag: rate, delay |
| Default | Tick to trade | 6,828–6,961 | S5 diag: rate |
| Current-thread | NOS/ER | 9,985–9,988 | **S5 pass** |
| Current-thread | MD | 8,549–8,760 | S5 diag: rate |
| Current-thread | MD + NOS/ER | 4,902–5,074 | S5 diag: rate, delay |
| Current-thread | Tick to trade | 9,987–9,989 | **S5 pass** |

Two admitted current-thread compact cells are copied byte-exactly under
`truefix+callback+complete-rtt+current-thread`. NOS SHA-256:
`f88269bf4cbc79f9e9bf6e2eed9abd7354ac45260e5053dfb6e679f5625f8ee8`;
tick SHA-256:
`4f70919d1c905d428d93a5e92f09f003064371a3e6034af5ec2ab3310171c138`.
Failed cells replace no result. Original shortened-clock cells stay superseded.
The successful collection is not retried unchanged. No raw working output is
copied into Git, the archive or the website. Four-flow matched admission stays
blocked, and HP's fresh corrected-port matrix remains queued at high priority
behind shared conformance. New working roots on both hosts:
`~/build/codex-fullbench/libhft/truefix-callback-physical-20261017-v2`.

## HP follow-up completed

The queued high-priority HP matrix now completes under a quiet lease. Its
original source/runtime remains unchanged; actual outcomes, admitted compact
cells and retained failures are in the [HP follow-up](RECORD_2026-10-17_hp_canonical_followup.md).
This appendix supersedes the earlier pending-HP scheduling status, while
preserving the original SCAN outcomes and campaign-specific comparison scope.