# TrueFix runtime diagnostic

The [retained bundle](evidence/truefix-current-thread-20261002/SHA256SUMS)
contains the original tiny and medium manifests, process logs, full raw primary
and scheduler-delay series, canonical wire audits and matching compiled source.
All original artifact and source hashes were verified before retention.
**These are HP localhost diagnostics, not physical Stage 5 results.**

The Stage 4 builder now accepts `TRUEFIX_BENCH_RUNTIME=current_thread` as an
explicit build-time experiment. It replaces only the adapter's Tokio entry-point
attribute with the public current-thread runtime. The original adapter source,
compiled `main.rs` hash and runtime selection are recorded separately. The
pinned upstream source, canonical fields/groups, optional dictionary validation
setting, log hooks, pacing and sample boundaries remain the same. The default
remains two Tokio workers plus the foreground task. The immutable root payload
and its pending physical jobs retain that tested default configuration.

The high-priority HP job used an `isol` lease, allowing concurrent shared
conformance work. Tiny has five excluded warmups and 20 samples at 100/s;
all four flows pass session, raw, wire and rate checks. Medium uses 1,000 paced
excluded warmups and three fresh process pairs with 10,000 samples each at an
offered 10,000 cycles/s. Every flow retains 30,000 primary and 30,000 scheduler
samples, bilateral logon/logout and complete canonical application extraction.

| Flow | Achieved cycles/s, three runs | Mean run p99, us | Stability | Local gate |
|---|---|---:|---|---|
| `nos_er` | 9991.922, 9991.299, 9991.245 | 67.278 | PASS | pass |
| `md` | 9988.931, 9986.517, 9985.459 | 96.432 | PASS | pass |
| `md_nos_er` | 6729.310, 6660.320, 6671.546 | 165.068 | PASS | fail |
| `tick_to_trade` | 9991.620, 9990.777, 9991.174 | 60.285 | PASS | pass |

Current-thread market data and tick-to-trade pass their local rate/stability
gates in this diagnostic. Mixed flow remains limited to 6,660–6,729 cycles/s,
below the report target. This experiment alone does not resolve the plan's
TrueFix mixed-flow rate requirement and is not admitted as a physical replacement.
The workload keeps one cycle outstanding; it does not measure pipelined capacity.