# Canonical .NET single-traversal book physical outcomes

The [actual compiled qualification](RECORD_2026-10-17_dotnet_book_fold.md)
completes all **128 physical tiny/report pairs**: Native/Managed, inherited/
first CPU, on each accepted DAC rig. Both **Native and actual public Managed
first-CPU configurations pass all four Stage 5 flows on both rigs**, within
one declared source/generation per configuration and host. No collection
borrows a successful flow from another profile or source.

All four HP configurations pass all flows. SCAN Native inherited tick retains
p50 stability failure; Managed inherited mixed retains one rate failure. All
other timing/functional/physical gates pass. Thirty successful producer-written
compact cells are imported byte-exactly: twenty-nine latest replacements and
one newly admitted Managed first-CPU SCAN mixed key, giving 643 compact cells.
The two failed fresh cells retain older successful bytes. No current .NET peer
comparison, ratio or whole-harness release admission is inferred.

| Rig | Configuration | Flow | Full grade | Achieved cycles/s, min–max | Blockers |
|---|---|---|---|---:|---|
| hp | native-inherit | nos_er | S5 pass | 9987.795–9987.917 | none |
| hp | native-inherit | md | S5 pass | 9988.013–9988.231 | none |
| hp | native-inherit | md_nos_er | S5 pass | 9987.209–9987.671 | none |
| hp | native-inherit | tick_to_trade | S5 pass | 9987.453–9987.622 | none |
| hp | native-first | nos_er | S5 pass | 9987.698–9988.002 | none |
| hp | native-first | md | S5 pass | 9987.900–9987.986 | none |
| hp | native-first | md_nos_er | S5 pass | 9987.647–9987.674 | none |
| hp | native-first | tick_to_trade | S5 pass | 9987.428–9987.497 | none |
| hp | managed-inherit | nos_er | S5 pass | 9983.532–9983.918 | none |
| hp | managed-inherit | md | S5 pass | 9914.055–9917.229 | none |
| hp | managed-inherit | md_nos_er | S5 pass | 9916.183–9916.954 | none |
| hp | managed-inherit | tick_to_trade | S5 pass | 9980.664–9981.430 | none |
| hp | managed-first | nos_er | S5 pass | 9983.543–9984.169 | none |
| hp | managed-first | md | S5 pass | 9916.831–9917.506 | none |
| hp | managed-first | md_nos_er | S5 pass | 9915.956–9916.895 | none |
| hp | managed-first | tick_to_trade | S5 pass | 9981.173–9981.447 | none |
| scan | native-inherit | nos_er | S5 pass | 9981.642–9982.274 | none |
| scan | native-inherit | md | S5 pass | 9982.129–9982.429 | none |
| scan | native-inherit | md_nos_er | S5 pass | 9979.925–9981.068 | none |
| scan | native-inherit | tick_to_trade | S5 diag | 9981.457–9981.925 | full_percentile_stability |
| scan | native-first | nos_er | S5 pass | 9981.921–9982.452 | none |
| scan | native-first | md | S5 pass | 9981.632–9982.363 | none |
| scan | native-first | md_nos_er | S5 pass | 9980.657–9981.161 | none |
| scan | native-first | tick_to_trade | S5 pass | 9981.774–9982.962 | none |
| scan | managed-inherit | nos_er | S5 pass | 9976.479–9976.603 | none |
| scan | managed-inherit | md | S5 pass | 9975.855–9976.397 | none |
| scan | managed-inherit | md_nos_er | S5 diag | 9791.182–9912.660 | achieved_rate_below_98pct |
| scan | managed-inherit | tick_to_trade | S5 pass | 9974.598–9975.251 | none |
| scan | managed-first | nos_er | S5 pass | 9975.617–9976.474 | none |
| scan | managed-first | md | S5 pass | 9975.757–9976.096 | none |
| scan | managed-first | md_nos_er | S5 pass | 9857.178–9907.551 | none |
| scan | managed-first | tick_to_trade | S5 pass | 9975.508–9976.051 | none |

SCAN Native inherited tick's p50 sample CV is 9.095067%, above 5%; p99 and
p99.9 CVs are 6.207130% and 6.601388%, below their unchanged 10%/25% bounds.
Managed inherited mixed rates are 9,854.996, 9,912.660 and 9,791.182/s: the
third is below 9,800/s. Its first-CPU configuration independently holds
9,857.178, 9,907.551 and 9,896.874/s and passes all stability gates. These are
separate declared profiles, not an unchanged failed timing retry. Previous
failed gates remain retained. No causal allocation, traversal or CPU claim is
inferred from these separately collected observations.

## Exact source, original production session and compiled proof

- Physical source: `1847a22cea224e2d9ff2b1189e0cb960ad0917da`.
- Complete stripped source: 911 files, inventory SHA-256
  `1ab98d68c51b3ef67e3d6a04580fe7b6bf8c7a7dabc9fd10d9be5304b18b2d50`.
- Native DLL: `57c8c4d6ae99a64fa2cddc82da1f9ac35ddbbb47dba6a1b1587d562024d465fd`.
- Managed DLL: `0c55a7611bb9f77fdda9710aa677b113c08cde364dc52f14ce0ec68c84ad735a`.
- Exact current SDK DLL:
  `0a6731d3bd690ca48eafb11d00540d7474f5f8376a47739bd66ea361d5d2ff6f`.
- Unpatched SDK content pin:
  `uncommitted-functional-source-26310f12584fd0f392af231da8c2ea10067f50bb337db118cc79a2895da54386`.
- Unchanged native library:
  `91bdc88f8cd2d375262daafdc233ce2b16bc49e8fcae0789da12af999c737dc9`.

| Rig | Configuration | Immutable installed generation |
|---|---|---|
| hp | native-inherit | `2a3396bd25ddcdf389e45978b369b28afbc3731d52d8f611a7e1333e8fb0d3bb` |
| hp | native-first | `62a3687837c710a599a80602cda0928e144bc39b0237a2de47ae8a51a28073f4` |
| hp | managed-inherit | `30d1faeb93683f750b6d02ed1f4167efba7b87c4949825c7c37f5237ebc9f479` |
| hp | managed-first | `957362d88f60041b7181fe0a2ced27a058f7d374ce39b3218445186855de4fed` |
| scan | native-inherit | `fbae1a7f11e76411db0638c8e88d120dc85c19ece1f5031da8cf7814efe4222a` |
| scan | native-first | `de189b5c984db35a9f4a80819ce6655e888d21f5cb56ffee40cff24d5d61cb0e` |
| scan | managed-inherit | `f67357ac2625d04d3d11eb232f249b09ba7a2f504e19d188672554d1dfaa45b5` |
| scan | managed-first | `b680237690538ae2d11cd7b345f71bfc40eaccae29a95d6e3152fb9b958a00bc` |

Each bundle verifies its seven actual compilation inputs and every runtime
asset against the original 216-pair/1,338-refusal qualification. It additionally
binds the exact compiled extraction checker and its 16,320 real malformed-book
rejections per host, including previous/current Native and Managed products
with each exact SDK. No duplicate extraction helper can supply that proof.
The SDK and target binaries are not rebuilt during preparation or collection.
Source/SDK bytes and modes and all runtime files remain unchanged. The earlier
cold-checker SDK-property/binary-binding failures stay retained separately.

Native uses actual public NativeEngine/NativeSession, SDK batches and PollMessage;
Managed uses actual ManagedSession/IFastHandler, builders and the public pre-send
gate. SDKs own Logon, sequencing, parsing, serialization, actual completed writes
and Logout. The managed native engine remains unloaded. Every level, typed
price/size, side, order/count, identity and complete owner fold is still checked.
The shared formatter's exact generated bytes remain unchanged. No SDK/native
patch, root helper, sudo rule, clock or grading threshold changes.

HP uses Solarflare, inherited masks 8–12/13–17; SCAN Mellanox, 10–13/14–17.
First-CPU foreground observations are 8/13 and 10/14 respectively. Inherited
profiles keep their full actual masks. Every endpoint/run matches its declared
profile and fixed outer budget. Host probes retain actual kernel/NIC/firmware/
namespace/CPU facts. Preparation uses `isol`; timing holds `isol-no-shared`,
high priority on HP, normal on SCAN, after checking slot/process and free disk.
Fresh disjoint ports are 64000–64015, 64100–64115, 64200–64215 and 64300–64315.
Only existing fixed helpers use root; applications execute as the benchmark user.

## Complete clocks, pre-warmup capture and independent cold replay

Every report has 1,000 warmups, 10,000 measured cycles per run, three runs and
target 10,000 cycles/s. Complete RTT ends after final required parse/extraction/
fold/checks; the tick uses prepared SDK wire before actual write through public
polling-handler return or callback entry. All selected clocks remain process-local
and monotonic. Probe is integer zero; no measured callback writes a capture/trace.

Each endpoint captures only ten pre-warmup cycles, then closes before warmup.
Actual bilateral body/framing/session, complete owner sink/fold, SDK callback,
prepared/completed write, admin, CPU, raw primary/delay and physical counter
checks pass on all rows, including both diagnostics. Every host/configuration's
tiny and full cold readers independently reproduce all original quantiles,
true extrema, fixtures and every promotion decision exactly. Original bytes/
modes remain unchanged; no cold replay compiles, opens sockets or repeats timing.

Full proof covers 120,000 primary/120,000 delay samples and 330,300 application
frames per peer per configuration/host: **960,000 primary/960,000 delay samples
and 2,642,400 application frames per peer** across all eight collections. Tiny
adds 80/80 samples and 350 frames per peer each. Each admitted compact result
has at most 2,000 CDF points per latency kind. Old successful cells and original
failed records remain intact, without filling a failed fresh collection.

Both hosts retain `~/.cache/libhft/fullbench-dotnet-book-fold-physical-20261017-v1`
and `~/build/codex-fullbench/libhft/dotnet-book-fold-{native-inherit,native-first,managed-inherit,managed-first}-physical-20261017-v1`.
Only qualified compact results and this summary enter Git/website data; raw
working capture/sample files, logs and source archives remain on the hosts.
Both actual LibHFT .NET flavours now have a complete first-CPU configuration
on both accepted rigs. A current canonical peer, the remaining language/anchor
work and coordinated whole-harness release remain separate plan gates.