Document source: 51e93508f22f; code reference: 51e93508f22f. 中文. Source language: en; requested: en; actual: en; translation: reviewed.

Source documentation does not establish software or physical hardware qualification. See the support and qualification records for their exact scope.

SoC refinement execution record

This is the historical 3728b334 baseline. Later functionality and execution status are recorded separately in SoC extensions. The hashes and counts below do not qualify those later changes.

This record covers the 55-item SoC ledger. It distinguishes implemented mechanisms from the exact source and execution scope of each result. Subsequent source changes do not inherit earlier results. All physical tests remain not_executed.

Committed software gates

The following commits passed ordinary staged hooks, including the unchanged-index TDD gate and commit-message checks. No test filter, skipped-only report or hook bypass establishes acceptance. Counts describe their individual execution scopes; they must not be added together as a final-platform test total.

Commit	Change	Recorded software execution
e7deeecdb2dfc85d40f2f557cd655cfa34e1492d	Exact-package GPIO inventory and bounded G474 UART	Normal commit gates; source-specific RED/GREEN retained
7e48b879b327df19c6d7dd7f672a5898585bdb2c	Nineteen SoC HIL case recipes and captured-evidence validation	Normal tool gates; plans remain prepared and unbound
1f5ff587f00ce91882ffd455982f50191bc7907f	Startup, clock, memory and interrupt cleanup ownership	156 GoogleTest executables / 2,819 cases; 480 tool and 385 configuration tests
ae27e999f157fecd027db99bdfa99ce38582a940	Reviewed original H7/U5 CMSIS dependency closures	156 GoogleTest executables / 2,819 cases; 527 tool and 385 configuration tests
ee33f38d8120405b5bf6f383710f81d05b12371b	Immutable descriptors, cold admission, EXTI and timer start	157 GoogleTest executables / 2,847 cases; 576 tool and 397 configuration tests
5899c230c8063eb8832bb1dc7f171958d83227d5	Runtime timer ACK isolation, retained cleanup and narrow W1C models	157 GoogleTest executables / 2,862 cases; 576 tool and 397 configuration tests
c3c4a62ebdbd3661bb561392651e80c23dd5b1e3	H743/U575 static GPIO/UART references, SDK closure and HIL guards	163 GoogleTest executables / 2,982 cases; 585 tool and 415 configuration tests
23cde7e57a4781db8ffe2e75849c02f976c4304d	Selected GPIO leases and terminal time-domain failures across five MCU families	165 GoogleTest executables / 3,040 cases; 585 tool and 423 configuration tests
3728b334b02eb2eb70bcf5d4caafbb9746837dd3	Complete private pin-helper dependencies and aligned startup DATA loads	165 GoogleTest executables / 3,040 cases; 585 tool and 428 configuration tests

All numerical suites above completed without failures or skips. Their staged source-tree identities and raw hook outputs are retained in the external nexus-soc-20261012/evidence/batch-*/ and nexus-soc-qualification-20261012/batch-*-hook-passed/ records. An earlier failed hook remains a failed attempt; its output is not rewritten as a successful run.

The dependency-only commit ae27e999f157fecd027db99bdfa99ce38582a940 additionally passed 47 actual source-SDK boundary tests with zero failures or skips. Source and dependency state were clean before and after execution. Relocated public consumers exercised Native C/C++, the existing platform ARM profiles and ThreadX within their declared scopes. This run did not qualify H743/U575 firmware. Its raw command receipts and consumer artifacts are recorded in nexus-soc-qualification-20261012/sdk-dependency-boundaries-v2/.

Baseline source qualification

The implementation qualification source is 3728b334b02eb2eb70bcf5d4caafbb9746837dd3, tree 48ba820e5c2d4c372bf7eb6e909c49850cc1fe76. Subsequent documentation commits describe that exact code source; they do not replace its qualification identity. The portable result index records exact report hashes and archive-relative paths without copying raw logs into Git.

Normal hooks passed on that source. The same complete model execution binds 203 reviewed I/O properties to assertion anchors and current production inputs; this is bounded executed evidence, not a formal or exhaustive proof. All five maintained ARM gates passed, covering sixteen memory/reference images and 87 selected-provider images. The same code source passed all 58 installed SDK boundary tests. Its 236 actual commands include expected nonzero negative-path results; the test suite has zero failures, errors or skips. Earlier results, including 58 SDK tests and 22 legacy contract programs on 23cde7e, are retained as source-specific history and are not promoted to the new source.

Scope	Maintained execution entry	Final-source status
Production C, generated factory and failure paths	scripts/ci/tdd_gate.py --all	165 executables / 3,040 cases; zero failures or skips on the code source above
Configuration, provider descriptors and tools	tools/testing/run_tool_tests.py	585 tool / 428 configuration cases; zero failures or skips on the code source above
F407 default/CCM and GD main/additional/TCM RAM	scripts/ci/stm32_memory.py, scripts/ci/gd32_memory.py	Eight actual ARM images passed resources and selected retention
H743 empty/selected, baremetal/FreeRTOS	scripts/ci/stm32h743.py	Four actual ARM images passed
U575 empty/selected, baremetal/FreeRTOS	scripts/ci/stm32u575.py	Four actual ARM images passed
Every maintained authored controller assembly	scripts/ci/provider_assemblies.py	All 87 actual fixtures passed; clean source and unchanged inputs
Installed SDK and actual public consumers	tests/contracts/test_source_sdk.py, test_device_sdk_payload.py	58 actual tests passed; clean source before/after; relocated Native, selected MCU and ThreadX consumers
Hot paths and image/resource costs	tools/measurement/soc_costs.py	Eleven representative final profiles observed: eight H743/U575 plus F407 EXTI, GD EXTI and G474 LPUART
Derived capability projection	tools/maintenance/capabilities.py --check	87 rows checked; actual exit code zero
External examples on the same platform code	X-Gen-Lab/nexus-examples	Eight CTest programs, three GoogleTest programs / 43 cases, 20 ARM images and budgets; ten HIL plans remain prepared and unbound
Offline reproducibility and candidate sealing	tools/evidence/reproduce.py, candidate.py, proof.py	Not established by this iteration record
Physical SoC cases 37–55	tools/hil/soc_acceptance.py	Tooling implemented; physical execution absent

The H743 ELF gate inspects the actual reset-to-power-entry closure before RAM is usable, in addition to vector bytes, selected retention and resources. The U575 gate preserves the declared TZEN=0 single-state boot boundary. Its vector/count checks do not automatically prove the complete Reset/SystemInit/data/BSS/MSP call closure. Independent review has now checked all eight actual H743/U575 images, including the complete linked startup path and four-byte DATA load alignment; 2,297 identified inputs stayed unchanged before and after capture. The gate’s source_and_call_closure_review_required boundary is closed by that separate source-bound report, not by relabelling the vector test. Neither ARM gate executes a scheduler on the MCU or establishes a measured electrical result.

The actual selected-reference costs are below. Flash is the sum of LOAD file bytes, not the largest occupied address span. RAM includes reserved MSP exactly once; every row reserves 2,048 MSP bytes and zero heap. Static reservations are not measured stack high water or product application budgets.

Family	Backend	Assembly	Flash LOAD bytes	RAM reserved bytes
H743	baremetal	empty	2,437	2,104
H743	baremetal	GPIO + UART	8,908	2,364
H743	FreeRTOS	empty	7,564	3,716
H743	FreeRTOS	GPIO + UART	14,108	3,976
U575	baremetal	empty	2,225	2,112
U575	baremetal	GPIO + UART	8,712	2,372
U575	FreeRTOS	empty	7,300	3,724
U575	FreeRTOS	GPIO + UART	13,868	3,984

The selected H743 and U575 images expose 60 and 68 bytes respectively of GPIO lease/gate objects. These are not total Platform storage. An inlined or unretained helper without a separate function symbol is not assigned zero cost. These observations describe exact linked selections; different backends, capacities or routes are not comparable optimization percentages. No cycles, interrupt blackout time or WCET was measured.

The same observer also inspected three existing images from the completed 87-fixture matrix without recompiling them:

Assembly	Selected storage	Flash LOAD bytes	RAM reserved bytes
F407 VE EXTI	Two lines, four event slots each	6,152	2,344
GD32 EXTI	One line, eight event slots	6,956	2,260
G474 GPIO/LPUART	64 RX bytes	9,012	2,316

These RAM figures also include the 2,048-byte MSP reservation and zero heap. The supplementary report records exact symbol sizes, decoded instructions and static branch sites. Different routes and capacities make these separate baselines. The eleven measured profiles do not represent every maintained mode or establish cycle counts, blackout duration or a cross-chip optimization rate.

External examples are qualified at code commit f02b91867bb64abfa31e3c1d4b6004bd1a38b091, pinned to the platform code above. The actual 46-command clean execution and its source record are indexed separately. The ten HIL preparations do not mean ten physical station executions.

Interpreting the ledger

implemented records a mechanism in source. contract_implemented and documentation_updated record narrower contract/documentation work. qualification_record_authority delegates acceptance to the actual external record for the exact source/configuration/artifact tuple; it is not a source file’s self-reported pass. final_source_pending means that execution is still required for the completed source scope.

selected_profiles_observed records actual static cost measurements only for the profiles named above and in the index. tooling_software_validated qualifies the HIL software contracts; the corresponding physical test remains unexecuted. The partial-reference implementation states preserve their limited GPIO/UART scope even after host, ARM or installed-SDK checks pass.

SOC-05 records versioned datasheet, manual and errata provenance. Actual silicon revision and full per-mode errata applicability remain unqualified. SOC-23 and SOC-27 require observed actual ELF costs; an observer’s unit tests alone do not complete measurement. A changed capacity, backend, Board route or memory layout must be reported as a distinct selection rather than a like-for-like saving.

SOC-28, SOC-30 and SOC-31 contain only the exact H743/U575 GPIO/UART/time reference subset. SOC-29 and SOC-32–36 retain explicit missing features. Their original exit conditions remain unchanged. A rejected unsupported mode is not a completed mode implementation. SOC-37–55 contain executable software preparation, admission, capture and validation tools; every physical exit remains open until real station records are captured and validated.

No result here establishes MCU cycles, interrupt latency, Flash power-loss behavior, cache/DMA coherence, TrustZone isolation, product safety, certification or a product release. Exact support remains defined by the support matrix and SoC contracts.