Document source: e47887d3a11f; code reference: e47887d3a11f. 中文. Source language: zh_CN; 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.

Software verification, source identity and evidence

This page navigates existing execution records; it does not establish platform-software qualification for this documentation HEAD. Compare source, assembly, dependencies, tools, actual execution and the same ELF/BIN/map bytes together, rather than reading passed alone. The qualification contract maintains qualification rules.

Current P2 unsigned software delivery and independent QA followup

Software qualification for this round is fixed to public source ab53225033dfb759e93c90beb6301a3048bbcfda, tree 2309ae92e019fac68c6f463cbee205c3726fe73d, with 23 clean public dependencies. The ordinary complete-index gate passed 420 Google programs, 7342 cases and 3964 cases across four tool domains, with zero failures, errors or skips. Two independent complete 159-stage plans actually passed; each Native mode had 420 programs / 7342 cases / 440 CTest registrations, and counts across the four modes are not added. Both tidy and cppcheck retained 1193 compiler contexts / 1303 owned TUs and returned 0.

SDK v15 passed 64 roots / 67 images / 270 commands and independent verification. The eight H743/U575 × baremetal/FreeRTOS × empty/UART candidates completed 16 fresh network-disabled builds, with byte-identical ELF/BIN for each pair, successful sealing, actual restoration and restored verification. HIL for the four UART profiles is only prepared_unbound. Strict documentation for exact P2 completed 132 canonical pages and 264 reviewed records; Doxygen/gettext/EN/ZH, the official package and its actual restored verification all passed.

The complete-plan comparison on the same P2, tools and environment retains original per-stage costs and failure diagnostics. Baseline/repeat cumulative stage durations were 6742.878869825/6387.338375682 seconds, while attempt interval-unions were 6742.710856041/6387.179537077 seconds; these are different measures. Maximum observed process RSS was 730820608/730624000 B, not a simultaneous process-tree peak. Cache reuse and storage relocation make this comparison observational; fresh_cold and controlled_causal_speedup are both false. F01 retains 157 same-name checks, two added tool domains and the original case mappings for three retired registrations. N09 also has actual fresh-build-directory configure/build and warm configure/build within the same CodeQL trace. All four compilation databases have identical 2642 records, and the warm build actually compiled 15 times and linked 10 times; this does not close independent exits for other MCUs, CLI or F05.

The unsigned public delivery is obtained through the pinned 12-part transport descriptor. It binds a 401386855 B outer archive; an independent join actually restored 6657 files / 433814378 B. At recovery time, the same producer and qualification artifacts that were still live then passed verify-local. Read the restored local-release.json first for archive/member mappings and original evidence identities. The remote package does not contain all Native binaries/objects; 42+4 local stores/fullindexes retain complete recoverable bytes/modes. No remote Native byte reconstruction or automatic replay of original paths/nsmtime is claimed. It is unsigned and local_observation_no_oidc, and establishes no product Release, signatures, hosted CI or physical-board qualification.

P2 Python actually completed 174 queries, with all 45 findings individually reviewed. C++ completed a cross-attempt analysis on the same immutable inputs: 109 existing results validated by full decoding plus 74 actual additional evaluations, followed by a fresh full-suite interpretation returning 0. All 183 queries map one-to-one, and all 1234 original findings were individually reviewed without missing or duplicate entries. This is neither a single fresh 183-query run nor a finding-free result. Original OOM, timeout and cancelled attempts remain failed. Native and portable restoration of the Python/C++ databases, retirement of derived expansions and restoration of temporarily relocated Native/TMP evidence have separate receipts; early live fields are not current-live assertions.

Actual RED/GREEN evidence covers the Python FD failure paths, GoogleTest RE copy assignment and lwIP fixture fatal-assertion cleanup defects. Machine dependency inventory also corrected confusion between the official upstream base and selected organization patch. The independent QA followup commit is 706ee80b12a653a1820685c530b21e9668d82064, tree 5c39d9cfce75ed33354a13e1329c6e668b0463d2; its 13-path ordinary complete-index gate passed 420 Google programs / 7342 cases and 3968 tool cases, with zero failures, errors or skips. Its GoogleTest pin is organization patch 58ec32c0 on official version 1.16.0 base 6910c9d9. A new Python database was extracted for this SHA and all 174 queries actually completed: create/analyze took 36.224848/96.868403 seconds and both returned 0. The 38 findings comprise 30 notes / 6 warnings / 2 errors. Seven old findings disappeared: two actual FD failure-path defects and five equivalent cleanups, with no new or ambiguous matches. All 38 survivors uniquely map to previously reviewed dispositions: 36 findings have unchanged source blobs, and the two survivors in changed files were independently reread. The two errors retain their raw severity and Root false-positive dispositions bound to the exact source identities; this is not a zero-error or finding-free result. Firmware, C++ analysis, SDK, eight candidates and both complete 159-stage runs retain their P2 identities without transferring qualification; no third 159-stage run was performed in this round.

Current private E is fixed to f59debf1a15913353462e69ba0039db1c010f284, with platform pin P2 ab532250. All 17 presets / 116 maintained CI commands returned 0. Native execution had 28 CTest tests / 18 Google programs / 204 cases, and installed execution had 2 CTest tests; counts across domains are not added. Source and installed builds each passed five real C broker scenarios. Exit code 4 for a deadline and exit code 6 for a missing client certificate were expected rejections, with join/release/cleanup actually completed. Three usage recipes were covered by this complete execution. The public installed-SDK walk completed 12 commands and 1 CTest test, with SDK payload and origin unchanged. Private application source, payloads and original logs remain private; only approved aggregates are public. The broker helper retains qualification=false; actual C cases and ownership cleanup were checked separately. The first source-broker Docker bridge environment failure is preserved. After restoring the original docker0, the existing maintained bridge contract actually passed; the first failure was not rewritten. After authenticated confirmation that the repository is private, E completed a normal nonforce push and exact readback of the same f59debf1. Private main is unchanged, and physical execution remains unperformed.

Later D independent bilingual review/strict documentation package/restoration, actual deployment readback and final FF must each bind their actual documentation source. Original server-rule HTTP403 and Release-upload HTTP401 remain unchanged. All 42 physical items remain not_executed; Flash/ECC event-owner/sticky boundaries, seven vendor-leaf proposals and the GD303 reference review are not closed by this software result. Complete data and obtainability boundaries are in source_version_records_20261103.final_public_source_P2_ab532250 of the same results index.

Historical eb4 development records and recovery progress

The accepted public development source at that historical stage was eb4aba660df27b6ecedf52fcb13dc1c95652ca0d, tree 7586629e789ab5e7cb12cff8609694c04d684a09. Its ordinary complete-index Native Debug gate actually passed 420 Google programs, 7335 cases and 3950 tool cases, with zero failures or skips. The I/O-property report records 233 properties with bounded execution and separately lists 8 unproven entries; unproven entries are not part of the executed count. These scopes are not added together and do not establish the complete 159-stage formal plan or a new candidate.

The clean Source SDK v14 at earlier 6c3a380a actually completed 59 roots, 62 images, 246 commands and separate SDK verification; clean English and Chinese builds also completed. Some original TMP/SHM receipt copies were lost in the outage. The retained ordinary-gate archive f8da57a5…cca7 and SDK archive db2576f7…c41f keep their original identities. This update only checked those archives against their original SHA-256/bytes; it does not reconstruct lost originals or claim another v14 verification. Full digests, original restore records and obtainability limits are in source_version_records_20261103 in the same results index.

The current combination includes RX payload loans, MCUboot control, TX error observation, PCAP, resolver/CLI refinements and SDK v15; the ordinary gate completed for the exact source above. The current exact-source record in the same results index separately states progress for all 64 installed-SDK roots/67 images, independent verification and archive audit. Earlier 205-case focused TLS checks, model-only NativeBoard admission of FLASH=0 and resolver RED→145-case GREEN retain their original development scopes; they do not replace actual current-source records.

The recovered F407 network development reference linked with effective -Os on all 217 actual C translation units, no LTO, 134304 B Flash footprint, 116024 B total SRAM and 0 B libc heap. It exceeds the fixed MCUboot 128 KiB reference slot’s 129536 B payload allowance by 4768 B; there is no fit claim. Five real host-broker scenarios and ten PCAPs have a separate host-protocol scope. These results do not establish physical PHY/DMA, trusted UTC, physical entropy, performance, stack high-water or a new software candidate.

The earlier private examples baseline 2e932bbc (PR 24) pins SDK 7fd36f8a; its ordinary gate has 86 Python cases, 23 CTest tests, 13 Google programs and 146 cases. Later accepted private development commit ff86eb49 pins public SDK 6c3a380a; its ordinary gate has 89 Python cases, 28 CTest tests, 18 Google programs and 204 cases, and source and installed builds each ran five real-broker scenarios, all with zero failures or skips. This private scope is separate from public eb4aba66/v15. Earlier 11 peer cases, 4 main cases, 5 broker cases and style checks on 24 files retain their original source scopes and are not added to the new executions. The public platform records only aggregates and commit/PR metadata; private source, payload paths, receipts and archives remain private. PR 24 was a draft at that time.

This historical eb4aba66 record had no complete plan, fresh offline reproduction or new firmware candidates at that time. Original v0.2.1 and 7590b2d4 executions remain unchanged. Later P2 qualification and the QA followup appear in the current section above; server rules and HIL retain their independent original acceptance conditions.

0.2.1 software preview: b1b903da

The public v0.2.1 Release fixes platform source b1b903da47eda722a6be3e98edced5cfa36221b1, tree 56fc077ca7048d4758bdc050a98046de77824d08. It is a software preview, not a new formal software candidate, production qualification or physical HIL result. Subsequent source iterations do not inherit the executions recorded here.

Scope	Evidence entry and actual limits
Exact-source CI	The 31 platform CI jobs and four Security jobs actually succeeded; this does not replace a complete candidate plan or physical measurement
Source SDK	The maintained manifest and SDK archive; the 50 consumer roots, 53 images and 202 policy-v11 commands actually passed, without qualifying other combinations
Release	The actual publication workflow uploaded 13 assets; the release manifest binds their bytes, while the old v0.2.0 tag retains its identity
English and Chinese documentation	The actual Pages deployment and documentation manifest describe the same b1b903da source; later site updates follow their actual manifests
Formal candidates and physical scope	This preview establishes no new software candidate; physical status remains not_executed. See the original sealed baseline below for historical candidates, and judge OCI recipe and server rules by their separate actual results

The new 71 planning entries map to the same original delivery ledger. Added features, configuration or dependencies require their own actual regressions, SDK and candidate records; navigation changes alone close no software, physical or server-rule exits.

SDK v15 plan requirements for current development source

The maintained nexus-sdk-consumers-v15 adds the following five roots to v14’s 59 exact consumer roots, requiring 64 independent roots and 67 images. Selections come from the same plan authority; counts describe plan structure rather than execution receipts. Published v0.2.1 v11 and accepted 6c3a380a v14 retain their original identities.

Added profile	Planned consumption scope and limits
f407 loaned rx	External F407 C11 payload-loan link with explicit Storage/profile selection; physical MAC/PHY and complete SRAM fit require separate acceptance.
gd470 loaned rx	The independent GD470 ARM context checks the same selection without inheriting F407 execution or physical conclusions.
native mcuboot control	Native-model explicit operation binding and permanent-request readback; no physical Flash or boot conclusion.
f407 mcuboot control	Explicit application-side generic Control and concrete Flash Port composition, without linking the full boot core.
f407 mcuboot boot	Separate boot/payload images and reference-layout links; product partition, trust and physical recovery obligations do not close automatically.

All five new consumers have actual observations across two nonpublishable development prefixes: two loaned-RX consumers at snapshot 188d1b54 and three MCUboot consumers at snapshot e2ac91bd. They retain their own source and failure/success scopes and cannot be combined into one clean 64/67 execution. Interpret the complete producer report, separate SDK verifier and archive audit through the current exact-source record in the same results index; development prefixes or plan counts do not replace acceptance results.

Current acceptance of the installed SDK, strict bilingual documentation and later private executions is recorded in the exact source-version entries of the results index above. Documentation records bind the actual clean source, four build commands and recoverable HTML/receipt archive; later status-text checks cannot impersonate that HTML build or firmware qualification.

Interpret three identities separately

The following fields apply to successfully packaged formal sites and offline packages. For build previews, first inspect source, status and original command receipts in documentation-build.json. A preview has no formal documentation-manifest.json yet and does not establish release completion.

Identity	Location	Meaning
Current code source	code_source in the site’s documentation-manifest.json	The checkout, tree, dirty status and code paths described by this site
Current documentation source	documentation_source in the same manifest	Exact commit/tree for this formal documentation package; not a replacement for preview-build records
Referenced qualification source	qualification_references[] in the schema 2 manifest; each entry retains its original candidate identity and scope	Earlier exact-code execution; no qualification automatically inherited by current code/documentation HEAD

After successful packaging, manifests for the public site and offline package derive from actual build inputs. Documentation commits may change Git HEAD; this page does not manually insert a rapidly drifting HEAD constant. An unpackaged working tree can run git rev-parse HEAD to check its checkout, but that command output alone is not a qualification receipt.

Schema 2 no longer implicitly references a historical index. The packager explicitly selects each original sealed candidate; the tool invokes the same candidate verifier to check its original closure before recording source, configuration, ELF/BIN/map, qualified scopes and the candidate file digest. With no candidate selected, the list is empty; this does not mean software failed. Offline documentation verification protects these reference bytes and structure; it neither reexecutes nor replaces original candidate qualification verification. The original locator identifies its storage location at that time, not a public download URL. Historical schema 1 packages retain read-only verification of their original qualified_code; it cannot produce a new default qualification reference.

Final cold-lifecycle delivery: 7590b2d4

This section describes the completed sealed baseline; later code and documentation commits do not automatically inherit its qualification. Platform 7590b2d457d6042112abb78a4f4df74d613df6ff, tree 17c18aea10066d5d7418f3a7fd3b42e619ee3414, was integrated into the then-current main through PR 41, without creating another source merge commit. The associated external examples source is b029cc4b7e8e7a7d3e5e1f1be080a0aa24b1ee09; its SDK gitlink pins that platform commit, and examples PR 18 was merged. The examples repository is currently private; public platform access does not include examples access.

The public platform delivery index is fixed at evidence commit ecb86832fced0192d620b9029ff2d54795fc6430. Read public-recovery-proof.json in the same directory first: the initial payload index preserves observations made at upload time, while the recovery proof records subsequent actual downloads, digest checks and independent restorations. Do not interpret an initial pending as the final result. Both retain their original bytes and do not rewrite each other.

Scope	Actual completion and limits
Formal software plan	All 159 maintained stages actually passed on the same clean platform source, and the owned verifier passed. The original execution report retains software_qualified=false; plan execution and candidate sealing are assessed separately.
Four Native modes	Debug, Release, ASan/UBSan and TSan each executed 406 CTest entries, 386 Google programs and 6819 cases, with no failures, errors or skips. Four tool domains executed 3481 Python cases in total; do not count them again per Native mode.
Static analysis	Clang-Tidy and Cppcheck each analyzed 1135 distinct compiler contexts. This software analysis does not replace current-source CodeQL association or physical measurement.
Source SDK	50 independent consumer roots, 202 commands and the maintained exact policy-v11 selection actually passed; the original plan constrains all 53 full-policy images individually. A count alone cannot authorize a substituted selection.
Eight software candidates	H743 HSI64/Cache off and U575 MSIS4/TZEN=0, each with empty/UART and baremetal/FreeRTOS: 16 fresh network-disabled builds, candidate sealing, verification and recovery completed. This does not qualify advanced DMA, Cache or low-power modes.
English and Chinese documentation	For the original platform source, 132 pages, 264 language records, strict API/gettext builds, offline packaging and independent recovery completed. The documentation package has no implicit candidate references; actual Pages deployment remains unexecuted.
Remote evidence recovery	Exact-commit downloads and SHA-256/byte checks passed for this increment’s 335 public payload members, and all 90 original archives were independently restored. Original failed and interrupted observations remain retained.

The original outer candidate-sealing execution session was lost; its original outer receipt remains running/null. The 24 terminated child commands and later independent recovery proofs establish their own actual completion. Do not invent an outer exit 0 or discard completed child originals. Main integration only synchronized identities and pinned dependencies; it neither reran nor claimed a new software plan, build or sealing run.

The examples authenticated recovery entry is fixed at private evidence commit 1baac7b7946a3d71e855e41365972b0781af5291. Read it using existing GitHub repository authorization; do not copy examples source, private execution reports or archives into public platform artifacts. Readers without examples permission can use the public Source SDK. An inaccessible private link does not indicate platform recovery failure.

Exits still open at sealing time

Work	Acceptance that still needs actual execution
Three-board HIL and performance	Connect real stations and execute startup, clock, UART, DMA, stop and recovery, then measure IRQ latency, throughput, power and peak stack use. These are currently all not_executed.
OCI recipe and a fresh daemon	Independently complete two fresh builds from the locked recipe and actually load, inspect and run in a fresh daemon. Existing OCI-byte recovery does not establish recipe reproducibility.
CodeQL, Pages and server-side rules	Obtain and disposition actual current-source CodeQL results, deploy and verify Pages, and read back branch rules from the server. Local gates do not replace these facts.
Advanced external usage paths	The 7590 baseline still needs independent external recipes for H743 Flash/RAM and U575 retained-time/Stop2. Existing C mechanisms and SDK link checks do not establish a complete usage tutorial or physical execution.
Historical examples evidence boundary	Classify existing public examples evidence before deciding retention, migration or withdrawal. A metadata inventory does not establish review of every archive member; history has not been rewritten.
Formal Release	Define trusted admission for local CI and actually publish. Do not invent Actions runs or treat an unsigned software candidate as a product release.

Current firmware hardware algorithms remain in the maintained C11 paths. Python performs strict configuration, static object assembly and fixed call wiring. Subsequent work prioritizes usage, delivery and actual measurement; this sealed summary does not establish whole-family coverage, every peripheral or completion of all original exits.

Subsequent maintenance observations: 2026-10-30

This maintenance continues on branches from the platform and examples baselines above. The existing 78-task ledger updates only statuses and indexes proved by existing evidence, and the original renderer derives its readable view. All original IDs, ordering, exit conditions, physical statuses, historical Git versions, original reports and candidate bytes are preserved. Statuses bind their original tested sources separately. Formal qualification still belongs to 7590b2d4; subsequent development executions are recorded separately. A new documentation HEAD does not replace qualified source or relabel that baseline’s 159-stage result as qualification of new source.

Platform increment dee8be5bb1052442a7853cf792b3fd589e17a9c9 actually passed its ordinary complete-index gate: 3565 Python cases (1755 testing, 1156 configure, 493 CI and 161 workflows), plus 386 Google programs and 6819 cases, with no failures or skips. This is development-gate execution for that commit, not a new complete 159-stage formal qualification or an expansion of old candidate scope.

Subsequent private examples commits ac9f07ba7494f472144f37660aabbd4185710394 (PR 19) and 5fb06175aab72292fb075e8a9ea67096203a86dc (PR 20) were both merged. Current examples main is the latter, while the SDK gitlink still pins verified 7590b2d4. The final ordinary gate actually passed 85 Python cases, 21 CTest entries, 12 Google programs and 129 cases, with no skips. Independent projects, actual ARM links and resource checks now exist for H743 Flash/RAM and U575 retained WUT, SRAM4 copy and single-tuple LPTIM1 LPBAM; the H743 RAM execution closure was also audited. This completes E04’s currently maintained selected advanced recipes as software development execution. It does not establish HIL, general LPBAM, new platform formal qualification or a new software candidate.

The initial LPTIM draft preceded its corresponding tests, as disclosed in PR 20; later genuine RED-to-GREEN fixes retain their original execution. Reconstructed source is not used to invent earlier test chronology. Parallel TDD now requires the test owner to retain actual RED, source and binary before releasing the implementation owner’s handoff barrier.

Exit	Actual observation in this maintenance
GitHub branch rules	Writing still returns HTTP 403 Resource not accessible by integration; rule readback is empty. Protection cannot be claimed as enabled.
CodeQL association	The current API read returns HTTP 403; current-source scan results have not been obtained for disposition.
Pages deployment	Not actually deployed. Local HTML and archive recovery do not replace remote-site verification.
OCI rebuilds and a fresh daemon	Available capacity does not meet the existing preparation estimate. Two fresh recipe builds and fresh-daemon execution have not started. The existing OCI byte identity and non-equivalence boundary remain unchanged.
Launch-board tooling for three boards	Three-board prepare and measurement-plan operations completed six actual CLI commands with exit 0; status remains prepared_unbound. Physical acquisition and performance measurements have not executed.
Historical publication boundary	Private retention and withdrawal of 21 members from the current public branch tip actually completed. Old commits, caches and clones may still retain original data; history was not rewritten.

These external exits remain open while software implementation continues. Original rule-write commands and diagnostics are at build/completion-20261030/remote/ on the execution host. This is an original host location, not a public download URL. For current complete records, first inspect the actual delivery index for the subsequent commit, then restore its fixed identities.

Actual historical-boundary mitigation is recorded in the public withdrawal notice. The private retention authenticated recovery index is fixed at 8aa825c7d2cd8f58f4fa84869f16700aea400783. Ordinary hooks, non-force pushes, exact-commit readback and member checks actually ran. Old commits were not rewritten; withdrawal from the current tip is not historical erasure. The current 7590b2d4 platform recovery proof and private examples 1baac7b7 entry retain their original identities.

F04 retains the original review’s contract/oracle mapping level; it does not claim that every historical project case was manually proved semantically independent. The same ledger preserves original exits for real hardware, complete controlled cold/warm comparison and environment reconstruction. A complete software-plan pass cannot change all of them to completed.

Historical independent result indexes

Scope	Exact code source	Original authority and execution explanation
C cold-lifecycle and engineering delivery increment	194ca6cb; d21dfcab; 364f3665; later worktrees are recorded separately	The current_iteration_20261028 object in the same results index retains each scope’s source, ordinary gate, ARM, SDK, documentation and archive; the original 64 exits are unchanged. Development results do not establish full qualification of the final integrated source
Organization-hosted dependencies and controlled delivery	N2 ce616ea9b30fd2bbd9511f73e2f3b5d8028224b4；E1 cef24bd1c7bc500fb28a17a6e1ff9cc7d5511711	The original formal software plan passed all 159 stages; external examples passed 98 commands. Six scopes passed for the exact-configuration candidate. The original candidate remains unsigned; physical HIL and product promotion were not executed.
Typed IR, cold integration and delivery unification	P2 fc01d863c3374643c276195ae19a24e6f8b8c8ee; E acb5179ec09e0c2fae1911d32655296df39acedc	Exact execution record; Q02 passed all 159 stages and E passed its complete local execution; the exact-configuration candidate passed its six scopes; Q01 failure remains separate, remote status publication is unconfirmed, and physical HIL is unexecuted
SoC/IP layout refactor and original candidate	P 92e669233957a2e450a528394e8f4dd0794c7049; E 0f15022e4d5f5c51991b134d0b9d6082dcf88de9	Execution explanation, 48 original ledger items and derived view; A90+B69 segmented coverage of the same P, the original candidate covers only its exact configuration, and physical HIL was not executed
Developer-experience iteration P and examples E	P c8df1c49d4f771a55490b3a8be32a9d1f24faa1e；E acb923cae407c8cfc4ac5788f8026d35c84cc841	The results index preserves complete local gates, independent SDK consumption, examples ARM results and original archives; it is not a formal software candidate
Build system and its actual software scopes	cc6a065c002041234a212a964be863a69198f9b4	Results index, execution records
SoC extension software slices	98270512ea2ff7eefe8ff79720018e582be3a949	Results index, execution records
Earlier SoC refinement baseline	3728b334b02eb2eb70bcf5d4caafbb9746837dd3	Results index, execution records

The original N2/E1 local candidate navigation is recorded in the external examples repository at build/controlled-dependencies-ce616ea9/delivery-index.json and build/documentation-controlled-dependencies-ce616ea9/delivery-index.json. The former locates original candidates, artifacts and portable evidence; the latter locates bilingual documentation packages and independent offline recovery records. Documentation publication remains published=false. These are original host locations, not public download URLs. Nexus merge commit 3ea252c6 has the same Git tree as tested N2, and examples merge commit 4a9f5f75 has the same Git tree as tested E1; the association retains the tested source without rewriting candidates. In the later C lifecycle increment, 194ca6cb completed exact-source SDK consumption and strict bilingual documentation delivery; d21dfcab completed its ordinary complete-index gate. Later pending fields in that index are development observations from their original time. Final gates, SDK, examples, documentation, dual builds and sealing for 7590b2d4 are described earlier on this page. Family development counts must not be added together or treated as coverage of all four Native modes or physical acceptance.

Each index preserves its original schema, scope, failed attempts, receipts, archive and member identities. Do not add test counts, ARM images or task counts, or calculate a project-wide completion rate. Later documentation commits only record these baselines. Current documentation checks, rendering, offline recovery and deployment have their own records.

The build-system record did not execute one complete 72-stage software plan, an OCI recipe rebuild, CP311 offline installation, Windows/macOS integration or MCU HIL. Interpret separately executed scopes through the original index rather than expanding an aggregate status. Its eight unsigned software candidates are H743 HSI64/Cache-off and U575 MSIS4/TZEN=0 empty/UART baremetal/FreeRTOS combinations.

Acquire and verify earlier evidence

1. Open the corresponding results index and confirm the canonical code commit, configuration and recorded scope.

2. Locate the archive through the index and verify SHA-256 and bytes; retain the original identities of failed attempts.

3. Locate members using original fields such as archive_relative_path and archive_member_relative_path, then verify each member’s hash and bytes. Field names vary by original schema; interpret them according to that index.

4. Inspect actual execution records, raw logs and the same artifacts. Software models, linking, independent consumption and two network-disabled rebuilds prove their own scopes; success in one does not replace another.

5. Assess physical records separately according to HIL status.

Original /workspace/... and /dev/shm/... paths locate the execution host; they are not public downloads elsewhere. The build-system index’s evidence_navigation specifies candidate archives, candidate documents, ELF/BIN/map and complete member-inventory locations. Archives are currently retained locally as recorded; do not invent download URLs for unpublished remote evidence. Documentation packages include original JSON for checking; they do not copy large raw archives or re-sign historical bytes.

FIG-11 expresses qualification relationships without generating receipts. See the diagram atlas for standards and source authorities, and the ledger views for original per-task exit conditions.