The original architecture plan now has 18 accepted tasks and 86 in progress. KL-065 is newly accepted for the exact recorded Linux source below. Acceptance is scoped to the original task criteria; current whole-product coverage, RF3, endurance and performance qualification remain open.
Source: f0d869c08f1a9efe30e93911c4631ad26eb87fb0. The successful Linux task job, attempt 1, executed 78 cases in 20 selections: 45 Unit, 25 process recovery and 8 Aspire RF3. All passed; none failed or skipped. RF3 operations used the actual .NET client and official MCP SDK against the Aspire-owned Docker cluster.
| Original architecture acceptance criterion | Executed operation evidence |
|---|---|
| Cross-tenant IDs do not collide in cache | GenerationVisibilityTests proves complete tenant/epoch-scoped SQL and graph sets, held positive/negative leases, retirement, shared admission, cancellation, original replay and same-root cold. The ZoneTree point-cache authorization flow and cross-tenant RF3 whole flow independently exercise current persisted scope. |
| Stale grants do not pass final authorization | Security/query, equality access paths, SQL inner join, partition query, document mutation and team policy operations prove persisted owner/team/project grants, removal, revocation, higher-epoch repair and complete original outcomes. Public SQL/partition and graph capability RF3 operations verify SDK/MCP refusal and authorized continuation. |
| Hidden graph paths do not appear in response or Explain | Incoming privacy, traversal, shortest-path and graph-search operations preserve hidden source/target/intermediate policy and field-use authorization. Five graph RF3 cases exercise real SDK and official MCP responses, persisted revocation, repair and revisioned projections. |
Root independently checked all 189 original length/SHA references, the exact old registry from Git at this source, unique genuine native UIDs and declarations, all original TRX identities/outcomes, exact Git source/PDB mappings for 6,592 files, compiled-source receipts, unchanged prepared/before/after images, native exit/readers/disposal and the source-bound Docker image. The original CI acceptance verifier itself exited successfully with all readers joined. DLL/PDB binary digests are authenticated CI receipts; binaries are absent from this artifact, so local binary digest recomputation is not claimed.
Authenticated artifact: task-acceptance-KL-065-normal, 11711112499.
| Original | SHA256 |
|---|---|
| GitHub ZIP | baeb291d60863d793ea873bbfdc1a8c16bacd784d0ff2ed64e8e6f07925dbc5b |
| Original acceptance receipt | c10e39fd5129e46a9517f453c7a6a93cd3f3eaf462761e72ec0fd967d9df1f74 |
| Original execution manifest | 6b479297111e2c08098e179dc489ffc847dd56954fc32a8295a6bcbebdcd193c |
| Independent observation | 52d23f3f12c4f7a02708f913ed94a57519f301a9393f6b2a57e65b68df7aec37 |
The independent observation is retained at /private/tmp/keyload-stage58-f0-linux-intake-r1-20261011/kl-065-independent-intake.original.json. The durable status records the source, job, artifact, hashes and criterion mapping. Later shared execution changes require their own regression checks.
The command inbox now preserves control admission while serving actual admitted tenants in bounded round-robin turns. It retains the ordered apply gate, original limits, cancellation, shutdown outcomes and lease ownership. The new whole operation proves A1/B1/A2/A3 order, principal/tenant refusal without effects, cancellation, shutdown, complete receipts, replay and same-root cold continuation. This is supporting KL-076 implementation, not its complete CPU/RSS, mixed-client or Linux performance gate.
Replica security fixtures explicitly borrow the owning TimeProvider. The signed expiry operation retains exact original MAC/refusal/reply/replay bytes. It is a required transport control and remains a noncontributor: verifying the accepted envelope does not dispatch a database write.
Due-work algorithm fixtures now use an explicit logical provider while preserving the actual native ZoneTree reader, range accounting, cancellation, original deadline predicate and iterator/gate ownership. Their deadline operation proves exact BudgetExceeded, complete no-effect state, healthy continuation and cold replay. The original Linux system-clock failures remain evidence; no product latency repair or enlarged deadline is claimed.
CodeQuality uses native filesystem enumeration without changing path/quota/link checks or source digests. Functional report admission requires globally unique genuine native identity and a complete required-case mapping before publication. Known optional nonempty source rawUrl is supported with required line/range links; unknown or malformed fields still refuse. The owning whole operation preserves originals, tests duplicate/swapped/source/status/unknown-field refusals and joins every native child.
The strict task registry includes the three previously omitted credential-status operations for KL-014 and the original ClientRemoteLink telemetry instance for KL-015. This fixes task-selection inventory; it does not claim their subsequent Linux operation lane passed. KL-015's original held-write/producer-settlement RF3 failures remain open.
R40 registers only the actual two new native instances: tenant admission in existing functional group 05 and the signed clock transport control as a required noncontributor. The existing 3,478 identities/classifications, all five selectors and 61 contributors are preserved; the five due-work entries only refresh observed source/span bindings. The registered inventory is 3,053 functional plus 427 required-control instances. The seven-case discovery is selected actual metadata, not a complete current census or numerical coverage result.
The full Release solution build passed with 0 warnings and 0 errors, approximately 150 seconds, with 8,696 source inputs and zero drift. Its receipt is /private/tmp/keyload-stage59-native-orleans-owner-rule-r1-20261011/main-build-r3-completion.original.json. Native Roslyn analysis also reported zero errors/warnings; informational findings are retained and are not all declared resolved.
The focused native TUnit cohort executed eight whole-operation cases with configured parallelism 50 and normal SIMD. Seven passed and one failed, in approximately 95 seconds. The tenant case failed during genuine snapshot-directory admission because macOS's original /var/folders/... temporary path traverses the /var symlink; the product guard correctly refused it. No storage guard or test assertion was weakened.
The same unchanged tenant case subsequently passed with TMPDIR set to the existing canonical /private/var/folders/... directory, in 1.449 seconds. The native runner exited zero; source drift was empty. This separate one-case result does not rewrite the original eight-case failure or manufacture a new passing cohort. Originals are in TestResults/stage59-focused-originals/, including native discovery, DLL/PDB/compile identity, both process reports and TRX results. The five due-work cases, borrowed-clock control and CodeQuality whole operation passed in the original cohort.
After those executions, registry/classification and durable acceptance documentation changed. The canonical formatter checked all 18 changed C# files and found one whitespace-only IDE0055 failure in DueDiscoveryDeadlineFlow. The native Roslyn formatting action corrected that brace placement without changing operation semantics. A strict owning Unit project Release rebuild then passed with 0 warnings and 0 errors in approximately 105 seconds, retaining the previously built product images. Its original completion is TestResults/stage59-focused-originals/post-format-unit-build-completion.original.json.
The matching one-file dotnet format whitespace --verify-no-changes --no-restore successor passed in 7.377 seconds with native exit zero and both readers closed. Its original is TestResults/stage59-focused-originals/formatter-whitespace-completion.original.json. The original all-file formatter failure remains preserved separately; the unchanged operation semantics do not require rerunning the 92-second CodeQuality flow. Repository governance and git diff --check also passed; governance preserved the full existing root-policy prefix, 27 projects and four modules.
The original full Linux Unit run at the same f0d869c source executed 3,478 cases: 3,461 passed and 17 errored. The full process-recovery suite passed 349/349. The three-node database job failed; full functional coverage and its strict successful merge remain open. These results are distinct from the successful KL-065 task lane.
KL-035's new exact lane passed 49/49 with all 45 originals independently verified. Its earlier fc4e490 ordered-tail absence remains an unresolved regression; a later passing run does not establish a causal repair. Its status stays in progress. KL-079's original reconnect failure remains open with the full QueryPage and Outbox comparison preserved.
Native Azure/Orleans membership, six-silo physical trust, discovery, custom peer-HTTP/provider retirement and autonomous ANN original-result ownership are separate prepared implementation waves. They are not part of this built/tested source stage and do not receive runtime or acceptance credit here.
The installed quality skills were reviewed in the required order. Root .editorconfig, latest-all SDK analyzers, warnings-as-errors and KeyLoad semantic maintainability rules retain ownership and unchanged severity. The Release build and native diagnostics are actual first-party/semantic analysis checks. dotnet format remains the formatter; CSharpier is N/A by explicit repository choice. Roslynator, Meziantou and StyleCop package installation/CLI execution are N/A for this stage because the repository uses its existing SDK and KeyLoad analyzer baseline; none was introduced or substituted. The existing 400/200/64/3 limits are unchanged.
The added operations cover original refusal, healthy, replay and cold flows rather than field getters. CodeQuality admission/source binding is stricter and the tenant scheduler owns one bounded responsibility. Whole-product numerical coverage and source-matched CRAP are unmeasured until the authentic successful coverage gate; missing data is not fabricated as zero or a passing score. Canonical formatter and repository governance receipts are retained in TestResults/stage59-focused-originals/ as original process results before the stage commit.