Remove redundant ANN reciprocal scan with native parity evidence

Preserve exact graph/options/distance/edge results in actual 128-row and 10000-row normal/scalar controls. Broader ANN normal cases passed 79/79; scalar cases passed 78/79 with the original default all-metric Euclidean build deadline failure retained. Full qualification and acceleration remain unclaimed. Include current requirements, native development receipts and exact-source CI cancellation status.

Author: KSemenenko

docs/ADR/ADR-019-managed-ann.md

@@ -202,3 +202,12 @@ all strict/Aspire gates. A partial failed build cannot prove graph equivalence
 or speed. Migration is N/A: R1 remains an internal immutable candidate with no
 serialized/public contract; R2/R3 require separate accepted contracts. Rollback
 restores only the removed scan and helper.
+
+Root's successful-control observation joins AC-ANN-018 before code: fixed v1
+SHA-256 framing covers the actual complete native IDs/revisions/levels and
+ordered adjacency plus entry point/maximum level. Build-only ticks/allocations
+are captured before independent snapshot/assertion/output; actual counters and
+fixed options are retained. Identical control/test source runs on the original
+and candidate in each native mode. Failed builds cannot supply a graph hash or
+equivalence baseline. This remains test-only evidence with no public or
+persistence contract, and does not qualify acceleration by itself.

docs/ADR/ADR-082-native-cqrs-streams.md

@@ -206,3 +206,13 @@ the live independent consumer, normal original captures, exact caller
 cancellation, early stream completion and genuine SDK/MCP revocation outcomes.
 Rollback removes the test-only context. This addition is diagnostic evidence,
 not proof of the old cancellation cause or completion of this ADR.
+
+The accepted DIAG-006 implementation also observes failures at the original
+cleanup owner: each actual failure append notifies the same read-only lifecycle
+owner before later completion/fallback/join/disposal work mutates its states.
+This narrow optional feature-local callback covers captures, subscriber and
+independent-consumer owners and authority cleanup stages, without changing the
+shared failure observer, original ordering, tasks, exceptions or deadlines.
+Record observer failures without skipping native cleanup. Root freezes this
+ordering correction before private implementation and owns native regression
+evidence; dependency_closeout owns the guarded diagnostics packet correction.

docs/ADR/ADR-100-local-partition-query-merge.md

@@ -3,6 +3,16 @@
 Status: Accepted; implementation and qualification pending.
 Date: 2026-10-05. Related work: KL-037 prerequisite only.
 
+The accepted TASK-DQUERY-CANCEL-OBSERVATION amendment implements
+AC-DQUERY-PREREQ-005 with an already-armed, owned native observation thread and
+the unchanged synchronous real ZoneTree query. Exact case/helper ownership,
+15-second observation/join bounds, original token/budget, no-result and healthy
+follow-up requirements are frozen in DistributedQueryExecution before code.
+Root owns review, joins and Aspire normal/scalar/full-gate evidence;
+dependency_closeout owns the private test packet. Preserve the earlier failed
+original report. No product boundary, data/wire format or migration changes;
+rollback may revert this test synchronization but cannot weaken its assertions.
+
 ## Context
 
 KL-037's original target requires query fan-out across logical shards. The

docs/ADR/ADR-103-scaled-fair-comparisons.md

@@ -72,10 +72,41 @@ Consumers retain the exact sidecar and authenticate its original artifact/provid
 
 Worker owns feature-local AppHost Contracts/Observation/Processes roles, minimal composition/lifecycle joins, bounded parser/identity/native regressions and scale-only script consumers. Root owns durable feature/ADR freeze, source integration, actual Aspire native Linux/Docker qualification, CI/publication, receipts and commits. Private implementation may rely on the existing 45 engine + 10 Aspire + 25 CI packets as explicit predecessor source, never silently overwrite their bases. A reviewable patch and base/post manifests are required. No checkout edits, gates or Git by the worker.
 
+The accepted SCALE-016 native-root correction follows the documented cgroup-v2
+root semantics: CPU/memory maximum interfaces exist on non-root cgroups. Preserve
+all actual non-root ancestor minima, effective cpuset and verified hierarchy
+identity; terminate at the real root rather than requiring nonexistent root
+limits. Missing/malformed/unreadable required child observations remain explicit
+unqualified evidence. The independent actual Linux oracle and supported-envelope
+regression must expose that distinction. Root owns integration/original Linux
+gates; partition_pages owns the guarded repair. No sidecar or stored schema
+changes occur and rollback cannot fabricate limits or reduce complete cohorts.
+
 ## Accepted prerequisite: Aspire scale forwarding (REQ/AC-SCALE-014)
 
 Introduce only the separate KeyLoadTests:ScaleProfile test-harness selector. TestSuiteSettings accepts one exact canonical scaled ID only for Suite=comparison, the exact /*/*/IsolatedNativeComparisonTests/* filter, present native Benchmarks:Target, matching Benchmarks:EvidenceProfile, disabled Benchmarks:Enabled, no direct Benchmarks:ScaleProfile and no workload overrides. Reject missing suite, other suites, unknown/blank/case-mismatched IDs or mixed modes before any resource creation. Existing direct Benchmarks:ScaleProfile plus a suite remains rejected.
 
 run-workload passes --KeyLoadTests:ScaleProfile=<id> to the outer AppHost. Its owned runner alone receives Benchmarks__ScaleProfile=<id>; clear KeyLoadTests__ScaleProfile alongside KeyLoadTests__Suite so the harness selector does not leak into the nested native AppHost. IsolatedNativeCase reads the exact ordinary ComparisonWorkerSelection from the runner environment and passes --Benchmarks:ScaleProfile=<id> to its own nested resource-owning AppHost. Preserve exact evidence/source/native topology and all control behavior. IsolatedNativeCase uses exactly 140 minutes for the closed scale profile and its existing 60 minutes for controls; all original tasks/resources are still joined.
 
 Worker owns the narrow TestSuiteSettings/TestSuiteResources/IsolatedNativeCase/run-workload joins and real Aspire model plus independent argument/environment regressions. This is a prerequisite repair to the accepted SCALE-014 path, not a new alternate test caller. Durable docs join before live implementation.
+
+## Accepted stage 11: original teardown settlement, 2026-10-05
+
+REQ/AC-SCALE-017 and TASK-SCALE-ORIGINAL-TEARDOWN repair the inspected existing
+IsolatedNativeTeardown path, which detached a pending task after 30 seconds and
+replaced or suppressed actual cleanup failures. Freeze the exact owner/order,
+original-task join, native fatal classification and primary/cleanup preservation
+contract in ScalingQualification before implementation. The collector joins from
+stage 10 use the same lifetime; the existing control success path/report schemas
+remain unchanged.
+
+Implement in order: retain the original case failure; settle the original
+collector/capture and report writers; stop and dispose actual owners; delete data
+only after safe ownership release; write bounded safe categories; propagate the
+original ordered failures after all safely reachable stages. Keep 30 seconds as
+an escalation/failure threshold, never detached completion. Add genuine native
+Cases/Helpers regressions for pending settlement and simultaneous primary plus
+cleanup failures, and run them through the canonical Aspire comparison entry.
+ComparisonTests owns this code, partition_pages owns its private guarded packet,
+and root owns integration/gates/evidence/commit. There is no data or wire migration;
+rollback cannot convert an unfinished original task into a passing qualification.

docs/ADR/ADR-106-partition-owner-movement.md

@@ -173,3 +173,51 @@ qualify it. Native membership CAS, persisted authority, schema/byte limits and
 public data-admission closure remain unchanged. Root owns the join and gates;
 Luna lifecycle_wave owns only the guarded implementation/test packet. There is
 no data/wire migration; rollback disables the new profile.
+
+## Accepted six-generation oracle and native parameter handoff, 2026-10-05
+
+REQ/AC-MEMBERSHIP-008 in PartitionTransfer closes the source oracle gap before
+implementation. The already-generated B physical ID/incarnation and secret are
+the same named Aspire ParameterResources used by the actual six-container model
+and resolved privately through native GetValueAsync in its owned test wave.
+No independently generated verifier credential, identity or membership provider
+is permitted. Decode the secret only within an owned joined/zeroed lifetime.
+
+The profile-only membership health reply carries a bounded closed fingerprint
+and counts derived from one actual native ReadAll call per node, including B's
+real proxy path. Its domain-separated length-framed SHA256 binds the six exact
+current Active SiloAddress generations without exposing addresses or row data.
+The independent integration oracle authenticates discovery from all six actual
+nodes using the shared A ClusterId and each distinct group incarnation/key,
+computes the expected active-generation digest, and compares all six native-view
+health responses. Boolean readiness alone does not prove this acceptance.
+
+Implement in order: same-resource AppHost handoff; bounded native view digest and
+closed health reply; six-node signed-discovery verifier; actual Aspire oracle
+and negative native controls; strict build/format and original exact-source Linux
+RF3 evidence. All code remains inside ClusterRouting responsibility folders.
+Existing three-node discovery helper, persisted membership, native CAS, MAC and
+serializer identities, wire caps, public closure, failover and original task
+teardown stay unchanged. Root owns integration/evidence; lifecycle_wave owns the
+guarded source/test packet. The additive health shape is confined to the new
+membership-only profile. Rollback disables that profile; Stage 1B data readiness
+and physical movement acceptance remain unqualified.
+
+## Accepted native authority route and provider ownership, 2026-10-05
+
+REQ/AC-MEMBERSHIP-009 and TASK-MOVE-1A repair two inspected source gaps before
+integration: the authority handler was not mapped, and externally registered
+disposable instances had no actual disposal owner. The exact native POST route,
+provider-owned factory/type registration, handler/DNS/call joins before endpoint
+credentials/gates and owner CTS disposal, failure preservation and negative
+admission contracts are frozen in PartitionTransfer before code.
+
+Implement in order: map the existing handler once in server composition; register
+its exact owner and endpoint with native DI ownership; preserve existing silo
+close/join then Kestrel stop/join then root-provider disposal; add native
+composition/lifecycle regressions and execute the existing actual six-container
+Aspire startup/authentication/teardown flow. lifecycle_wave owns the private
+guarded ServerConfiguration/transport/test correction; root owns source review,
+integration, strict checks, original Linux RF3 evidence and commit. No wire/data
+migration or public database opening occurs. Rollback disables the new profile;
+an unmapped route or undisposed owner cannot be reported as delivered membership.

docs/Features/BenchmarkComparisons/ScalingQualification.md

@@ -81,10 +81,57 @@ Consumers retain the exact sidecar and authenticate its original artifact/provid
 
 Worker owns feature-local AppHost Contracts/Observation/Processes roles, minimal composition/lifecycle joins, bounded parser/identity/native regressions and scale-only script consumers. Root owns durable feature/ADR freeze, source integration, actual Aspire native Linux/Docker qualification, CI/publication, receipts and commits. Private implementation may rely on the existing 45 engine + 10 Aspire + 25 CI packets as explicit predecessor source, never silently overwrite their bases. A reviewable patch and base/post manifests are required. No checkout edits, gates or Git by the worker.
 
+REQ/AC-SCALE-016's native cgroup-v2 walk must honor the actual hierarchy root:
+the kernel defines `cpu.max` and `memory.max` only on non-root cgroups. Walk
+every admitted non-root ancestor and retain its minimum effective limits; stop
+at the verified real hierarchy root without inventing missing root limit files.
+Observe the actual effective cpuset and supported hierarchy/controller identity.
+An unreadable, malformed or missing required non-root observation still makes
+evidence unqualified; it is never silently replaced by `max`. The independent
+native Linux oracle must use the same documented root semantics without copying
+the production parser. A supported Linux envelope test must assert its actual
+value instead of conditionally omitting a null result. This source correction
+changes no sidecar schema, workload, bounds or qualification requirements.
+
 ## Accepted prerequisite: Aspire scale forwarding (REQ/AC-SCALE-014)
 
 Introduce only the separate KeyLoadTests:ScaleProfile test-harness selector. TestSuiteSettings accepts one exact canonical scaled ID only for Suite=comparison, the exact /*/*/IsolatedNativeComparisonTests/* filter, present native Benchmarks:Target, matching Benchmarks:EvidenceProfile, disabled Benchmarks:Enabled, no direct Benchmarks:ScaleProfile and no workload overrides. Reject missing suite, other suites, unknown/blank/case-mismatched IDs or mixed modes before any resource creation. Existing direct Benchmarks:ScaleProfile plus a suite remains rejected.
 
 run-workload passes --KeyLoadTests:ScaleProfile=<id> to the outer AppHost. Its owned runner alone receives Benchmarks__ScaleProfile=<id>; clear KeyLoadTests__ScaleProfile alongside KeyLoadTests__Suite so the harness selector does not leak into the nested native AppHost. IsolatedNativeCase reads the exact ordinary ComparisonWorkerSelection from the runner environment and passes --Benchmarks:ScaleProfile=<id> to its own nested resource-owning AppHost. Preserve exact evidence/source/native topology and all control behavior. IsolatedNativeCase uses exactly 140 minutes for the closed scale profile and its existing 60 minutes for controls; all original tasks/resources are still joined.
 
 Worker owns the narrow TestSuiteSettings/TestSuiteResources/IsolatedNativeCase/run-workload joins and real Aspire model plus independent argument/environment regressions. This is a prerequisite repair to the accepted SCALE-014 path, not a new alternate test caller. Durable docs join before live implementation.
+
+## Accepted original teardown settlement (REQ/AC-SCALE-017)
+
+REQ-SCALE-017 requires the actual isolated AppHost owner to settle every original
+collector, capture, report-write, application-stop and disposal operation before
+releasing its dependencies or deleting its owned data. This applies to controls
+and scaled runs; successful workload results and control report schemas are
+unchanged. The existing 30-second teardown limit is a recorded failure threshold,
+not permission to detach the task with a continuation. After that threshold,
+request cancellation or escalation through the original owner's supported API
+where available, retain the threshold failure, and ultimately await that same
+original task. Do not start a replacement operation, use an uncancellable shadow
+task, or dispose an owner while its observation/writer still uses it.
+
+AC-SCALE-017 requires genuine native lifecycle regressions to prove that cleanup
+does not pass an unfinished original operation, and that cancellation and failure
+paths settle the original process/readers/capture before directory deletion.
+Retain the actual workload primary exception and each actual cleanup exception
+object, including nested aggregates and original cancellation tokens; do not
+flatten, replace them with a category string, or suppress them when a primary
+failure exists. Cleanup continues through all safely reachable stages. At final
+propagation, one original failure retains its stack, multiple failures retain
+ordered primary then cleanup objects, and native ManagedCode fatal classification
+remains visible and takes priority. The receipt contains only the existing safe
+stage categories and primary-presence flag, never exception text or payloads.
+
+TASK-SCALE-ORIGINAL-TEARDOWN is owned by ComparisonTests BenchmarkComparisons
+Helpers (`IsolatedNativeCase`, `IsolatedNativeTeardown` and populated settlement
+helpers), with Cases/Helpers for real native regressions. The resource collector
+joins from SCALE-016 participate in this same lifetime. Root owns the pre-code
+contract, review, Aspire execution, original evidence and commits; partition_pages
+owns the private guarded implementation packet. No change to provider topology,
+measurement timing, ACK/durability, source provenance or publication eligibility
+is authorized. Rollback may revert the repair but cannot claim successful
+settlement or qualification for a detached original task.

docs/Features/ClusterRouting/NativeCqrsRequestV2.md

@@ -820,6 +820,21 @@ extend deadlines or change the parser, artifact schema or success conditions.
 The diagnostics owner may expose a read-only lifecycle snapshot of its native
 cleanup owner; no alternate collector or inferred task state is permitted.
 
+The existing native cleanup owners must notify this same lifecycle evidence
+owner immediately when an actual operation first appends a failure, before the
+next resource completion, fallback cancellation, join, retry or disposal changes
+the observed state. This includes each of the three resource completions,
+bounded capture drain, fallback/capture join, observer and independent-consumer
+close/join, and each actual authority cleanup stage. A snapshot taken only after
+an entire failing cleanup owner returns is not first-failure evidence. Use an
+optional feature-local read-only observation callback and closed cleanup-stage
+labels; keep the original tasks, operation order, joins, failures and deadlines.
+An observation failure is itself retained and must not prevent safely reachable
+native cleanup stages. Do not change the shared ServerFailureObserver, add a
+second collector or filter native cancellation. The terminal snapshot remains
+after all original joins; actual callback timing and bounded context are part of
+the native diagnostics regressions, not proof of the old CI initiating cause.
+
 TASK-CRS-DIAG-FIRST-FAILURE first freezes this contract and ADR-082, then
 privately implements bounded feature-local lifecycle helpers and read-only
 joins in the existing diagnostics scope, cleanup, subscriber observer,