Record qualified SQL BETWEEN and bounded native benchmark failures

Author: KSemenenko

docs/ADR/ADR-065-full-sql-client-compatibility.md

@@ -243,7 +243,9 @@ REQ-SQLC-006 / AC-SQLC-006A / TASK-SQLC-BETWEEN is the next additive Q1
 expression stage under [SqlBetween](../Features/QueryExecution/SqlBetween.md).
 The owning feature fixes the complete truth/error/budget/test and ordered
 execution contract; strongest TASK-SQLC-R23 review joins before delegated code.
-Status remains Accepted, with all runtime gates pending.
+Status remains Accepted. [Original7d1196 CI](../implementation/sql-client-qualification-7d1196.json)
+qualifies the bounded stage's44 execution rows and complete ordinary CI;
+coverage, full SQL/native transport and complete performance gates remain open.
 
 1. Review the current parser/normalizer/evaluator/candidate/permission/cursor
    sources and retained ca7 baseline. R22 corrects partial-bound truth: UNKNOWN

docs/Features/BenchmarkComparisons/NativeGateRepair.md

@@ -13,7 +13,7 @@ BenchmarkComparisons guarantees,1/2/3node cells and270scenario scope remain.
 |REQ-NGR-002 exact Kurrent event preservation|AC-NGR-002 / TASK-NGR-R2/K1: source-contract freeze then genuine SDK original/custom/system metadata equality and unchanged55,378owned-stream/foreign/cleanup proof on1/2/3. StageA source delivered; selected f403 native1/2/3 originals pass. Empty tracing characterization remains open; a hard-coded empty assumption is not a valid completeness oracle.|
 |REQ-NGR-003 actual native leader routing|AC-NGR-003 / TASK-NGR-R2/K1: pinned SDK/advertised-endpoint source freeze, actual leader/follower/native setup proof and observed members/copies/ACK. Selected f403 native1/2/3 preflight proof exists; D1 diagnostic source is delivered. Original routing cause and diagnostic failure-path qualification remain open; no guessed leader, independent nodes or measured write retry.|
 |REQ-NGR-004 genuine complete publication|AC-NGR-004 / TASK-NGR-Q1: full source/normal/scalar/recovery/RF3/native preflight270/cohort/site original source/run/attempt/upload/digest/report evidence. Failed/cancelled/skipped/unavailable data remains honest.|
-|REQ-NGR-005 settled resource-failure observation|AC-NGR-005 / TASK-NGR-E1: Authored OutboxFailureDiagnosticClassifierTests and OutboxFailureDiagnosticFormatterTests map exact eligibility and typed privacy/bounds; execution is pending. Existing real isolated1/2/3 failed-case runner/worker originals must prove the one authenticated SDK read outside timing, cancellation/settlement and original-result preservation. No quotas, purge or retry changes.|
+|REQ-NGR-005 settled resource-failure observation|AC-NGR-005 / TASK-NGR-E1: thirteen genuine classifier/formatter cases pass in each3ae unit mode; the7d1196 n1 delete original proves a numeric native failure observation. Complete1/2/3 failure-path and cancellation/settlement gates remain open. Preserve the authenticated SDK read outside timing and original result. No quotas, purge or retry changes.|
 
 ```mermaid
 flowchart LR
@@ -120,3 +120,68 @@ numeric budget cause and the complete isolated cohort remain open. Automatic
 Benchmarks37129421599 cancelled all27 preflight and270 scenario jobs without
 steps, so it supplies no native measurement or website refresh. Later source
 changes and working-tree SQL BETWEEN are outside this exact3ae qualification.
+
+TASK-NGR-K2R / REQ-NGR-003 is read-only source diagnosis of the genuine7d9852
+KurrentDB3node preflight failure in Benchmarks37130907091/attempt1/job111227060126.
+Root independently verifies two original ZIPs and0/1 TUnit error; all five worker
+repetitions have null measurements and zero samples. First setup reports
+NoStreamSemantics/NotLeaderException; remaining repetitions record
+PreviousSetupFailure. Teardown retains primaryFailure=true and no failed stages.
+The source/run/upload/report/worker evidence is retained under seal
+c50787a2279f9a8525df8794aaafa786e5a2cb2f7dd37bac623e36f524dd0a54;
+root verification receiptf12f7c4f4e4fd068e6b78fb4c6d13aa77d0f12976f7c6fedc525a157a2d6f1cb.
+Bounded analysis worker reads the Kurrent client/setup/verifier/endpoint and
+Aspire topology paths, compares actual7d to current owned bytes, and returns
+precise causal evidence or an explicit uncertainty with candidate file ownership
+and real fault/test requirements. No implementation, retry/quorum/timeout change,
+package, runtime, Git mutation or provider control is authorized by this diagnostic task.
+Root owns the contract/review/qualification join before any repair. This failed
+native gate cannot qualify the complete cohort or website metrics.
+
+K2R source review is complete: eight inspected consumer/topology files match the
+failed revision; server logs show convergence near setup, but the sanitized
+original omits RPC status, selected destination and redirect details. Cause
+remains unknown, with no proven consumer or pinned-SDK defect. No implementation
+is justified by that packet. Root reviewed analysis SHA-256
+`da2be25c51d66e5b242554ced508c58715732a94fc7ae74888b4439c8ed741de`.
+[The original7d985 receipt](../../implementation/sql-client-qualification-7d985.json)
+retains the green CI baseline and cancelled native cohort:26/27 preflight jobs
+have provider success, one genuine Kurrent setup failure;19/270 scenario jobs
+have provider success and251 are cancelled. These metadata counts do not prove
+complete case or measurement success. Nineteen cancelled jobs have failed result
+upload steps; no workload failure is inferred from upload failure. Full cohort,
+scale and website qualification remain open.
+
+The successor [7d1196 n2 preflight original](../../implementation/native-kurrent-preflight-failure-7d1196.json)
+also fails NoStreamSemantics with NotLeaderException and no measurements. Root
+independently verifies both original ZIPs; the repeated stage/type does not
+establish a shared cause. [Pinned SDK source review](../../implementation/kurrent-sdk-routing-source-review.md)
+confirms supported multi-seed settings and future-call route updates while the
+original failed append is surfaced. Selected destination/status/leader snapshot
+remain absent; no repair or replay is approved from this source review.
+
+The [7d1196 KeyLoad failure originals](../../implementation/native-keyload-failures-7d1196.json)
+retain three later genuine failures from the same live run. The n1 delete's first
+four repetitions each complete10k operations; final preparation rejects with
+ResourceExhausted and the stageE SDK observation records100000 retained outbox
+entries and192581709 bytes. This reaches the exact default100000-record cap,
+below the1GiB byte cap. It qualifies that native numeric failure observation;
+fixture accounting, reclamation and a repair remain separate unresolved work.
+The n2 update completes its first10k operations, then four repetitions fail
+OwnershipLost before timed sample collection; warmup versus preparation and the
+underlying cause are absent. The n2 PointRead completes all50k
+samples, then its separate native fault check fails quorumAssertions. The harness
+erases the original exception and assigns receipt fields only after all SDK/MCP
+probes return: null/false defaults do not prove which probe failed or that MCP ran.
+Container restoration is observed, while logical fault recovery remains failed.
+Root independently verifies all six original ZIPs, provider/upload/source bindings,
+reports and raw case hashes. The live snapshot is bounded by its recorded time;
+later failures need fresh review. No new passing cohort, quota/retry change,
+required-scale result or website refresh is established.
+
+[Exact delete source accounting](../../implementation/keyload-outbox-benchmark-source-review.md)
+predicts95620 entries before fifth measured-input preparation and115620 for the
+unchanged complete fixture. The observed100000 count matches the guard; hidden
+ordinal and inner cause remain unexported. Server defaults have no current
+benchmark limit-injection seam. Any new setting requires a separate replicated
+resource/provenance/restart contract, not a change to stageE observation.

docs/Features/QueryExecution.md

@@ -4,8 +4,10 @@ REQ-SQLC-006 / AC-SQLC-006A / TASK-SQLC-BETWEEN adds the bounded typed
 [SqlBetween stage](QueryExecution/SqlBetween.md) under Accepted ADR065. Existing
 scalar/null/eager-error semantics, expanded AST budgets, field authority and
 cursors remain canonical. First-authored real ZoneTree and genuine RF3 SDK/MCP
-tests must qualify the new syntax; full SQL/native protocol and measured
-performance remain unfinished.
+tests qualify the new syntax at source7d1196 with44 original execution rows;
+[the receipt](../implementation/sql-client-qualification-7d1196.json) retains
+full exact-source CI. Coverage, full SQL/native protocol and measured performance
+remain unfinished.
 
 REQ-SQLC-003 / AC-SQLC-003P adds independent full public/native value and
 same-instance native writer tests under the Accepted

docs/Features/QueryExecution/SqlBetween.md

@@ -1,8 +1,8 @@
 # QueryExecution bounded SQL BETWEEN
 
 Status: Accepted contract under [ADR-065](../../ADR/ADR-065-full-sql-client-compatibility.md);
-source implementation and development checks complete; exact-SHA runtime proof
-pending. Existing Q1 semantics
+source implementation, development checks and original GitHub runtime proof
+complete at7d1196. Coverage and performance gates remain open. Existing Q1 semantics
 remain the target. Full PostgreSQL syntax/semantics and native client sessions
 are separate mandatory unfinished stages.
 
@@ -121,12 +121,25 @@ joined all eleven owned source files without a source blocker. Initial frozen
 7d9852b source build failed with five IDE0005 unnecessary using directives in
 new unit fixtures,0warnings,elapsed00:01:39.70. The owning worker removes only
 those imports; no assertion, diagnostic severity or execution contract changes.
-Repeat full Release, formatter and static checks before delivery. Runtime still
-requires the new delivered-source GitHub originals.
+Full Release, formatter and static checks passed before delivery. Original
+runtime proof is retained below separately from those development checks.
 
 The [source-stage receipt](../../implementation/sql-client-source-stage-003.json)
 binds all eleven owned files to frozen7d985 base, full26project Release with
 0warnings/0errors, canonical formatter and static26project/4module governance.
 The final import/whitespace corrections preserve all assertions and non-whitespace
-bytes. The21unit and2RF3 cases still require authentic delivered-source CI proof.
-No runtime, coverage or measured gain is inferred from development checks.
+bytes. No runtime, coverage or measured gain is inferred from development checks.
+
+[Original CI proof](../../implementation/sql-client-qualification-7d1196.json)
+now closes the runtime gate at source7d1196/run37132491153/attempt1: all21 new
+unit cases pass in each normal/scalar mode, and both RF3 SDK/official MCP cases
+pass. Root independently verifies original ZIP/report/source/job/upload hashes,
+all44 execution rows and the same21 unit IDs in both modes; all11 owned source
+files match the frozen development source. Full CI passes2667/2667 normal and
+scalar,194/194 recovery,69/69 RF3 and118/118 unique analyzer cases, with no
+nonpass, Release warnings/errors, formatter or governance failure. This source
+includes9 additional governance tests and is a descendant of deliveredbe2e;
+its cancelled predecessor supplied no executed cases.
+Coverage collector/thresholds, complete native comparisons, full SQL and native
+sessions remain open. The benchmark run has a genuine Kurrentn2 setup failure;
+it cannot qualify scale or website performance. ADR065 remains Accepted.

docs/implementation/keyload-outbox-benchmark-source-review.md

@@ -0,0 +1,54 @@
+# KeyLoad delete fixture outbox accounting
+
+TASK-NGR-D1R / REQ-NGR-005 / AC-NGR-005 is read-only source diagnosis.
+[The genuine failed job](https://github.com/managedcode/KeyLoad/actions/runs/37132491118/job/111232263528)
+uses source `7d1196db51f682f64ffcf4560e7b3f12b49b499b`; its
+[original receipt](native-keyload-failures-7d1196.json) retains provider/upload,
+ZIP/report/raw hashes, all case counts and the redacted numeric observation.
+No implementation, quota, retention or measurement change is delivered here.
+
+The observed head has100000 retained entries,192581709 bytes and no active
+consumer. The exact source's default bounds are100000 entries and1073741824
+bytes. [AppendOutbox](https://github.com/managedcode/KeyLoad/blob/7d1196db51f682f64ffcf4560e7b3f12b49b499b/src/KeyLoad.Core/ProjectionOutbox.cs)
+rejects the next ordinary append when the retained count reaches its cap.
+Only projection progress uses the separate reserve. This supports record-cap
+exhaustion strongly; the original inner exception and rejected ordinal were not
+exported and remain unproven runtime details.
+
+The exact4096-document,10000-operation,256-warmup,five-repetition,16-worker
+source yields the following independent entry accounting:
+
+|Source work|Outbox entries|
+|---|---:|
+|4096 document puts and4096 vector puts|8192|
+|256 graph vertices with3 outgoing edges|768|
+|4096 event appends|4096|
+|Successful mutation-probe effects|4|
+|Each completed delete repetition:256 preparation puts,256 warmup deletes,10000 measured-input puts and10000 measured deletes|20512|
+|Initial seed/probe plus four completed repetitions|95108|
+|Fifth repetition's preparation/warmup before measured-input puts|512|
+|Head predicted before fifth measured-input preparation|95620|
+|Complete unchanged five-repetition fixture requirement|115620|
+
+The source therefore predicts4380 successful preparation appends to reach100000,
+then rejection before timed samples are allocated. Four original repetitions
+have10000 successes/zero failures/10000 samples each; the fifth has
+ResourceExhausted/null measurement/zero samples, consistent with that prediction.
+This arithmetic does not prove the hidden runtime ordinal. Source review compares
+18 inspected owning files byte-for-byte with the failed commit and current tree.
+Analysis SHA-256 `2c21daf4fde8dfc5a26fe621760e5f112b8e75103b8fc6b5f1c45dc27cf0c7bc`;
+source/evidence inventory SHA-256
+`9cbce0b5d23dd1cfdb4cccb249c5abfe7eeccf467a603fd94376e6a33069c496`.
+
+The actual server PartitionHost constructs DatabaseEngine with default limits;
+NodeOptions and isolated AppHost currently expose no database-limit seam.
+Direct Core constructor injection does not configure the real server fixture.
+The115620-record minimum is a requirement for this fixture;120000 is only a
+bounded candidate, with the byte cap unchanged. A separate implementation
+contract must freeze effective limits on every replica, provenance, restart
+compatibility, admission, regression boundaries and actual default/candidate
+native proof before introducing a server setting. ADR-068 stageE authorizes
+observation only. No automatic quota growth, history purge or write retry follows
+from this review. The mandatory100k/1m/5m corpus and100k-operation cells require
+their own checked capacity and retained-byte accounting; this small control
+fixture cannot qualify those scales, the complete cohort or website metrics.

docs/implementation/kurrent-sdk-routing-source-review.md

@@ -0,0 +1,52 @@
+# Pinned Kurrent SDK routing source review
+
+TASK-NGR-K3R / REQ-NGR-003, read-only source evidence under Accepted ADR-068.
+NuGet metadata binds KurrentDB.Client1.4.0 to official commit
+`e041cf8f459d6abe05f39918ecd62a71038b9ddc`. This source review does not prove
+the destination or cause of an original failed call, or an SDK defect.
+
+The pinned [connection parser](https://github.com/kurrent-io/KurrentDB-Client-Dotnet/blob/e041cf8f459d6abe05f39918ecd62a71038b9ddc/src/KurrentDB.Client/Core/KurrentDBClientSettings.ConnectionString.cs)
+accepts the configured multi-host esdb scheme and leader preference.
+[GossipChannelSelector](https://github.com/kurrent-io/KurrentDB-Client-Dotnet/blob/e041cf8f459d6abe05f39918ecd62a71038b9ddc/src/KurrentDB.Client/Core/GossipChannelSelector.cs)
+uses the first successful gossip response; the
+[node selector](https://github.com/kurrent-io/KurrentDB-Client-Dotnet/blob/e041cf8f459d6abe05f39918ecd62a71038b9ddc/src/KurrentDB.Client/Core/NodeSelector.cs)
+ranks available members. Leader preference can fall back to a follower when
+that response contains no usable leader. This is a possible mechanism, not
+evidence that it caused either preserved KeyLoad failure.
+
+[TypedExceptionInterceptor](https://github.com/kurrent-io/KurrentDB-Client-Dotnet/blob/e041cf8f459d6abe05f39918ecd62a71038b9ddc/src/KurrentDB.Client/Core/Interceptors/TypedExceptionInterceptor.cs)
+maps the NotLeader trailer key and leader-host/port trailers to the typed
+exception with its original RpcException. Its actual path does not invoke the
+separate Google RPC detail parser in NotLeaderException.FromRpcException.
+The exception exposes a public DnsEndPoint LeaderEndpoint.
+[ReportLeaderInterceptor](https://github.com/kurrent-io/KurrentDB-Client-Dotnet/blob/e041cf8f459d6abe05f39918ecd62a71038b9ddc/src/KurrentDB.Client/Core/Interceptors/ReportLeaderInterceptor.cs)
+reports that endpoint for future channel selection; raw Unavailable errors
+instead request rediscovery. The
+[batch append receiver](https://github.com/kurrent-io/KurrentDB-Client-Dotnet/blob/e041cf8f459d6abe05f39918ecd62a71038b9ddc/src/KurrentDB.Client/Streams/KurrentDBClient.Append.cs)
+faults the original pending append and replaces its appender for later calls.
+No transparent replay of that same failed append was found.
+
+The [7d985 n3 original](sql-client-qualification-7d985.json) and
+[7d1196 n2 original](native-kurrent-preflight-failure-7d1196.json) both retain
+NoStreamSemantics/NotLeaderException, all five null measurements and clean
+teardown. Identical stage/type does not prove identical causes. Their sanitized
+exports omit the selected destination, reported LeaderEndpoint value, gRPC
+status code and SDK's own selected gossip snapshot. Cause remains unknown;
+no consumer workaround, SDK repair or routing change is justified by this review.
+
+The next real native proof must correlate the original selected/receiving node,
+reported leader and status-code-only diagnostic with contemporaneous all-node
+roles. Any added capture needs a bounded redacted contract before implementation.
+Preserve the failed original; success still requires unchanged NoStream ACK,
+expected conflict and exact original-event readback on actual1/2/3 members.
+Do not export messages, raw trailers, credentials, endpoints or user payloads,
+or retry the failed append to make its qualification green.
+
+Research report SHA-256
+`49b3fe94a528e942db05fbcef6c75a82ac40649f9d1b203f6c9f052a8f4a73e3`;
+source inventory SHA-256
+`890d18e3f8ae1119afc4401d7da0ae4ceec62ffa552822f179683b1e4a686d51`.
+The inventory records immutable Git SHA-1 blob IDs, not a claimed local SHA-256
+of retrieved source. There was no package installation, runtime, native SDK
+execution or repository mutation by the research worker. This record is manual
+architecture evidence and cannot qualify performance or the complete cohort.