Retain machine-key argument-role regression baseline

Preserve all required tests and coverage gates. Fix native display parsing and numeric test spans/cohesion; add meaningful real semantic regressions before the separately reviewed production repair. This is a tests-first source cut, not completed qualification.

Author: KSemenenko

docs/ADR/ADR-033-code-quality.md

@@ -506,3 +506,130 @@ per file and explicit aggregate covered-line budgets, with independent exact cou
 exit and critical-pipeline assertions. It must not pad production sources, copy
 invented execution into native evidence, or make absence of a threshold label
 alone count as a successful parse. These bounded corrections precede TASK011 join.
+
+### Observed native candidate failure implementation contract
+
+Run36988549282 atb402dc50785028cc5b09a3928c5b28ba8a721a6e preserves the first real
+AnalyzerTests baseline:88executed/84passed/4failed/0skipped and native Cobertura
+SHA256983f7de7132938062db791723cdfba1ca683eceb98b3cecb191b2452af938091. Related
+REQ/AC-BC-027, REQ-CQ-006 and AC-CQ-008/009 keep all original thresholds/sources.
+
+1. Strongest lead freezes the observed failure acceptance and task graph before
+   writing workers. Read-only native grammar/numeric span/cohesion discovery is
+   complete; the missing native report or test failure never becomes a pass.
+2. TASK-SITE-NATIVE-FORMAT-014 owns only
+   `scripts/Features/CodeQuality/site-analyzer-coverage.shared.ps1` and new
+   `tests/KeyLoad.Analyzers.Tests/Features/CodeQuality/SiteAnalyzerCoverageBranchDisplayTests.cs`.
+   Author real-PowerShell TUnit native decimal/display-rounding/malformed cases
+   first; change only the full-token display regex. Observed0..100 integer or
+   one/two decimal spellings are valid; covered/valid groups stay integer-only
+   and retain all bounds/overflow/denominator checks. Rounded display never
+   controls a threshold. No other parser, contract or configuration change.
+3. TASK-SITE-NUMERIC-REPAIR-015 owns existing ControlFlowNestingAnalyzerTests.cs
+   and ExecutableUnitCodeLineCountAnalyzerTests.cs, and new
+   ControlFlowNestingExtendedConstructTests.cs and
+   ExecutableUnitSpanAndTriviaAnalyzerTests.cs in the same test slice. Preserve
+   all registered cases/assertions; move cohesive construct/trivia scenarios into
+   separate nonpartial types under200code lines. Expected method spans use the
+   known declaration token's source index, preserving all location/end assertions.
+   NumericAnalyzerFixture, production analyzers and self-inventory stay unchanged.
+4. TASK-SITE-DIAGNOSTIC-FLOWS-016 follows a strongest reviewed source/case map
+   before implementation. It owns only explicitly new CodeQuality test files;
+   no014/015/shared source overlaps. It exercises real analyzers through SDK
+   Roslyn/BCL/Orleans inputs and diagnostic/no-diagnostic assertions for actual
+   KLD0001/KLD0022 uncovered flows. Source denominators remain frozen; no direct
+   helper calls, stubs, padding, reclassification or threshold alteration.
+5. Root alone integrates docs, project/workflow/config/candidate ownership. Join
+   every complete source/method-registration/hash packet; development Release
+   build and scoped formatter precede GitHub qualification. Compose an ordinary
+   descendant using a reviewed separate index; preserve the shared dirty checkout.
+   Run complete AnalyzerTests/native collection/gate and complete SiteTests at
+   the exact candidate SHA, retain raw native/TRX/PDB/SARIF/run/job/hash receipts,
+   then require strongest joined final review. No local tests or qualifications.
+
+Stages014/015 are safe parallel writes after frozen approval;016 waits for its
+read-only case map and an available capable economical worker. Workers must end
+complete/blocked/failed/cancelled with exact diff/hashes and preserved case maps;
+only reviewed complete packets unblock integration. Stop on ambiguous semantics,
+actual analyzer defects, unsupported APIs, overlapping ownership or new contracts.
+Rollback restores the coherent candidate tooling/test unit while keeping mandatory
+policy and failed artifacts. Runtime/data/public migrations are N/A. ADR remains
+Accepted until required exact-SHA evidence and every acceptance criterion pass.
+
+TASK016's exact new-file ownership is now frozen under
+`tests/KeyLoad.Analyzers.Tests/Features/CodeQuality/`:
+LiteralMachineKeyInvocationTests.cs (actual keyed invocation and nested-literal
+flows), LiteralMachineKeyConstructionTests.cs (actual key-value construction and
+initializer flows), LiteralMachineKeyAttributeContractTests.cs (native qualified
+attribute key contracts), and SystemClockSemanticTests.cs (real clock/nonclock
+property symbols). Its precise positive/negative/edge map is in site acceptance.
+Use existing AnalyzerFixture compilation and public analyzer entry points only;
+each positive proves exact count/ID/severity/path/known token location, negatives
+prove no finding. No shared fixture changes or artificial framework types.
+Known potential reordered named-value/multiargument WriteString false positives
+must be retained as regressions and escalated, never copied into a passing oracle.
+Any production repair requires a separate explicit strongest-reviewed contract
+before writes. Root queues this task on the economical014 worker after its complete
+join, while015 retains disjoint numeric-test ownership; no blocked dependency is
+treated as complete. Native90% pipeline proof remains required after all joins.
+
+TASK015's strongest source join found a latent executable-unit violation in
+ElseIfEveryLoopUsingStatementAndFixedAddNestingAsync: its multiline literal is
+part of the unit's token line span, exceeding50 even after aggregate-type splits.
+The same worker may edit only ControlFlowNestingExtendedConstructTests.cs to
+extract that literal into a named private const in the same class. Preserve exact
+decoded compiler-input bytes, all assertions and the sole original registration.
+No counter/threshold/source exclusion or test case changes. Root verifies the
+raw-string byte-equivalence and refreshed hash before enabled build/format/CI;
+this source-only remedy cannot close the GitHub self-inventory qualification.
+
+### Accepted machine-key argument-role repair contract
+
+TASK016 stopped with a real compiler-input regression for Utf8JsonWriter.WriteString:
+the existing nested AlwaysKey helper classifies both property-name and human value.
+Related REQ-CQ-002/006, AC-CQ-004/008/009 and REQ/AC-BC-027 require this test-first
+TASK-SITE-ARGUMENT-ROLE-018 repair, not an oracle accepting the false positive.
+
+1. Root freezes the detailed semantic acceptance and explicit task graph. The
+   economical capable worker authors only new LiteralMachineKeyArgumentBindingTests.cs
+   and LiteralMachineKeyCreationBindingTests.cs under the CodeQuality test slice.
+   Keep016's retained Invocation regression unchanged and blocked. Verify actual
+   compiler-valid SDK/BCL/ASP.NET inputs and independent exact literal spans for
+   positional/reordered/nested key versus value, key-value construction, dynamic
+   fallback, reduced/static extensions and expanded params.
+2. Root joins those sources plus completed014/015, enabled build/format and a
+   reviewed ordinary descendant temporary-index source cut. Run complete real
+   AnalyzerTests/native collection in GitHub with production unchanged. Retain
+   exact failed assertions/TRX/XML/source/PDB/run/job/artifact hashes. This red
+   baseline cannot unblock017/final acceptance; no test or branch exit is ignored.
+3. After reviewed GitHub regression evidence, the same worker may edit only
+   src/KeyLoad.Analyzers/Features/CodeQuality/MachineKeyLiteralClassifier.cs and
+   MachineKeySemanticSymbols.cs. Resolve named arguments by actual parameter
+   identity; positional and expanded params by bound parameter. Explicit key
+   names qualify; current method/container/constructor heuristics qualify only
+   the first logical parameter, skipping an unreduced extension receiver.
+   Reduced instance extension calls already omit that receiver. A resolved value
+   parameter returns false before any textual-position heuristic.
+4. Apply one shared decision to nested AlwaysKey literals, ordinary invocation
+   fallback and known key-value construction. Preserve nearest argument context.
+   Unresolved/dynamic invocations retain recognized named-key or unnamed argument0
+   method-name fallback; unresolved construction stays non-key. Keep catalogs,
+   analyzer entry point/ID/severity, attributes/indexers/initializers/generated/
+   empty-string behavior and all25 executable paths/12 pipelines unchanged.
+5. Root/strongest join every source diff/case/hash and mandatory numeric limits,
+   then resume016's exact semantic scope. Enabled build/format/governance precede
+   complete exact-SHA GitHub analyzer/native/site/browser qualification. Freeze
+   new source hashes and use actual new native denominators for80/70/90; historical
+   native counts cannot qualify changed source. Final017/TASK007 waits for every
+   completed source stage and all required runtime evidence.
+
+Worker018 owns its two new tests and later two production files only;014/015/016,
+shared AnalyzerFixture, catalog, contracts/config/packages/policy/docs stay other
+owned. Root exclusively owns shared integration/candidate delivery. Completion
+requires complete/blocked/failed/cancelled stage packets; stop on unsupported SDK
+metadata, ambiguous bindings or changed diagnostic contract. No local tests,
+stubs, direct helper coverage, suppression, exclusions or weakened thresholds.
+This changes compiler diagnostic classification only, with no public runtime,
+data or deployment migration. Rollback restores the coherent analyzer plus
+regressions under unchanged mandatory policy and retains all red evidence.
+ADR remains Accepted until actual joined implementation and verification pass.

docs/Features/CodeQuality.md

@@ -33,6 +33,24 @@ sources, exact integer80/70/90 thresholds and complete diagnostic regressions.
 Successful evidence establishes only the first analyzer-module baseline; global
 AC-CQ-009 and RF3/container/no-decrease coverage remain pending.
 
+Observed candidate run36988549282 executes88 regressions with84 successes and4
+failures, retained native XML/TRX and no skipped analyzer case. Its next bounded
+failure loop maps BC-027/CQ-006/CQ-008/009 to TASK-SITE-NATIVE-FORMAT-014 (native
+decimal-display grammar with integer-only decisions), TASK-SITE-NUMERIC-REPAIR-015
+(independent token-start span oracles and cohesive200-line test classes), and
+TASK-SITE-DIAGNOSTIC-FLOWS-016 (meaningful KLD0001/KLD0022 flows for real coverage
+gaps). Acceptance, exact disjoint ownership, staged join and GitHub verification
+are in the site acceptance/plan and ADR-033 below. No native numeric pass or
+broader CodeQuality completion follows from source repairs or local builds.
+
+TASK016's genuine WriteString regression exposed a value-argument false positive.
+Accepted TASK-SITE-ARGUMENT-ROLE-018 preserves that failing case, adds exact real
+argument-binding regressions and first runs an unchanged-production GitHub red
+baseline. The later two-file bound-parameter classification repair and016 semantic
+join remain mandatory before complete native/site qualification. See the detailed
+site acceptance/plan and ADR-033 implementation contract; blocked source never
+unblocks final acceptance.
+
 ## Canonical slice map
 
 - Backend/tooling: `src/KeyLoad.Analyzers/Features/CodeQuality/`.

docs/implementation/site-design.json

@@ -59,7 +59,11 @@
     "coverage": "TASK013_COMPLETE_strongest_source_approval_ae21793f1c58424a507dbeb2b8bdf4b227cd84821f4034b7d576966d6e5ad352; integrated_development_build_passed; real_GitHub_native_numeric_qualification_pending",
     "browser": "TASK012_COMPLETE_strongest_source_approval_7424fb8d25f303c3c94ca95870de76431de473e5317a76654e028a8b23bf1082; preserved_otherchat_KL-SITE-BUILD-003A_handoff; integrated_build_and_format_passed",
     "candidateAudit": "TASK-SITE-CANDIDATE-AUDIT-010_COMPLETE; repair source-preparation review COMPLETE, all required joins still pending",
-    "analyzerCoverage": "TASK011_COMPLETE_strongest_source_approval; native_boolean_BigInteger_checked_sum_physical_line_gate_and_complete_12_pipeline_oracles_joined; nesting3; integrated_build_and_format_passed"
+    "analyzerCoverage": "TASK011_COMPLETE_strongest_source_approval; native_boolean_BigInteger_checked_sum_physical_line_gate_and_complete_12_pipeline_oracles_joined; nesting3; integrated_build_and_format_passed",
+    "nativeFormatRepair": "TASK014_COMPLETE_root_and_strongest_source_approval; native_decimal_TUnit_execution_pending",
+    "numericRepair": "TASK015_COMPLETE_root_and_strongest_source_approval;17registrations_preserved;699byte_fixture_d402a5f38f1eb28daa90714c92ec45906c833ea7eef936893fd44875c1eb3f01;21unitlines; runtime_selfinventory_pending",
+    "diagnosticFlows": "TASK016_BLOCKED_on_real_WriteString_value_false_positive; retained_meaningful_regression; no_production_edits",
+    "argumentRoleRepair": "TASK018_RED_source_COMPLETE_11cases_plus_retained016regression; integrated_build0warnings0errors; strongest_source_and_red_candidate_review_pending; production_unchanged"
   },
   "verification": {
     "source": "All_bounded_worker_packets_complete_joined_full_root_and_strongest_review; 51_source_manifest_98c457ee0333c816795665f742101989b9a0638b2382372313afefd823de4617; source_only_not_runtime_qualification",
@@ -106,7 +110,7 @@
       ]
     },
     "githubSiteSuite": {
-      "state": "pending_descendant_retry_after_pretest_workflow_failure",
+      "state": "test_first_argument_role_red_baseline_pending_before_scoped_production_repair",
       "attempts": [
         {
           "run": 36987695573,
@@ -119,7 +123,40 @@
           "siteTests": "skipped_not_passing",
           "coverage": "missing_native_report_failed_closed",
           "publication": "skipped",
-          "artifact": "site-qualification-f56d511b4a343eff99d3117e9414490a05cfb8ae"
+          "artifact": "site-qualification-f56d511b4a343eff99d3117e9414490a05cfb8ae",
+          "artifactReceipt": {
+            "id": 11218301726,
+            "digest": "sha256:be2cc09ec751e64ff51fe257ea3c91b8a90ef9331f8a7c78300fd0dfdf66ac9a",
+            "localEvidence": "/private/tmp/keyload-site-run-36987695573",
+            "log": "/private/tmp/keyload-site-run-36987695573/workflow.log"
+          }
+        },
+        {
+          "run": 36988549282,
+          "sourceRevision": "b402dc50785028cc5b09a3928c5b28ba8a721a6e",
+          "url": "https://github.com/managedcode/KeyLoad/actions/runs/36988549282",
+          "jobUrl": "https://github.com/managedcode/KeyLoad/actions/runs/36988549282/job/110779101338",
+          "conclusion": "failure",
+          "state": "completed_failure",
+          "runtimeSourceReceipt": "passed",
+          "publication": "skipped",
+          "analyzerTests": {
+            "total": 88,
+            "executed": 88,
+            "passed": 84,
+            "failed": 4,
+            "skipped": 0
+          },
+          "siteTests": "skipped_not_passing",
+          "coverage": "native_report_retained_decimal_display_pair_rejected",
+          "nativeReportSha256": "983f7de7132938062db791723cdfba1ca683eceb98b3cecb191b2452af938091",
+          "artifactReceipt": {
+            "id": 11217729758,
+            "name": "site-qualification-b402dc50785028cc5b09a3928c5b28ba8a721a6e",
+            "digest": "sha256:9ed687f54b29ba15de64653993cb5cb2b93227eb64a8295621eeffcddc6a3640",
+            "localEvidence": "/private/tmp/keyload-site-run-36988549282",
+            "log": "/private/tmp/keyload-site-run-36988549282/workflow.log"
+          }
         }
       ]
     },
@@ -149,15 +186,39 @@
         "kind": "raw_compiler_SARIF; includes suppressed/generated diagnostics; terminal error count is separate",
         "terminalBuildErrors": 3
       }
-    ]
+    ],
+    "analyzerNativeAnalysis": {
+      "state": "strongest_independent_diagnostic_analysis_only_not_gate_pass",
+      "run": 36988549282,
+      "nativeReportSha256": "983f7de7132938062db791723cdfba1ca683eceb98b3cecb191b2452af938091",
+      "unionLinesCovered": 841,
+      "unionLinesValid": 942,
+      "branchesCovered": 399,
+      "branchesValid": 570,
+      "belowCriticalThreshold": [
+        {
+          "diagnostic": "KLD0001",
+          "covered": 163,
+          "valid": 224
+        },
+        {
+          "diagnostic": "KLD0022",
+          "covered": 25,
+          "valid": 30
+        }
+      ],
+      "otherTenPipelines": "arithmetic_at_least90percent; genuine_gate_parser_failed"
+    }
   },
   "delivery": {
-    "candidate": "validation_branch_created_first_attempt_failed_before_tests_retry_pending",
+    "candidate": "second_validation_failed_analyzer_tests_and_native_format_repair_pending",
     "branch": "codex/site-qualification-20261002-a",
     "initialCandidateRevision": "f56d511b4a343eff99d3117e9414490a05cfb8ae",
     "qualifiedRevision": null,
     "sourceSnapshotReceiptSha256": "271d8415d29c69e8af84f604858079ed6effd4ce586d06220237ff9ac57b9684",
     "workspace": "shared_dirty_checkout_preserved",
-    "main": "not_changed"
+    "main": "not_changed",
+    "currentCandidateRevision": "b402dc50785028cc5b09a3928c5b28ba8a721a6e",
+    "descendantReceipt": "/private/tmp/keyload-site-candidate-20261002-b.delivery.json"
   }
 }

scripts/Features/CodeQuality/site-analyzer-coverage.shared.ps1

@@ -51,7 +51,7 @@ $script:CoverageTokens = [ordered]@{
     XmlBranchTrue = 'true'
     XmlBranchFalse = 'false'
     XmlConditionCoverage = 'condition-coverage'
-    BranchCoveragePattern = '\A[0-9]+% \(([0-9]+)/([0-9]+)\)\z'
+    BranchCoveragePattern = '\A(?:100(?:\.0{1,2})?|(?:0|[1-9][0-9]?)(?:\.[0-9]{1,2})?)% \(([0-9]+)/([0-9]+)\)\z'
     XmlPackageSelector = './packages/package'
     XmlSourceSelector = './sources/source'
     XmlClassSelector = './classes/class'