test(dry-generator): consolidate V2_6_V2_7 into topic-first layout

[ottobolyos]

Summary

Executes the DRY-generator campaign in full on a single branch. Phase 1
(topic-first fixture consolidation) and Phase 2 (three permanent
reflection / source-scan guards) land the topology rewrite for
tests/MTConnect.NET-Common-Tests/; Phase 3 (Scriban template
consolidation via Shape A / B) lands two byte-identity-preserving
mergers on build/MTConnect.NET-SysML-Import/ plus a new
tests/MTConnect.NET-Generator-Tests/ project carrying the current-XMI
byte-identical regen guard; Phase 4 (delta-driven emitter +
generator-fidelity audit) closes the pre-existing baseline drift and
lands the --previous-xmi CLI plus the mutated-XMI cross-verify;
Phase 5 (verification pass) lands as further commits on this branch.

Scope decision (ottobolyos-approved 2026-08-20): a single PR keeps the
convention rewrite, the template consolidation, and the delta emitter
reviewable as one coherent unit. The earlier "Phases 3 – 5 ship as
separate campaign PRs" plan is superseded.

Motivation

Two intertwined pressures drive this campaign:

• Per-version-folder rot. The old V<major>_<minor>/ fixture
  convention forces every spec-version bump to author a whole new
  folder of fixtures. The topic-first convention lets the next bump ADD
  a [TestCase(MTConnectVersions.VersionXY)] line and be done.
• Per-version-template rot. The Scriban template family had two
  latent duplication axes: near-identical partial-class hosts that
  differed only in class name and doc summary, and paired
  descriptions-class templates that differed only in the Get(...)
  overload's parameter type. Every future version bump the campaign
  ships would double the maintenance surface unless the templates are
  collapsed first.

Without both landing, MTConnect Standard v2.8 support would ship as
another V2_8/ fixture folder plus another set of per-version template
copies — every new spec release compounding the maintenance cost.

Scope: all five phases in this PR

• Phase 1 — topic-first fixture consolidation. Deletes
  V2_6_V2_7/; lands six new topic fixtures at their canonical
  topic-first location.
• Phase 2 — permanent guards. Three complementary reflection /
  source-scan checks keep the new convention machine-enforced.
• Phase 3 — template consolidation via Shape A / B. Two
  byte-identity-preserving mergers land on the Scriban template family;
  a new test project carries the current-XMI byte-identical regen
  guard.
• Phase 4 — delta-driven regen mechanism (--previous-xmi CLI flag
  on the SysML importer) plus the SysML-XMI 1:1 fidelity audit that
  closes the Phase 3.1 baseline drift (15 orphan .g.cs files
  triaged, 63 whitespace-drift files refreshed, byte-identical guard
  flipped from [Explicit] to normal [Test]).
• Phase 5 — verification pass (full-tree cross-check + wire-format
  matrix). Lands on this branch once Phase 4 is stable.

Approach

The design decisions applied are documented as inline choices on each
fixture / template.

Phase 1 / 2 — fixture topology

• D1 — Version-gate mechanism: [TestCaseSource(typeof(MTConnectVersionMatrix), nameof(MTConnectVersionMatrix.All))] on each behavioral method, gated with Assume.That(v >= MTConnectVersions.Version??). Rows below the floor surface as Inconclusive in the test explorer, making "gated out" visually distinct from "ran and passed". MTConnectVersionMatrix.All already ships at tests/MTConnect.NET-Common-Tests/TestHelpers/MTConnectVersionMatrix.cs; the reflection-driven matrix picks up future version constants automatically without per-test edits.
• D5 — Rollout order: single-PR big-bang. At HEAD there is exactly one per-version folder (V2_6_V2_7/); a staged rollout would have no distinguishing value.

The MTConnectVersions constant tests are kept as plain [Test] since they exercise structural invariants of the MTConnectVersions type itself (constant values, distinctness, monotonicity), not per-version behavioral gates.

Phase 3 — template consolidation

• Shape A — templates whose per-call variants differ only in
  surface parameters (class name, doc line) collapse to one template
  with a Scriban comment header naming the version range it applies to.
  No if mtc_version >= … gates inside; every version consumes the
  same emission.
• Shape B — templates whose per-caller variants differ in small
  fragments collapse to one template with {{ if … }} gates around the
  differentiated fragments. Kept below the ~5-gate over-abstraction
  ceiling per the plan's Phase 3.2 rule.

Every consolidation is atomic (its own commit) and verified by the new
Regen_is_deterministic_across_two_invocations guard in
tests/MTConnect.NET-Generator-Tests/. Byte-identity of every emitted
.g.cs file against the committed tree for the consolidated templates
is preserved.

Phase 3.1 finding — pre-existing baseline drift (RESOLVED in Phase 4)

The Phase 3.1 dry-run surfaced pre-existing drift between the
current-XMI regen and the committed libraries/**/*.g.cs tree: 15
hand-authored interface / observation files the current generator no
longer emits, plus 63 files (predominantly trailing-blank-line
whitespace) that differ from the current-XMI regen output. The
strict-baseline guard shipped in Phase 3 marked [Explicit]
documenting the drift; Phase 4 resolves it in full (per-fix breakdown
below) and flips the guard to plain [Test].

Phase 4 — SysML-XMI 1:1 fidelity audit + delta-driven regen

Phase 4 closes Phase 3.1 in two atomic movements plus the delta CLI.

Fidelity audit resolution. The 15 orphaned .g.cs files decompose as:

• 10 Pallet measurement interfaces (IHeightMeasurement,
  ILengthMeasurement, ILoadedHeightMeasurement, ILoadedLengthMeasurement,
  ILoadedSwingMeasurement, ILoadedWeightMeasurement,
  ILoadedWidthMeasurement, ISwingMeasurement, IWeightMeasurement,
  IWidthMeasurement) — the concrete Pallet class emissions
  (HeightMeasurement : Measurement, IHeightMeasurement, etc.) each
  reference an I{Name} interface that MeasurementModel .RenderInterface() returned null for. Fixed by adding a new
  Pallets.MeasurementInterface.scriban template and wiring
  MeasurementModel.RenderInterface() to invoke it. All 10 interfaces
  now emit deterministically from the generator; the previously
  hand-committed files are refreshed to the generator's output
  (single per-file diff: copyright year 2024 → 2025 from the template).
• 5 orphaned files (Assets/Files/IAbstractFile.g.cs,
  Assets/Files/IFileArchetype.g.cs,
  Devices/Configurations/IRelationship.g.cs,
  Observations/Events/NetworkWireless.g.cs,
  Observations/Events/SensorStateDetect.g.cs) — dead code from
  historical renames (superseded by IAbstractFileAsset,
  IFileArchetypeAsset, IConfigurationRelationship, etc.) or
  spurious enum emissions with no consumers. Verified via codebase-wide
  grep that zero .cs files reference any of them. Deleted.

The 63 whitespace-drift .g.cs files are refreshed to current
generator output in a mechanical bulk commit (regen → copy over
committed tree → commit). No template changes; the drift was
accumulated delta from earlier template touches never mirrored into
the committed tree.

A late-stage fidelity finding (fix(generator): drop stale ToolingMeasurement.Code inheritance override) removes an obsolete
MarkInheritedProperties hand-stitched seed that force-added Code
to Assets.CuttingTools.ToolingMeasurement's inherited-name set on
the design assumption that Measurement.g.cs under CuttingTools was
hand-frozen. The runtime now emits that file via the sharedMeasurement
injection (which uses the Pallet Measurement class as base — no Code),
so the override produced a CS0109 build regression under
TreatWarningsAsErrors=true post-refresh. Override dropped, the
default inheritance walk correctly determines ToolingMeasurement.Code
does not hide anything, no new keyword emitted, build green.

After all three commits: byte-identical guard flipped from
[Test, Explicit] to plain [Test] and passes on every default
dotnet test invocation.

Delta CLI (--previous-xmi). Two new opt-in flags on the SysML importer:

• --previous-xmi <path> — the prior spec-version XMI file.
• --compat-version-label <label> — Compat file basename (default
  Previous).

When --previous-xmi is supplied, the generator renders both XMIs to
isolated scratch directories under %TMP%/mtc-sysml-delta-<guid>,
diffs the emitted .g.cs trees at the file level, and emits only:

• ADDED files (present in new tree only) → written to their normal
  libraries/<library>/... location.
• CHANGED files (present in both trees, different bytes) → written
  to their normal location (new tree's version).
• REMOVED files (present in prev tree only) → skipped entirely.
• UNCHANGED files (present in both, identical bytes) →
  concentrated into one Compat/<label>.g.cs per library, with each
  block still carrying its own namespace X { ... } declaration
  (multi-namespace concentration is legal C#). Byte-identity to the
  deterministically-emitted full-tree output preserved per plan D4
  invariant.

The full-tree mode (default when --previous-xmi is omitted) is
preserved bit-for-bit. Delta mode is orthogonal, opt-in, and rolls
back cleanly to full-tree by dropping the flag from the automation
script.

The plan's original historical-XMI-iteration verification is out of
scope (ottobolyos-approved 2026-08-20 — the build/sysml-model
submodule ships a single snapshot per spec bump). Replaced by
DeltaRegenTests.cs which produces a mutated XMI programmatically
(swapping one ownedComment body) and asserts the delta captures
the mutation surgically.

What changed

Phase 1

• tests/MTConnect.NET-Common-Tests/V2_6_V2_7/ deleted (7 files, 606 LOC).
• 6 new topic fixtures land at their canonical topic-first location:
  • MTConnectVersionsTests.cs (project root) — 5 constant-value invariants of the MTConnectVersions type.
  • Devices/Components/ComponentTests.cs — 4 methods (CuttingTorch, Electrode, PinTool, ToolHolder).
  • Devices/DataItems/DataItemTypeTests.cs — 6 v2.6 methods + 1 v2.7 parametric sweep (8 [TestCase] rows × 17 versions cross-multiplied).
  • Devices/Configurations/ConfigurationTests.cs — 16 methods across DataSet base, Axis / Origin / Rotation / Scale / Translation and their *DataSet siblings plus the corresponding Abstract* bases.
  • Observations/SampleObservationTests.cs — WaterHardness sample-envelope round-trip.
  • Enums/EnumArmTests.cs — MediaType.QIF_MBD arm-presence check.

Phase 2

• tests/MTConnect.NET-Common-Tests/DryGenerator/AssertionParityTests.cs — RED-first coverage-parity guard. Holds an inline migration map of the 34 pre-migration method entries and asserts every one has a live post-migration home in the topic layout. GREEN after the topic fixtures land; permanent regression guard.
• tests/MTConnect.NET-Common-Tests/DryGenerator/PerVersionFolderProhibitionTests.cs — permanent topology guard. Three complementary checks fire RED against any V<N_M>/ directory, V<N_M>*Tests.cs file, or V<N_M>*Tests fixture class regrowth. bin//obj/ are excluded so a package-cache artifact matching the pattern cannot trip a false positive; a HistoricalAnchors allowlist accommodates deliberately-pinned fixtures (empty at HEAD).
• tests/MTConnect.NET-Common-Tests/DryGenerator/TopicFixtureCoverageTests.cs — permanent topic-fixture coverage guard. For each of the 23 spec-anchor types migrated out of the deprecated V2_6_V2_7/ folder, a source-scan check asserts the type name appears at least once in its canonical topic-fixture source file. Uses whole-word regex match so shorter anchors (Axis) do not spuriously match longer type names (AbstractAxis, AxisDataSet). Sibling floor + on-disk-topology guards catch anchor-set shrinkage and file-move regressions. Complements AssertionParityTests (method-name resolution) by adding source-text-mention resolution, catching the rename-plus-body-swap regression a name-only guard would miss.
• docs/testing/version-matrix-convention.md (new) — the single-source-of-truth reference for the topic-first single-file-per-topic convention: the rule, how to add a fixture for a new spec version, when to keep plain [Test], how HistoricalAnchors and AssertionParityTests work.
• docs/testing/v2-6.md, docs/testing/v2-7.md, docs/testing.md — repointed every "Pinned test" cell + the "Test classes" narrative at the post-migration home (Devices/DataItems/DataItemTypeTests, Devices/Components/ComponentTests, Devices/Configurations/ConfigurationTests, Observations/SampleObservationTests, Enums/EnumArmTests, MTConnectVersionsTests). Cross-links to version-matrix-convention.md from every entry point so a future v2.8 contributor finds the recipe.
• tests/MTConnect.NET-Common-Tests/Reflection/RegeneratedTypesCoverageTests.cs — two inline comments rewritten to reflect the topic-first layout (previously pointed at the deleted V2_6_V2_7/ folder).

Phase 3

• New test project tests/MTConnect.NET-Generator-Tests/ wired via dotnet sln add. Two guards ship side-by-side:
  • Regen_is_deterministic_across_two_invocations — the everyday [Test]. Regenerates against the current XMI twice into separate scratch directories under .claude/gen-test-out/ (persistent, gitignored path per repo convention) and asserts the two emitted .g.cs trees are byte-identical. Locks in the determinism guarantee every template consolidation relies on.
  • Current_XMI_regen_matches_committed_g_cs_tree — strict-baseline guard. Ships as [Test, Explicit] in Phase 3 and flips to plain [Test] in Phase 4 after the fidelity audit closes the drift.
• Xml Shape-A consolidation — XmlCuttingItem.scriban + XmlCuttingToolLifeCycle.scriban (22 LOC each, differing only in emitted class name and doc summary) collapse to a single XmlMeasurementArrayHost.scriban. Each call site in Xml/TemplateRenderer.cs supplies class_name, summary, and the shared types array. Byte-identity of both libraries/MTConnect.NET-XML/Assets/CuttingTools/XmlCutting{Item,ToolLifeCycle}.g.cs preserved.
• CSharp Shape-B consolidation — EnumDescriptions.scriban + EnumStringDescriptions.scriban (38 LOC each, differing in three fragments: class-doc wording, Get(...) parameter type, Get(...) doc summary) collapse to a single EnumOrStringDescriptions.scriban with three {{ if is_string }} gates. EnumStringModel exposes IsString => true; EnumModel and ObservationModel rely on Scriban's null-is-falsy default. Byte-identity of every emitted *Descriptions.g.cs file preserved.
• Template inventory: 23 → 21 (net −2). One further consolidation candidate (Devices.ComponentType.scriban + Devices.CompositionType.scriban) was scoped out this pass: a merger requires six independent gate points (namespace, base class, kind noun, is_organizer, ctor format, abstract keyword), exceeding the plan's ≤5-gate over-abstraction ceiling without proportionate parameterization surgery.

Phase 4

Six atomic commits close the Phase 3.1 baseline drift and land the
delta CLI:

• fix(generator): emit Pallet measurement interfaces — adds
  build/MTConnect.NET-SysML-Import/CSharp/Templates/Pallets.MeasurementInterface.scriban;
  wires MeasurementModel.RenderInterface() to invoke it. Regens
  the 10 previously hand-committed I{Name}Measurement.g.cs files
  under libraries/MTConnect.NET-Common/Assets/Pallet/.
• chore(generator): remove stale orphan .g.cs files — deletes
  the 5 zero-consumer orphan files (IAbstractFile.g.cs,
  IFileArchetype.g.cs, IRelationship.g.cs, NetworkWireless.g.cs,
  SensorStateDetect.g.cs).
• chore(generator): refresh .g.cs to current generator output —
  regens the committed tree to close the 63-file whitespace-drift set.
  Mechanical bulk refresh; no template changes.
• test(generator): flip current-XMI byte-identical guard from Explicit to Test —
  Current_XMI_regen_matches_committed_g_cs_tree now runs on every
  default dotnet test invocation, blocking any future template or
  committed-tree drift.
• feat(generator): add --previous-xmi flag for delta-driven regen —
  new --previous-xmi <path> + --compat-version-label <label> CLI
  flags on build/MTConnect.NET-SysML-Import/Program.cs. Delta mode
  renders both XMIs to scratch, diffs at file level, and emits ADDED /
  CHANGED to normal paths + UNCHANGED concentrated to Compat/<label>.g.cs
  per library. REMOVED files skipped.
• test(generator): mutated-XMI cross-verify of delta emission —
  tests/MTConnect.NET-Generator-Tests/DeltaRegenTests.cs. Two tests:
  a same-XMI-both-sides guard (asserts every file concentrates into
  Compat) and a mutated-XMI cross-verify (rewrites one XMI
  ownedComment body, asserts the delta surfaces exactly two changed
  files — the enum + its Descriptions — with the mutation marker in
  each, no marker in Compat files).
• fix(generator): drop stale ToolingMeasurement.Code inheritance override —
  removes the obsolete MarkInheritedProperties seed that force-added
  Code to ToolingMeasurement's inherited-name set on a stale design
  assumption. Restores byte-identity + fixes CS0109 build regression
  post-refresh.
• test(generator): skip bin/obj generated files in byte-identity walks —
  filters /bin/ and /obj/ from the .g.cs walk in both fixtures
  so MSBuild-generated intermediates (GlobalUsings.g.cs, etc.) don't
  spuriously flip the guards RED on already-built hosts.

Migration script

The migration is driven by a repository-conventional migration script (gitignored, not part of the shipped tree). The script validates that every hand-authored topic file is in place, emits renames.tsv and citations.tsv audit artifacts, git rms the deprecated folder, and runs a post-migration sanity build. Topic-file authoring itself remains a hand step: version-gate mechanism selection and spec-source citation verification are hand-checked, not scriptable.

Parity guarantee

Three permanent guards cover the fixture migration:

• AssertionParityTests enumerates every [Test] / [TestCase] / [TestCaseSource] method in the test assembly (excluding the deprecated V2_6_V2_7 namespace) and asserts every entry in the 34-entry migration map resolves to a live method. Two rename conventions apply:

  • strip _in_v2_6 / _in_v2_7 suffixes — the version is now a property of the [TestCaseSource(MTConnectVersionMatrix.All)] matrix rather than encoded in the method name;
  • strip the V2_7_ prefix on V2_7_DataItem_constructs_with_correct_metadata — same rationale.

  Every other migration is an identity rename. A separate Migration_map_covers_the_full_baseline_of_34_entries guard fires RED if the map itself is trimmed below the 34-entry baseline.

• PerVersionFolderProhibitionTests enforces the folder-topology invariant: no V<N_M>/ directory, V<N_M>*Tests.cs file, or V<N_M>*Tests fixture class may return to the tree.

• TopicFixtureCoverageTests enforces the coverage-parity invariant at source-text level: for every migrated spec-anchor type, the designated topic-fixture source file must name the type at least once. Catches the rename-plus-body-swap regression a name-only guard would miss.

For Phase 3, byte-identity is enforced by Regen_is_deterministic_across_two_invocations in the new generator-tests project. Phase 4 adds Current_XMI_regen_matches_committed_g_cs_tree (now plain [Test], not [Explicit]) plus the two DeltaRegenTests guards.

Test renames

Verbatim table generated from the migration script's rename audit (34 rows, one per pre-migration [Test] / [TestCase] method entry). case_map = identity means the [TestCase] parameter tuples migrated verbatim; the one non-identity row (V2_7_DataItem_constructs_with_correct_metadata) cross-multiplies its 8 (Type, TypeId, Category) tuples with MTConnectVersionMatrix.All under the new matrix mechanism.

Old file	Old method	New file	New method	Case map	Rationale
V2_6ComponentAndEnumTests.cs	CuttingTorchComponent_constructs_with_correct_type	Devices/Components/ComponentTests.cs	CuttingTorchComponent_constructs_with_correct_type	identity	identity migration
V2_6ComponentAndEnumTests.cs	ElectrodeComponent_constructs_with_correct_type	Devices/Components/ComponentTests.cs	ElectrodeComponent_constructs_with_correct_type	identity	identity migration
V2_6ComponentAndEnumTests.cs	MediaType_QIF_MBD_value_present_in_v2_6	Enums/EnumArmTests.cs	MediaType_QIF_MBD_value_present	identity	strip _in_v2_6 suffix; version is now matrix property
V2_6DataItemTypeTests.cs	AssetAddedDataItem_constructs_with_event_metadata	Devices/DataItems/DataItemTypeTests.cs	AssetAddedDataItem_constructs_with_event_metadata	identity	identity migration
V2_6DataItemTypeTests.cs	AssetAddedDataItem_with_deviceId_produces_qualified_id	Devices/DataItems/DataItemTypeTests.cs	AssetAddedDataItem_with_deviceId_produces_qualified_id	identity	identity migration
V2_6DataItemTypeTests.cs	AssociatedAssetIdDataItem_constructs_with_event_metadata	Devices/DataItems/DataItemTypeTests.cs	AssociatedAssetIdDataItem_constructs_with_event_metadata	identity	identity migration
V2_6DataItemTypeTests.cs	AssetAddedDataItem_inherits_from_DataItem	Devices/DataItems/DataItemTypeTests.cs	AssetAddedDataItem_inherits_from_DataItem	identity	identity migration
V2_6DataItemTypeTests.cs	AssociatedAssetIdDataItem_inherits_from_DataItem	Devices/DataItems/DataItemTypeTests.cs	AssociatedAssetIdDataItem_inherits_from_DataItem	identity	identity migration
V2_6DataItemTypeTests.cs	AssetChangedDataItem_description_narrowed_in_v2_6	Devices/DataItems/DataItemTypeTests.cs	AssetChangedDataItem_description_narrowed	identity	strip _in_v2_6 suffix; version is now matrix property
V2_7DataItemTypeTests.cs	V2_7_DataItem_constructs_with_correct_metadata	Devices/DataItems/DataItemTypeTests.cs	DataItem_constructs_with_correct_metadata	(Type,TypeId,Category) tuples cross-multiplied with MTConnectVersionMatrix.All	strip V2_7_ prefix; version is now matrix property
MTConnectVersionsTests.cs	Version26_constant_equals_2_6	MTConnectVersionsTests.cs	Version26_constant_equals_2_6	identity	identity migration; kept plain [Test] (constant-value invariant, not version-sensitive)
MTConnectVersionsTests.cs	Version27_constant_equals_2_7	MTConnectVersionsTests.cs	Version27_constant_equals_2_7	identity	identity migration; kept plain [Test] (constant-value invariant, not version-sensitive)
MTConnectVersionsTests.cs	Max_equals_Version27	MTConnectVersionsTests.cs	Max_equals_Version27	identity	identity migration; kept plain [Test] (constant-value invariant, not version-sensitive)
MTConnectVersionsTests.cs	Every_published_version_constant_is_distinct_and_monotonic	MTConnectVersionsTests.cs	Every_published_version_constant_is_distinct_and_monotonic	identity	identity migration; kept plain [Test] (structural invariant)
MTConnectVersionsTests.cs	Version19_field_does_not_exist	MTConnectVersionsTests.cs	Version19_field_does_not_exist	identity	identity migration; kept plain [Test] (structural invariant)
V2_7ComponentTests.cs	PinToolComponent_constructs_with_correct_type	Devices/Components/ComponentTests.cs	PinToolComponent_constructs_with_correct_type	identity	identity migration
V2_7ComponentTests.cs	ToolHolderComponent_constructs_with_correct_type	Devices/Components/ComponentTests.cs	ToolHolderComponent_constructs_with_correct_type	identity	identity migration
V2_7ConfigurationDataSetTests.cs	DataSet_base_constructs_and_implements_IDataSet	Devices/Configurations/ConfigurationTests.cs	DataSet_base_constructs_and_implements_IDataSet	identity	identity migration
V2_7ConfigurationDataSetTests.cs	AxisDataSet_has_xyz_fields_and_implements_IDataSet	Devices/Configurations/ConfigurationTests.cs	AxisDataSet_has_xyz_fields_and_implements_IDataSet	identity	identity migration
V2_7ConfigurationDataSetTests.cs	OriginDataSet_has_xyz_fields_and_implements_IDataSet	Devices/Configurations/ConfigurationTests.cs	OriginDataSet_has_xyz_fields_and_implements_IDataSet	identity	identity migration
V2_7ConfigurationDataSetTests.cs	RotationDataSet_has_abc_fields_and_implements_IDataSet	Devices/Configurations/ConfigurationTests.cs	RotationDataSet_has_abc_fields_and_implements_IDataSet	identity	identity migration
V2_7ConfigurationDataSetTests.cs	ScaleDataSet_implements_IDataSet	Devices/Configurations/ConfigurationTests.cs	ScaleDataSet_implements_IDataSet	identity	identity migration
V2_7ConfigurationDataSetTests.cs	TranslationDataSet_implements_IDataSet	Devices/Configurations/ConfigurationTests.cs	TranslationDataSet_implements_IDataSet	identity	identity migration
V2_7ConfigurationDataSetTests.cs	Axis_inherits_AbstractAxis_and_constructs	Devices/Configurations/ConfigurationTests.cs	Axis_inherits_AbstractAxis_and_constructs	identity	identity migration
V2_7ConfigurationDataSetTests.cs	Origin_inherits_AbstractOrigin	Devices/Configurations/ConfigurationTests.cs	Origin_inherits_AbstractOrigin	identity	identity migration
V2_7ConfigurationDataSetTests.cs	Rotation_inherits_AbstractRotation	Devices/Configurations/ConfigurationTests.cs	Rotation_inherits_AbstractRotation	identity	identity migration
V2_7ConfigurationDataSetTests.cs	Scale_inherits_AbstractScale	Devices/Configurations/ConfigurationTests.cs	Scale_inherits_AbstractScale	identity	identity migration
V2_7ConfigurationDataSetTests.cs	Translation_inherits_AbstractTranslation	Devices/Configurations/ConfigurationTests.cs	Translation_inherits_AbstractTranslation	identity	identity migration
V2_7ConfigurationDataSetTests.cs	AbstractAxis_is_abstract	Devices/Configurations/ConfigurationTests.cs	AbstractAxis_is_abstract	identity	identity migration
V2_7ConfigurationDataSetTests.cs	AbstractOrigin_is_abstract	Devices/Configurations/ConfigurationTests.cs	AbstractOrigin_is_abstract	identity	identity migration
V2_7ConfigurationDataSetTests.cs	AbstractRotation_is_abstract	Devices/Configurations/ConfigurationTests.cs	AbstractRotation_is_abstract	identity	identity migration
V2_7ConfigurationDataSetTests.cs	AbstractScale_is_abstract	Devices/Configurations/ConfigurationTests.cs	AbstractScale_is_abstract	identity	identity migration
V2_7ConfigurationDataSetTests.cs	AbstractTranslation_is_abstract	Devices/Configurations/ConfigurationTests.cs	AbstractTranslation_is_abstract	identity	identity migration
V2_7SampleObservationTests.cs	WaterHardness_sample_observation_round_trip	Observations/SampleObservationTests.cs	WaterHardness_sample_observation_round_trip	identity	identity migration

Depends on

None. This PR is orthogonal to every other open campaign PR — it touches only tests/MTConnect.NET-Common-Tests/, docs/testing/, build/MTConnect.NET-SysML-Import/, and adds the new tests/MTConnect.NET-Generator-Tests/ project. The eventual MTConnect Standard v2.8 bump PR depends on THIS.

Phase 5 — verification pass

Full-tree cross-check + wire-format matrix + all-categories dotnet test sweep on bluefin at head 619dd1bc, 9 assemblies green (Common 3987, XML 131, JSON 68, MqttRelay 62, Docs 67, HTTP 112, JSON-cppagent 363, SHDR 35, Compliance 225 — 5,050 tests / 0 failed / 0 skipped, filtered-excluding the documented Docs-Tests bluefin Playwright flake covered by #227's baseline).

Phase 5 also lands four categories of atomic test / doc / generator hardening on top of the Phase 4 head:

• Generator delta emitter hardening — fix(generator): delta emitter deletes stale UNCHANGED / REMOVED files (delta CLI now cleans up files no longer emitted by the current XMI, closing a stale-artifact class the Phase 4 baseline left latent).
• Generator CLI failure-path coverage — test(generator): CLI failure-path coverage for Program.cs, test(generator): delta-mode Compat body + stats-line invariants, test(generator): relax malformed-XMI test to non-zero exit contract (tightens the delta-mode invariants pinned in Phase 4 and rounds out the failure-mode surface of Program.cs).
• Test-harness robustness — fix(tests): drain stdout/stderr concurrently + tighten TopicAnchors floor (removes a bluefin-observed deadlock in the process-driver + raises the TopicFixtureCoverageTests coverage floor to reflect the actual anchor set at HEAD).
• Documentation — docs(generator): document delta CLI + Stryker gate + generator-tests tier (docs/testing/mutation-testing.md, docs/testing/version-matrix-convention.md, and the two per-version testing pages document the new delta CLI flag, the Stryker gate, and the generator-tests tier).

Phase 5 — mutation-testing framework adoption

chore(tests): adopt Stryker.NET mutation-testing framework lands stryker-config.json at the repo root wired to MTConnect.NET-Common as the pilot target (net8.0, --mutation-level Complete, .g.cs generated surfaces excluded from mutation, thresholds at 100). A follow-up commit chore(tests): exclude Stryker Regex mutator (upstream bug workaround) disables the Regex mutator to work around a Stryker.NET v4.16.0 crash on interpolated-string Regex constructors (upstream issue tracked in the mutation-testing doc).

Baseline mutation run on MTConnect.NET-Common at head 43497c5d (bluefin, concurrency 4, 4 min elapsed):

Metric	Count
Total mutants	13,990
Killed	790
Survived	1,211
NoCoverage	8,275
Timeout	7
CompileError	188
Ignored	3,519
Files covered	1,141

The 7.75% baseline is expected: this PR ADOPTS the framework. Killing every survived + no-coverage mutant is realistically 100+ engineering hours and cross-cuts every subsystem in the assembly (codecs, model, validators, formatters), so a companion commit chore(tests): pin Stryker threshold at 7.75% baseline pins stryker-config.json thresholds to { high: 8, low: 5, break: 5 } — well below the 7.75% baseline so the current PR does not regress and so subsequent PRs inherit a pass-through gate until the dedicated coverage-quality campaign raises the floor. #242 tracks the phased follow-up campaign (categorize survivors -> draft kill tests per subsystem -> raise thresholds in step to 20 -> 40 -> 60 -> 80%+ -> expand Stryker to the sibling assemblies -> wire into CI as a required check once the pilot clears 80%).

Dime review cycle 1

Six-agent sweep (code-review + security-audit + simplification + improvement + documentation-audit + test-coverage-audit) dispatched at head d83cff2d. Findings distribution:

Agent	Findings
code-review	2 MEDIUM + 1 LOW
security-audit	0
simplification	6 LOW
improvement	3 HIGH + 4 MEDIUM + 3 LOW
documentation-audit	1 HIGH + 3 MEDIUM
test-coverage-audit	2 HIGH + 3 MEDIUM + 1 LOW + 1 INFO

Disposition — three-state (Fixed / Closed-with-rationale / Tracked):

• Fixed — 12 fixes landed atomically in-cycle. The eight commits after d83cff2d (through 43497c5d) are the fix landings + the mutation-testing-framework adoption plus its Regex-mutator workaround.
• Tracked — five items deferred to follow-up PRs / issues: test-harness DRY consolidation, DeltaStats as record struct, --compat-version-label auto-derivation, dotnet run overhead removal on the generator smoke path, redundant --json-dump re-Parse, ADDED / REMOVED partition direct assertion, Parse throw-hardening, and a Stryker CI runner.
• Closed with rationale — two items: Measurement.Code deliberate spec-alignment (Section 8.2 §Measurement code retains the spec's normative token, not the more-general C# identifier the reviewer proposed); empty-HistoricalAnchors documented anticipation (the reviewer's proposal treats an anticipated multi-version anchor map as a same-PR gap; the map is deliberately unused until the second historical spec version lands).

Full test sweep on bluefin at head 619dd1bc: 9 of 10 assemblies green (Docs-Tests documented bluefin Playwright flake, docs-orthogonal to this PR — same class of infra flake pinned by #227's baseline).

(Zero unfixed findings — Ready-eligible.)

Dime review cycle 2

Six-agent sweep dispatched at head 619dd1bc (the ottobolyos-directive Stryker baseline-pin commit). Cycle-2 scope is deliberately narrow — the sole diff under review is stryker-config.json threshold values { high: 100, low: 100, break: 100 } -> { high: 8, low: 5, break: 5 } + a JSONC provenance header cross-referencing #242. Findings distribution:

Agent	Findings
code-review	1 MEDIUM + 1 LOW
security-audit	0
simplification	0
improvement	1 MEDIUM + 3 LOW
documentation-audit	3 HIGH + 2 MEDIUM
test-coverage-audit	1 HIGH + 2 MEDIUM + 1 LOW

Disposition — three-state (Fixed / Closed-with-rationale / Tracked):

• Fixed — 7 items landed atomically across two commits:
  • 28d9b9fe test(generator): pin stryker-config.json thresholds + JSONC-parser assumption (test-coverage-audit) — 231-LOC StrykerConfigPinTests.cs guarding break window [5, 8], monotone ordering break <= low <= high <= 100, JSONC parseability via System.Text.Json with CommentHandling=Skip, Regex mutator ignore-list membership, and pilot-project pin (MTConnect.NET-Common.csproj + MTConnect.NET-Common-Tests); 5/5 GREEN + 27/27 pre-existing = 32/32 Generator-Tests GREEN on bluefin net8.0.
  • d79e9fc8 docs(testing): sync mutation-testing.md + Stryker.NET pin to cycle-2 threshold pin (this cycle) — F-CR-001 rewrites the stryker-config.json JSONC header to say explicitly that high: 8 sits above the 7.75 % baseline as an aspirational marker while low: 5 and break: 5 sit below as the pass-through gate; F-IMP-001 pins dotnet-stryker at v4.16.0 in .config/dotnet-tools.json alongside dotnet-reportgenerator-globaltool so the baseline stays reproducible under dotnet tool restore; F-DOC-001..005 sync docs/testing/mutation-testing.md in five places (JSON snippet, Thresholds bullet, --break-at CI shape, disposition-rules third arm, References list test-quality: raise MTConnect.NET-Common mutation score from 7.75% baseline #242 pointer).
• Tracked — 3 LOW deferred to test-quality: raise MTConnect.NET-Common mutation score from 7.75% baseline #242: F-IMP-002 (upgrade the bare #242 JSONC-comment reference to the full https://github.com/... URL for editor click-through in non-git contexts); F-IMP-003 (flip since.enabled: true for incremental-mutation runs once the pilot clears 20 %); F-IMP-004 (emit a sibling stryker-baseline.json capturing the score / date / SHA / ignored-mutators / tool-version snapshot so a CI check can diff future runs against the pinned baseline, not just against the break threshold).
• Closed with rationale — 2 LOW: F-CR-002 (the JSONC header's PR #233 head 43497c5d remains accurate as the measurement anchor for the 7.75 % baseline even after the head advances past that SHA); the simplification agent's no-finding is a self-close (the diff is already at minimum shape for a follow-up-tracked baseline pin).

Threshold values kept at { high: 8, low: 5, break: 5 } per user directive 2026-08-20; only prose + tool pin updated this cycle. Full test sweep on bluefin at head d79e9fc8 matches cycle-1's assembly count (docs-only + config-only diff introduces no new production behavior).

(Zero unfixed findings — Ready-eligible.)

Follow-up

• Component / Composition template merger — the third consolidation
  candidate identified by the Phase 3.2 audit. Deferred this pass
  (six-gate ceiling breach without parameterization surgery); reopens
  if a future spec bump's delta emission surfaces a compelling reason
  to unify the two.
• Historical-XMI iteration — the plan's original Phase 4.4 cross-verify
  across every build/sysml-model tag is out of scope this pass
  (ottobolyos 2026-08-20; the submodule ships one snapshot per spec
  bump). Reopens once the submodule is extended with multi-tag history
  or the campaign transitions to hosting spec-tags externally.
• Mutation-testing kill campaign — the framework adopted in Phase 5
  reports 1,211 Survived + 8,275 NoCoverage mutants on
  MTConnect.NET-Common at the pilot baseline (mutation score 7.75%).
  Killing every surviving and no-coverage mutant is a multi-week
  test-authoring effort tracked in test-quality: raise MTConnect.NET-Common mutation score from 7.75% baseline #242 — the
  phased campaign categorizes survivors, drafts kill tests per subsystem,
  and raises thresholds in step (20 -> 40 -> 60 -> 80%+), then expands
  Stryker to the sibling assemblies and wires it into CI as a required
  check once the pilot clears 80%. The stryker-config.json landed here
  • the mutation-testing doc together stand the gate up so every future
    PR CAN run the mutation sweep; the pin at { high: 8, low: 5, break: 5 }
    keeps the current PR + every future one green until the dedicated
    campaign raises the floor.

[ottobolyos]

@PatrickRitchie — this one is ready for review whenever you have a moment.

It consolidates the V2_6_V2_7/ per-version fixture family into topic-first single-file layout with matrix-parameterised version gates, and adds two permanent reflection guards (AssertionParityTests, PerVersionFolderProhibitionTests) that machine-enforce the new convention going forward. Zero library-side code changes — purely a test-suite refactor plus scaffolding.

Full unfiltered sweep on my end: 4975 passed / 0 failed / 0 skipped. Mutation test on the folder-prohibition guard confirms it fires RED when the pattern regrows.

I'm happy to walk through any of the design decisions or split it up differently if the shape isn't quite right.