[devin-ai-integration]

What and why

Pins the OMG pilot implementation to release 2026-09 (jupyter-sysml-kernel 0.63.0; pilot 38665073bbf02849bd309b106eba79cbb3a1ee1d, release b7a6b77000c9136f8f24822fc9d2cb237c76267f), re-provisions every pinned artifact at that pin, syncs the vendored standard library and the kpar fixtures, re-records every oracle baseline and moves every current-pin mention of 2026-08 / 0.62.0 in the README, docs and testing skills — the same file set as the 2026-08 bump.

What moved upstream between 2026-08 and 2026-09

27 pilot commits. Of note:

• Connection and flow definition ends must be usages (ST6RI-974, resolving SYSML21-681): the validator gains validateConnectionDefinitionEndsAreUsages and validateFlowDefinitionEndsAreUsages (217 → 219 constraints), and Systems Library/Flows.sysml is repaired in the same change — Message and messages own end occurrence source/target usages redefining Transfer::source/target, messages becomes an action rather than a message, plus a doc-comment typo fix. That is exactly the repair the errata overlay carried as Flows-46, Flows-99 and Flows-103.
• The Xpect suites rename FlowConnectionUsage_Invalid → FlowUsage_Invalid and reword its four messages to the SysML v2.1 names (A flow definition can have at most two ends., A flow usage must be typed by flow definitions.); three suites are new (MultiplicityAlias_valid.kerml.xt, ConnectionUsage_End_Invalid.sysml.xt, FlowUsage_End_Invalid.sysml.xt).
• Default-multiplicity handling, enumeration-valued metadata evaluation ("expression result processing"), parser post-processors and model-mutation API work — the pilot-side behaviour behind the differential movement below.
• Library project stamps 1.1.0-dev.20260801 / 2.1.0-dev.20260801 → …20260901; the only library text change is Flows.sysml. The three Xtext grammars are byte-identical between the two tags (fetched at both commits and diffed), so the anonymous-alias form the new Xpect suite exercises was always grammar-admitted; SysML.ecore (20250201) and SysMLv1Tov2.xmi did not move either.

Standard-library sync

internal/workspace/libs/stdlib is byte-identical to the release's sysml.library again: ten .project.json stamps and Systems Library/Flows.sysml (the upstream repair), snapshot regenerated (make stdlib-snapshot && make stdlib-snapshot-check). Per-file movement against develop: no new rows, no moved rows; three removed rows — the errata entries Flows-46, Flows-99, Flows-103 (registry 16 → 13 entries, 9 → 6 corrections) and the matching validationStdlibCorrections rows of internal/workspace/model, whose gates (TestStdlibCorrectionsAreAccountedFor, TestStdlibValidationCorrectionsClearTheirFindings) now hold over an empty correction list. tests/kpar/testdata/omg carries the ten …20260901.kpar archives.

Two small code changes the pin surfaced

• Anonymous alias members parse (internal/syntax/parser/namespace.go): alias for Base::exactlyOne; — AliasMember's memberName is optional in KerML.xtext — was rejected with "for" is a reserved keyword; the pilot's new MultiplicityAlias_valid.kerml.xt declares it clean. Golden fixture tests/parser/testdata/parse/kerml_anonymous_alias.kerml.
• Flow diagnostics use the v2.1 names: flow-def-too-many-ends and the flow usage-typing diagnostic said "flow connection definition" / "flow connection" (the v2.0 names of FlowDefinition / FlowUsage); they now say what the pilot's Xpect rows declare, so those four rows agree word-for-word as before the rename.

Oracle baselines — every movement and its adjudication

Oracle	develop	this PR	Adjudication
pilot-diff	920 files, 517 fully agreeing; 1098 agreed, 366 only ours, 2448 only pilot (pilot diagnostics 3557)	920 files, 514; 1098 / 366 / 2459 (3568)	Nothing of ours moved. The pilot now warns Bound features should have conforming types on ten enumeration-valued metadata bindings in its own Metadata Examples/{Requirement,Risk,Verification}MetadataExample.sysml (totalRisk = high; with RiskLevelEnum :> RiskLevel, kind = (test,demo); with kind : VerificationMethodKind) — the literal's enumeration is the attribute's type or specializes it, so the bindings conform and silence is right: sysml-examples fully agreeing 88 → 85, only-pilot 0 → 10. examples/self-model/document.sysml:403 draws one more pilot Duplicate of inherited member name 'result' row (17 → 18, the reference's inherited-result cascade, unmapped): examples only-pilot 61 → 62. Errata 16 → 13 / 9 → 6 as above.
pilot-xpect	429 files, 1264 assertions / 1326 expectations; 1293 agree (248 wording-only), 33 disagree	432 files, 1267 / 1329; 1296 agree (248 wording-only), 33 disagree	+3 files: MultiplicityAlias_valid (agrees after the alias fix), ConnectionUsage_End_Invalid and FlowUsage_End_Invalid (agree word-for-word). The four reworded FlowUsage_Invalid rows agree word-for-word after the wording change. Suite census test tools/referee/xpect/w5c_census_test.go moved to 304 + 128 files, errors 463 / noErrors 277 line notes.
pilot-reject	349 cases: 337 both reject, 12 ours only	349: 339 / 10	semantic/connection-def-end-not-usage.sysml and semantic/flow-def-end-not-usage.sysml move ours-only → both-reject: the pilot now implements the check our W10BEndKindPass already reported.
validation-census	217 constraints, 163 reported (156 faithful / 7 approximate, 7 deliberate)	219, 165 (158 / 7, 7 deliberate)	Two new rows, both faithful: validateConnectionDefinitionEndsAreUsages / validateFlowDefinitionEndsAreUsages → connection-def-end-not-usage / flow-def-end-not-usage, with probes under tools/census/validation/testdata/probes/. The manifest's errata list drops the three Flows entries.
grammar-coverage	pin 2026-08	pin 2026-09	Grammars identical, so no production or evidence verdict moves; the compact baseline's evidence-root sizes are re-measured: stdlib 31227 → 31233 lines (Flows.sysml), parser-fixtures 268 → 286 files (develop's fixtures added since the 2026-10-20 recording plus kerml_anonymous_alias.kerml), tests/testdata 30 → 40 files, examples 9653 → 9656 lines, spec-examples 6344 → 6242 lines (the extraction fix 3a9186342 landed after that recording; the PDFs are unchanged).
grammar-metaclass-map	agree 61, disagree 9, undecided 657	63 / 9 / 655	KerML.xtext 26 → 27 agree, SysML.xtext 35 → 36: the two productions the 2026-08 jar exposed no reflective metaclass for (no-reflective-metaclass 2 → 0) resolve against the 0.63.0 jar and agree with their assigned metaclass.
pilot-apijson-diff	930 files; 91870 matched (18396 ours-only, 37363 pilot-only); pilotFailed 56	930; 91667 (18389 / 37472); pilotFailed 59	The pilot's exporter now throws (Element.eResource() null) on the same three Metadata Examples/*MetadataExample.sysml files whose enumeration-valued bindings it evaluates — pilotFailed 56 → 59 (52 spec examples, 4 fixtures, these 3), so sysml-examples matched 26739 → 26536 and the headline moves with those three files. Scorecard rows 11730 → 11728; the launch-ready gap table in docs/project/api-json-differential.md is re-stated from the baseline (its gated test passes). Nothing of ours changed.
transformation-census	783 mapping classes	783	-check -require-xmi passes unchanged (SysMLv1Tov2.xmi did not move).
pssm-referee	—	—	-check passes, bucket counts unchanged (the PSSM suite is not pinned to the pilot).
ontology table	pilot 2026-08 header	pilot 2026-09 header	SysML.ecore 20250201 unchanged: 415 properties / 175 classes, only the provenance header moves.

Corpus gates: TestTrainingExamples, TestPilotCorpora, TestSpecExamples (with OPENSYSML_REQUIRE_* set), TestPilotLibraryXMI and TestStdlibConformance pass with no per-file movement; tests/kpar passes over the 20260901 archives. The spec-example fixtures' digest moves only because each fixture header records the release tag — the two PDFs they are extracted from are byte-identical.

Re-merges against develop

Develop was merged twice, each time once as a merge commit (no rebase, no force-push, no empty commits), and every baseline and derived document this PR records was regenerated on the merged tree with the same -check → read the diff → -update discipline.

First re-merge, develop ab1a9cf9a (after #1484, #1477, #1502, #1524 and #1523 landed): every conflict (the four testing skills, README, architecture, adjudications, the API-JSON / differential / Xpect baselines and their documents, spec-compliance, validation-constraints, stdlib.snapshot) was resolved by taking develop's current file and regenerating on the merged tree, then re-applying this PR's pin edits. Against develop's file and this PR's previous head 4713bf893: stdlib.snapshot regenerated over develop's extension-library pin 5c5deea6d (scripts/sync-extension-libraries.sh in sync); pilot-differential-baseline.json 521 → 518 fully agreeing, 359 → 359 only ours, 2450 → 2461 only pilot (pilot-side rows unchanged from this PR's adjudication; the only-ours drop 366 → 359 and fully agreeing +4 are develop's #1484, which withdrew part-usage-part-definition); pilot-xpect-baseline.json 1296/33 → 1299/30 agree/disagree (the +3 are develop's #1484, noErrors 265 → 266 of 277); pilot-rejection-baseline.json 337/12 → 339/10 (this PR's movement only); api-json-differential-baseline.json 92658 → 92455 matched, pilotFailed 56 → 59 (the pin's part is the three metadata examples the 0.63.0 exporter fails on; the rest, 91667 → 92455 against the previous head, is develop's #1477/#1502 corpus growth); constraint manifest 217 → 219; grammar baselines re-measured (metaclass map agree 61 → 63); ontology table and Python metamodel provenance header only.

Second re-merge, develop 9618afae5 then 0286c9845 (after #1537, #1536, #1539, #1534, #1538, #1542, #1541, #1544, then #1547, #1546, #1545 landed). The only conflicts were docs/project/spec-compliance.md and docs/project/validation-constraints.md, where #1536's Beta 2 constraint-name rows met this PR's end-usage row and pin wording: #1536's rows and formal names are kept as develop has them, this PR's 2026-09 / 0.63.0 mentions and its end-usage row are kept, and the derived census lines were rewritten by validation-census / doc-counts rather than by hand. The second merge (0286c9845) touched no oracle input — README, docs, the MDK Java plugin and the release-digest hygiene test — so every -check was re-run on that head and nothing moved. Each re-recorded artifact against develop's file and against this PR's previous head 9ca600a59, by category:

Artifact	vs develop 0286c9845	vs previous head 9ca600a59	Category
pilot-differential-baseline.json	521 → 518 fully agreeing, 359 → 359 only ours, 2450 → 2461 only pilot (pilot diagnostics 3559 → 3570); provenance 2026-08 → 2026-09	identical except the examples evidence-root digest, which develop moved (its own re-record over the grown examples/) and this run reproduces	regenerated, no row moved
pilot-xpect-baseline.json / pilot-xpect.md	1296/33 → 1299/30 (432 files, 1267 / 1329)	identical	regenerated, no row moved
pilot-rejection-baseline.json	337/12 → 339/10	identical	regenerated, no row moved
api-json-differential-baseline.json / .md	92658 → 92455 matched, pilotFailed 56 → 59; the document's scorecard line and sysml-examples row re-stated from the baseline	three properties of the pilot exporter's own output flip phase: scorecard.FeatureChainExpression.featuringType 284/230 → 287/227, pilot-examples oursMissing 312072 → 312075 / absentBoth 577330 → 577327, unkeyedReferences 111955 → 111958 (totals follow); plus develop's examples digest	re-recorded in the phase this run yields — see the note below
constraint-manifest.json / validation-constraints.md	217 → 219 constraints; 163 → 165 reported (156 → 158 faithful, 7 approximate, 7 deliberate, 47 unknown)	formal names and the deliberate rows are #1536's, now carried as develop has them; this PR's two faithful end-usage rows unchanged	regenerated, census agrees with manifest and compliance map
spec-compliance.md	this PR's pin mentions (2026-09 / 0.63.0), the end-usage row (both-reject under 2026-09; published Flows.sysml carries the repair) and the pilot's 2026-09 wording in the flow-ends row (A flow definition can have at most two ends.); no § number touched	#1536's Beta 2 rows merged in; the remaining "pinned pilot" 2026-08 / 0.62.0 mentions moved to 2026-09 / 0.63.0	merged + edited
grammar-coverage-baseline.json	pin and evidence roots re-measured	examples root 287 → 288 files, 6716 → 6729 lines (develop's examples added since the previous recording); no production or evidence verdict moved	regenerated
grammar-metaclass-map-baseline.json	agree 61 → 63	identical	regenerated
stdlib.snapshot	regenerated over develop's extension-library pin (sync-extension-libraries.sh in sync) and the 2026-09 standard library	byte-identical	make stdlib-snapshot-check passes
README / architecture / doc-counts figures	doc-counts current on the merged tree	README carries develop's two new lines only	regenerated, -check current
ontology table, Python metamodel	provenance header only	identical	pymetamodel -check current

Final re-merge, develop 1ed77285e (after #1550, #1551, #1553, #1554, #1555, #1556, #1557 and #1558 landed). The only conflicts were docs/project/api-json-differential-baseline.json and docs/project/api-json-differential.md, where #1555's canonicalized referee (the declared pilotNondeterminism list and the seventh, pilot-nondeterministic scorecard column) met this PR's re-record at 2026-09: develop's two files were taken whole, the baseline was then re-recorded on the merged tree with the canonicalized referee, and the document's pin wording, headline, pilotFailed sentence and five root rows were re-stated from that baseline (the declared-nondeterminism section is develop's, plus one sentence recording that the disagreement reproduces at 2026-09). spec-compliance.md and validation-constraints.md merged clean — the rows #1550, #1551, #1553 and #1554 added land beside this PR's — and every other -check was re-run on the merged tree. Each re-recorded artifact against develop's file and against this PR's previous pushed head 9ca600a59, by category:

Artifact	vs develop 1ed77285e	vs previous head 9ca600a59	Category
api-json-differential-baseline.json / .md	92658 → 92455 matched (18486 → 18479 ours-only, 37657 → 37766 pilot-only), pilotFailed 56 → 59, 11730 → 11728 rows, 514 pilot-nondeterministic on both sides; provenance 2026-08 → 2026-09	the FeatureChainExpression.featuringType row moves whole into the seventh column: [0, 0, 284, 0, 230, 0] → [0, 0, 0, 0, 0, 0, 514], so ours-missing 1074296 → 1074012 and absent-both 1973602 → 1973372 (−514 together), unkeyed references 111955 → 111949 (the row's six); every other row, root and total identical	re-recorded with the canonicalized referee; the phase flip this PR previously documented is now scored, not recorded
grammar-coverage-baseline.json	pin and evidence roots re-measured	parser-fixtures root 287 → 290 files, 6716 → 6751 lines (develop's fixtures from #1551 and #1558); no production or evidence verdict moved	regenerated
pilot-differential-baseline.json	521 → 518 fully agreeing, 359 → 359 only ours, 2450 → 2461 only pilot (pilot diagnostics 3559 → 3570)	identical except the examples evidence-root digest develop moved	-check reproduces, no row moved
pilot-xpect-baseline.json, pilot-rejection-baseline.json, grammar-metaclass-map-baseline.json	agree 1296 → 1299, disagree 30 → 30 (429 → 432 files, 1264 → 1267 assertions); 337/12 → 339/10; agree 61 → 63 (the two no-reflective-metaclass productions resolve against the 0.63.0 jar)	byte-identical	-check reproduces
constraint-manifest.json / validation-constraints.md / spec-compliance.md	217 → 219 constraints, 163 of 217 → 165 of 219 reported (158 faithful, 7 approximate, 7 deliberate, 47 unknown); develop's new rows (#1550, #1551, #1553, #1554) carried as develop has them	#1536's formal names carried as develop has them; this PR's rows unchanged	merged clean, census agrees
stdlib.snapshot, ontology table, Python metamodel, doc-counts figures	byte-identical / -check current on the merged tree	identical	no change

API-JSON nondeterminism. The three-element FeatureChainExpression.featuringType flip this PR documented against its earlier heads (284/230 ↔ 287/227 between ours-missing and absent-both, reproduced by this PR's own runs, by develop's history and by develop's nightly oracle-reproduction) is the row #1555 declares in tools/referee/apijson/nondeterminism.go; the canonicalized referee scores every comparison of that row pilot-nondeterministic, so the re-recorded baseline reproduces on every JVM and -check no longer depends on the exporter's phase. The disagreement still reproduces at 2026-09, so the declaration stays, and the document says so.

Gates on the final head (1ed77285e merged): go build, go vet, gofmt -l . empty; pilot-diff / pilot-xpect / pilot-reject / pilot-apijson-diff / validation-census -require-jar / transformation-census -require-xmi / pssm-referee -check all pass; grammar-coverage and grammar-metaclass-map reproduce their baselines byte for byte; corpus gates with OPENSYSML_REQUIRE_* set, TestPilotLibraryXMI, TestStdlibConformance, make stdlib-snapshot-check, pymetamodel -check pass; make test-quick, go test -C tools ./... and go test ./... (full run, -timeout 0) pass; doc-counts -check, doc ids/links and the changelog check are clean.

Specification PDFs

The four specification PDFs under the release's doc/ (1-Kernel_Modeling_Language, 2a-OMG_Systems_Modeling_Language, 2b-SysML_v1_to_v2_Transformation, 3-Systems_Modeling_API_and_Services) have identical blob ids at 2026-08 and 2026-09, so the clause numbers docs/project/spec-compliance.md cites do not move; that file's § numbers are left to #1526 — only its pin mentions, the end-usage row and the flow-ends row's quoted pilot wording change here; #1536's Beta 2 constraint-name rows are carried as develop has them.

Specification basis

The alias fix implements KerML.xtext AliasMember: … 'alias' ( '<' declaredShortName '>' )? declaredName? 'for' … (KerML 1.0 §8.2.3.4.3); the wording change follows the SysML v2.1 names of FlowDefinition / FlowUsage. No compliance row moves.

How it was verified

• go build ./..., go vet ./..., gofmt -l . (empty), go test ./... (full run, -timeout 0: 122 packages ok, 0 failed; cmd/sysml and tests/codegen exceed the 10-minute default under parallel load on this machine, as they did before this PR); go test -C tools ./... passes.
• go run -C tools ./cmd/doc-counts -check, python3 scripts/check-doc-ids.py, python3 scripts/check-doc-links.py, python3 scripts/changelog.py check all clean.
• Each oracle's -check run first, the movement read from the diff, then -update; the table above is that reading.

Checklist

• make test and make lint pass locally
• Tests added or updated for the change
• Documentation extended where it already covers the surface (see CONTRIBUTING.md)
• Changelog entry added as changes/unreleased/<slug>.<section>.md, not as an edit to CHANGELOG.md
• baselines regenerated and make docs-counts run if a gate count moved (compliance rows need nothing: the census is counted at docs build)
• No internal work-item labels (waves, slices, F4, K5) in the body, docs, or changelog

[devin-ai-integration]

I'll fix CI failures and address comments from users with write access. I'll skip comments containing "(aside)".

• Disable automatic comment, CI, and merge conflict monitoring

[devin-ai-integration]

🔍 Devin Review: 1 flag

Not posted on this PR by your GitHub settings — view it in Devin Review. (Configure)