fix(migrate): close v1 migration fidelity residuals in calls, results and run configurations

[devin-ai-integration]

What and why

The SysML v1 → v2 migrator, its -migration-results sidecar and the -compare-results harness lost fidelity on large Cameo Simulation Toolkit models in ways that were all general: a call whose required input receives no value was replaced by a placeholder performing nothing, result snapshots were matched to a configuration by package proximity rather than by the execution target's classifier, Monte Carlo summaries were read as single runs, configurations with no stored snapshot or no numberOfRuns were skipped or refused, parametric configurations were refused with a blanket message, and a few runtime defects (a case body's timed steps off the clock, a bodyless state behavior refused, a state machine's poll of the signals in flight re-done at every step) surfaced only on such models. Every change is expressed in UML/SysML v1 semantics, the MagicDraw profile (by namespace provenance) or the Simulation Toolkit's conventions; nothing is keyed to a particular model.

Migrator (internal/translate/migrate)

• Calls whose required input receives no value are performed. A call passing no argument, or a pin nothing reaches, for a parameter with no default and lower bound above zero is no longer written as a placeholder performing nothing; the parameter or pin is declared [0..upper], the call is performed, the absence propagates through pins, nested activity outputs and method parameters, and a write of a required feature from a may-be-absent one is guarded with if x->notEmpty(). A send of a signal with a required attribute unset still stands in for itself. The report says per parameter/pin why a value may fail to reach it.
• «Probability» is read by provenance. Only the OMG SysML profile's Probability (recognised by the namespace its application is serialised under, as every standard stereotype is) weights a decision's branches; a same-named stereotype from another profile weights nothing and the report says which profile it comes from: «Probability» from <profile> is not the SysML profile's; its probability is not read. Missing values, unmarked-branch remainders, scaling and out-of-range values behave as before. See the note under Leakage audit on why the gate is the SysML profile's namespace and not SimulationProfile's.
• Opaque bodies: console prints are left out. print(…), println(…), System.out.print[ln](…) statements change nothing of the model, so they are omitted (each noted as an approximation) while the rest of the body is translated; a body of prints alone becomes an empty action. A print whose argument assigns, counts or calls anything but a value-computing function of the table is refused rather than left out; a Java equals counts as such a function only on a receiver known to be a string (a literal, or a feature or local declared String), any other type's equals being that type's own method. Other unknown calls are still refused.
• Run configurations (simconfig.go): the execution target's classifier behavior is resolved through generalizations; a target with no classifier behavior whose parts hold constraint properties is reported per constraint property (which constraint block, whether its rule is a migrated constraint or which call of an opaque rule stops it) instead of a blanket refusal; treatAllClassifiersAsActive = false and autostartActiveObjects = false are typed notes; startTime/stepSize/timeUnit become the sidecar's clockStep.
• A constraint/requirement slot holding a VerdictKind literal is reported precisely as the run's verdict rather than an instance of the slot's type.
• State machines: a transition effect whose body is a console print migrates to do action log and no longer leaves a dangling ; before then.

Result sidecar and comparison (internal/translate/simresults, internal/frontend/repl/compare.go)

• Snapshots are matched to a configuration by the execution target's classifier (or a general/special of it); snapshots of another classifier stored in the same result location are set aside with a note.
• MonteCarloAnalysis summaries (N, Mean, Deviation, OutOfSpec) become the snapshot's statistics of the observable the analysis binds, standing for N runs — deviation and outOfSpec are written only when the snapshot records them, so a summary without a Deviation states none rather than a zero and the disagreement test needs both deviations; a statistic that is no number, or an OutOfSpec that is no count of the N runs, is noted and the snapshot read as an ordinary run; raw values and summaries are pooled by run count.
• Every configuration is compared: one with no stored snapshot prints a tool (no stored result to compare) row; one with no numberOfRuns runs once with a note (the tool's default); refusals name the configuration.
• Two new integrity notes, both typed: a summarising snapshot another configuration stores under the same name with the same statistics is a likely copy; summaries of one observable whose means lie more than three standard errors apart cannot be of runs of one and the same model, and the pooled mean blends them.

Runtime (internal/exec/runtime, internal/ir/lower)

• A case body's timed steps are driven by the clock (analysis_steps_wait_on_clock, action_case_step_waits_on_clock; typed ErrAcceptDeadlock / ErrCaseReadWaits).
• -clock-step <seconds> / %clock-step: a fixed-step clock on every engine, the external-engine protocol (clockStep) and gRPC; witnesses record and replay it. A wait comes due on the first tick not before its instant, within a rounding tolerance capped at a millionth of a tick; a step so fine that the tick count overflows a float64 leaves the instant as it is, and a tick past the last finite instant is a typed error before the wait is queued.
• A state behavior or transition effect written as a bodyless action usage (entry action hello;, do action log) lowers to an empty body and executes as nothing (state_behavior_action_of_no_content).
• A state machine's poll of the signals in flight is memoized until something it could see changes (perf; same traces).

Specification basis

SysML 1.6 §11.3.2.1 (Probability on ActivityEdge/ParameterSet) and §11.3.1.2 (standard stereotypes keyed by profile); UML 2.5.1 §16.2 (an InputPin/Parameter with lower bound 0 admits no value; a CallBehaviorAction with an unset argument still executes) and §14.2.3.4 (an entry/exit/do Behavior of no content); SysML v2 KerML 8.4.4 ([0..1] multiplicity on features), SysML v2 §7.16 (state behaviors as action usages), §7.20 (analysis/verification case steps on the clock). docs/project/spec-compliance.md gains rows for the bodyless state behavior, the case body's timed steps and the fixed-step clock; no row moves down.

How it was verified

Gates, all green locally: make build, make test, make lint, make docs-check, python3 scripts/changelog.py check, python3 scripts/check-doc-ids.py, make man-check, gofmt -l . (empty), go vet ./..., and with the three corpora downloaded OPENSYSML_REQUIRE_TRAINING_CORPUS=1 OPENSYSML_REQUIRE_PILOT_CORPORA=1 OPENSYSML_REQUIRE_PILOT_LIBRARY_XMI=1 go test -count=1 -race ./... (80 packages ok; training_examples_expected.txt unchanged; no ratchet moved).

Second-fixture proof of each mechanism (hand-written, non-benchmark XMI under tests/migrate/testdata/xmi/ or runtime conformance fixtures)

mechanism	fixture (v1 XMI or v2 conformance case)	test
call with a required input receiving no value is performed, [0..upper] propagation, guarded writes; allocated behavior-less call stays unresolved	inline XMI in tests/migrate/arguments_test.go; meter.xmi goldens	TestCallsWithoutRequiredArgumentsAdmitNone, TestAllocatedCallsWithoutBehaviorStayUnresolved, TestConstructFixtures/meter
«Probability» by profile provenance; user-profile same-named stereotype not read; remainder fallback	weighted_decision.xmi (+ goldens)	TestWeightedDecisionReadsOnlyTheSysMLProfilesProbability, inline cases in tests/migrate/behavior_test.go
console prints left out of opaque bodies; a transition effect of a print alone kept as do action log	plant_states.xmi (+ goldens); inline bodies in opaque_migration_test.go	tests/migrate/states_test.go, tests/migrate/opaque_migration_test.go, internal/translate/migrate/opaque_test.go
Monte Carlo summaries as statistics; pooled runs; likely-copy and disagreeing-summaries notes	montecarlo.xmi (+ goldens); sidecars in internal/frontend/repl/compare_test.go	TestMonteCarloAnalysisSnapshotsAreSummaries, TestDisagreeingSummaries, TestComparisonTablePoolsSummarisedResults, TestCompareNotesSnapshotsStoredTwice, TestSummarisedMigrationResultsThroughCLI
snapshots matched by target classifier, another classifier's set aside; inherited classifier behavior; test-case target; per-constraint-block refusal; clockStep from startTime/stepSize	inline XMI in tests/migrate/simconfig_test.go	TestSimulationConfigReportsWhatItCannotRun, TestSimulationConfigPerformsATestCaseClassifierBehavior, TestSimulationConfigNamesTheConstraintsOfAParametricTarget, TestSimulationConfigRecordsTheClockStepOfTheToolsInternalClock
no-snapshot configurations compared; no numberOfRuns defaults to one run; refusals name the configuration	sidecars in internal/frontend/repl/compare_test.go	TestCompareRunsConfigurationsWithoutStoredResults, TestCompareDefaultsToOneRun, TestCompareNamesTheConfigurationItRefuses
verdict literal in a constraint slot	inline XMI in tests/migrate/values_test.go	TestConstraintSlotHoldingAVerdictIsUnmappedAsOne
case body's timed steps on the clock	analysis_steps_wait_on_clock.sysml, action_case_step_waits_on_clock.sysml (+ trace goldens)	TestExecutionConformance, TestExecutionTrace, TestRuntimeRobustnessCaseClockWait
fixed-step clock	clock_step_waits_come_due_on_ticks.sysml (+ trace golden)	TestExecutionConformance, TestRuntimeRobustnessClockStep, internal/exec/runtime/clock_step_test.go, internal/frontend/repl/clock_step_test.go
bodyless state behavior executes as nothing	state_behavior_action_of_no_content.sysml	TestExecutionConformance
memoized signal poll	every existing state-machine trace golden unchanged	TestExecutionTrace, tests/perf

Benchmark model (Cameo Simulation Toolkit model, 44,136 elements) — used only as a test bench

Counts: before 28,626 mapped / 10,319 approximated / 5,191 unmapped → after 28,389 mapped / 10,556 approximated / 5,191 unmapped; validation errors 0 → 0. (275 elements moved from mapped to approximated — almost all input pins and parameters now declared admitting no value, with the reason a value may fail to reach them — and 38 the other way: object flows, decision nodes and run configurations that are now mapped outright.)

-compare-results -seed 1, all 62 configurations (before: 14 compared, 32 silently skipped for lack of a snapshot, 11 refused as "performs no migrated behavior", 5 refused for no numberOfRuns; after: 41 run, 21 refused with a per-configuration reason). Means of the numeric observable(s); n/a = the tool stored no result, so only our statistics are printed.

#	configuration	observable	tool	before	after
1	Acquisition Time	—		refused (blanket)	refused: neither the target nor any general of it has a classifier behavior
2–4, 6	Template Workflow Group 2/3/4/5	Time_Acq_Total	n/a	skipped	17.0
5	Template Workflow Group 1	Time_Acq_Total	18.0	16.6	17.6
7	Template Workflow Group 6	Time_Acq_Total	17.2	16.6	17.6
8	Template Workflow Group 0	Time_Acq_Total	10.8	11.1	12.0 (note: #28's snapshot bears this one's name and statistics, so one may be a copy of the other)
9	Acq Time with IRIS and NFIRAOS Group0	Time_Acq_Total	16.5 (4 summaries, means 13.3–21.3, > 3 SE apart)	21.9	13.8
10	Acq Time with IRIS and NFIRAOS Group4	Time_Acq_Total	n/a	skipped	241.2
11	Acq Time with IRIS and NFIRAOS Group3	Time_Acq_Total	208.4	209.2	222.4
12	Acq Time with IRIS and NFIRAOS Group5	Time_Acq_Total	n/a	skipped	176.2
13	Acq Time with IRIS and NFIRAOS Group6	Time_Acq_Total	210.4	184.8	213.0
14	Acq Time with IRIS and NFIRAOS Group2	Time_Acq_Total	n/a	skipped	109.2
15	Acq Time with IRIS and NFIRAOS Group1	Time_Acq_Total	157.9	150.0	141.6
16	Dither Target Time Group 4	Time_Dither	125.8	92.7	99.2
17, 18	Dither Target Time Group 3/5	Time_Dither	n/a	skipped	195.0
19	Dither Target Time Group 2	Time_Dither	86.5	165.0	195.0
20	Dither Target Time Group 6	Time_Dither	n/a	skipped	99.2
21	Dither Target Time Group 0	Time_Dither	149.0	92.7	99.2
22	Dither Target Time Group 1	Time_Dither	104.9	92.7	99.2
23	Preset with IRIS and NFIRAOS Group 0	Time_Preset_Total	109.0	1.0	74.4
24–27, 29, 30	Acq Time with WFOS Group 1–6	—		skipped	refused: names no execution target; its snapshots are not read (no target classifier they could be of)
28	Acq Time with WFOS Group 0	Time_Acq_Total	10.8 (its one snapshot is named and summarised exactly as #8's, noted as a likely copy)	88.8	104.6
31–35, 37	Preset with WFOS Group 1–6	—		skipped	refused: names no execution target; snapshots not read
36	Preset with WFOS Group 0	Time_Preset_Total	n/a (its result location's one snapshot summarises Time_Acq_Total, not the Time_Preset_Total its analysis binds, so it is of another configuration and set aside)	Time_Acq_Total 0 vs 10.8	57.0
38	APS M3 to APS-TMT Interface Point Alignment Error Analysis	—		refused (blanket)	refused per constraint property (10): one whose rule is migrated as a constraint, which a v2 run checks and does not solve; nine whose blocks state no rule
39, 40	APS Power Roll-up Online/Standby	—		refused (blanket)	refused: the classifier behavior (an interaction whose lifelines/messages are unmapped) is not migrated, so no action is performed
41	APS Satisfy Peak Power Analysis	—		refused: no numberOfRuns	run once (tool's default); snapshot slots hold instances, no numeric observable
42	Test Duration Trade Study	—		refused: no numberOfRuns	run once; runtime error uninitialized feature value: alternatives (trade-study driver reads a feature nothing sets)
43	APS Post Segment Exchange Duration Analysis	PhasingDit/RBDit/offAxis…	20/45/7/6	refused (blanket)	20/45/7/6 (4 of 8 match; tBB1/tFinal/tM3Align/tNB_Results 0 vs 612/3712/67.75/56.5)
44, 45, 47–50, 52, 54	APS Off-Axis / M3 Alignment / Warping Harness / Self Test / M1CS Sensor / Gravity / Pupil Registration / Template duration analyses	—		refused (blanket)	run once; snapshot slots hold instances (aPS Mission Logical, analysisDriver), no numeric observable
46	APS Maintenance Alignment Duration Analysis	PhasingDit/RBDit/offAxis…	20/45/7/6	refused (blanket)	20/45/7/6 (4 of 8 match; t*_Results 0)
51	APS Wavefront Calibration Duration Analysis	—		refused (blanket)	run; runtime error: a lookup-table action's result gets 0 values for a [1] pin
53, 56	APS Reference Point Tracking / Pre-Session Calibration	—		refused (blanket)	run; runtime error: transition effect Send Ack waits for the clock, which only a do behavior may
55	unnamed	—		skipped	refused: names no execution target
57, 58	APS interactive silent/animated	—		refused: no numberOfRuns	run once; accept deadlock (nothing posts Control::finish; the tool's run is interactive)
59–62	Test Run Case 2–5	—		refused (blanket)	refused: treatAllClassifiersAsActive = false has no v2 form (typed note); the target's classifier behavior is a state machine, which a run performs as no action

Residuals closed (mechanism in parentheses): 1 result-location wiring (snapshots matched by the target's classifier and by the observable the analysis binds; a snapshot of another observable or a copy of another configuration's is a typed note rather than a comparison of the wrong number); 2 identical dither groups — partly (the calls whose required inputs get no value are now performed, so Group 2 differs from 0/1/4; 0/1/4 remain identical, see below); 3 the +33 % IRIS/NFIRAOS Group 0 gap (calls with unset required inputs performed; the tool's four stored summaries for this configuration disagree by more than three standard errors, which the harness now says); 4 blanket "performs no migrated behavior" (per-constraint-property report, inherited classifier behaviors, interaction/state-machine classifier behaviors named); 5 no numberOfRuns (defaults to one run with a note); 6 no snapshot (run and printed with a no stored result row); 7 opaque ALH.*/console prints, bodyless state behaviors, treatAllClassifiersAsActive, VerdictKind slots (generic fixes or precise typed notes).

Remaining, with cause:

• Dither Groups 0/1/4 (99.2 vs 149.0 / 104.9 / 125.8) and Group 2 (195 vs 86.5, one run) — statistical and source-side: every stored dither summary is of 5 runs with a standard deviation of 34–94 s (standard error 15–42 s), and the tool's own two summaries of one group differ by as much as the groups do (Group 1: 92.8 vs 117.0; Group 4: 97.2 vs 154.4), so the stored results do not resolve the groups. Our three groups coincide under one seed because the migrated dither behavior reads no property the groups set differently; whether the source sets one through a construct not migrated is not established and is left as an open question rather than guessed.
• Acq Time with WFOS Group 0 (104.6 vs 10.8) and Preset with WFOS Group 0 — source-model defect: the Acq configuration's result location holds a snapshot named and summarised exactly as Template Workflow Group 0's (reported as a likely copy), and the Preset configuration's holds a summary of Time_Acq_Total rather than of the observable its analysis binds (set aside, reported); neither target has a stored result of its own.
• Acq IRIS/NFIRAOS Group 0 (13.8 vs 16.5) — the tool's four summaries lie between 13.3 and 21.3 and disagree by more than three standard errors (reported); ours lies within their range.
• t*_Results of the APS duration analyses (0 vs 612/3712/…) — runtime gap, reported per action: the duration comes from opaque JavaScript actions beyond the translated subset (ALH.inState, ALH.createList, lookup-table reads) which stay as refused opaque actions.
• Send Ack transition effects (Reference Point Tracking, Pre-Session Calibration) — no v2 equivalent: a transition effect cannot wait on the clock (only a do behavior may); typed runtime error.
• Trade study and interactive configurations — a trade-study driver reading a feature nothing initialises, and interactive runs that wait for a message a person sends; both by design of the source.

Leakage audit

rg -n -i 'tmt|Time_Acq|Time_Dither|IRIS|NFIRAOS|WFOS|ProbabilityAcq|Retries|GS_Found' internal --glob '!*_test.go' --glob '!**/testdata/**' returns only generic hits: stmt/Stmt (statement identifiers in codegen, migrate, runtime), IRIs/iris (RDF IRIs in translate/rdf, export, interop) and the English verb "retries" in four comments. No benchmark element, package, observable, property name or numeric shape appears in non-test code, docs or changelog.

«Probability» provenance: in the benchmark model all 83 «Probability» applications are serialised under the OMG SysML profile namespace (http://www.omg.org/spec/SysML/…/SysML), none under SimulationProfile — Probability is a SysML 1.x standard stereotype, not a Simulation Toolkit one, so the gate is the standard-profile namespace (the same check the migrator applies to «Block» etc.), not isSimulationProfile, which would have dropped every weight. With the gate in place the model's 187 @Stochastic::Probability annotations and its report entries are byte-identical to the pre-gate output (0 foreign-profile notes); the foreign-profile case is proven on weighted_decision.xmi, where a same-named stereotype from http://example.com/routing/profile is not read and its edge falls back to the remainder weight.

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