[devin-ai-integration]

What and why

A feature read at namespace (model) level now holds its value in the feature's effective multiplicity, the way the same feature read on an object already does. On develop, with

package P {
  private import ScalarValues::*;
  attribute xs : Integer[0..*] = 4;
  part def D { attribute xs : Integer[0..*] = 4; }
  part d : D;
}

sysml -e P::xs printed = 4 while sysml -e P::d.xs printed = [4] (same in the REPL): a [0..*] feature holding exactly one value read bare at model level and as a one-element sequence on an object. P::ys = (1, 2) → [1, 2] and an empty [0..*] → [] were already right, so the one-value case was escaping the shape rule: a declared value was evaluated and returned as written, and declaredCollection only reshaped values that were already collections.

Found while triaging #1575; this is a separate, small inconsistency and does not address that issue.

The fix is in the runtime read path, not the formatter (runtime.FormatValue and the REPL's formatValue are untouched):

• Context.admitted's shape logic is extracted into Context.heldShape(feat, val) — one predicate: an at-most-one feature holds a one-element collection's element bare, a multi-valued feature holds a collection however the value was written, with the element charging and set/sequence canonicalisation admitted already did.
• Context.declaredCollection (the namespace-level read of a declared value, reached from conformHeld and denotedValue) now builds the EffectiveFeature for the declaration and applies that same heldShape, so -e, REPL bare expressions and %eval, gRPC Evaluate and MCP all read [4].
• Consequence handled: a parameter stating no multiplicity is held to [0..*] (semantics.EffectiveParameterRange), so a calc-typed parameter's default (in calc f = twice, the library's in calc :>> eval = evaluationFunction) is now a one-element sequence holding the function. A call applies the one function such a feature holds: boundFunction and declaredFunction take soleElement of the read, as evalChainInvocation and invokeFunction already did.

Arithmetic and comparison over such a read are unchanged (P::xs + 1 → 5, P::xs == 4 → true).

Specification basis

KerML 1.0 §7.3.4.1 and §7.4.12: the values of a feature are a sequence its multiplicity constrains. SysML v2 §7.6.3: a parameter stating no multiplicity has no implied 1..1. The "performed action's parameter" row of docs/project/spec-compliance.md is extended with the namespace-level read (implementation and tests); its status is unchanged.

How it was verified

Fixture first, both failing on develop with 4 / int_value and passing here:

• internal/frontend/repl/namespace_shape_test.go TestEvalNamespaceReadHoldsEffectiveMultiplicity: %eval and the bare expression agree for [0..*] valued 4 ([4]), the same feature on an object ([4]), (1, 2) ([1, 2]), empty [0..*] ([]), [1], [0..1] and bare attribute x : Integer (4), [2] ([4, 5]), [0..*] valued (4) ([4]), P::xs + 1 (5), P::xs == 4 (true).
• tests/grpc evaluate_namespace_feature_multiplicity (P::xs → sequence[int_value 4]) beside evaluate_object_feature_multiplicity (P::d.xs, same).

Moved expectations, each adjudicated:

File	Movement	Category
internal/exec/runtime/robustness_usage_members_test.go (Probe::rig.fits.limit, a constraint's bare in limit : Real = 100.0)	100.0 → [100.0]	same value wrapped in a one-element sequence (a bare parameter is [0..*], as the existing performed-pin rule holds)

No golden, trace or expected.json file moved: go test -count=1 ./internal/exec/runtime is green without re-recording (the trade-study conformance cases and function_value_qualified_default, which call a calc-typed parameter by name, pass through the soleElement consumer fix above, with their recorded traces unchanged). No codegen target is touched.

Gate, all local on this head: go build ./..., go vet ./..., gofmt -l . empty, go test -count=1 -timeout 40m ./... green (122 packages, with the training and pilot corpora downloaded so TestTrainingExamples and TestPilotCorpora ran rather than skipped; tests/grpc, tests/codegen and the docs transcript checks are in that run), python3 scripts/changelog.py check, python3 scripts/check-doc-links.py (0 broken links). The quick CI tier's runtime, model/export/graph, static, WebAssembly, conformance and documentation jobs are green.

Merged develop (e19d9beb3) in with merge commits, never a rebase; the docs/project/spec-compliance.md conflict took develop's refreshed citation column and kept this PR's row content, so the file differs from develop by that one row alone. Re-ran the gate on the merged head: go build ./..., go vet ./..., gofmt -l . empty, go test -count=1 -timeout 40m ./... green (123 packages, corpora present), go test -C tools over the census and referee packages, check-doc-links, check-doc-ids, check-doc-figures, changelog.py check, mkdocs_census-test.py; sysml -validate -format sarif examples over the tracked examples exits 0 on this head.

Dry-merged clean against the open runtime siblings #1571 and #1572 and against #1580.

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 gate count moved
• No internal work-item labels (waves, slices, F4, K5) in the body, docs, or changelog

Link to Devin session: https://nasa-jpl-demo.devinenterprise.com/sessions/4b614bc5d6204ee693145e91142acb0f
Open in Devin Desktop: https://nasa-jpl-demo.devinenterprise.com/desktop/session/4b614bc5d6204ee693145e91142acb0f?variant=devin
Requested by: @HuiJun

[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