[avrabe]

hw : refined to system platform.single in an extending implementation
instantiated BESIDE the inherited hw : system platform instead of
replacing it. On main, DigitalControlSystem DCS.dualtier2 has three app
and three hw, one per extends level, where OSATE's own instance model
has one of each. 61 of the 2086 corpus files refine a subcomponent. Every
analysis that walks children saw the phantoms, and REQ-AADL-REF-CONTEXT-001
needed a resolver allowance to step around them.

Semantics, from OSATE source at osate2 @ 87468992:

• ComponentImplementationOperations.getAllSubcomponents removes the
  refined original and appends the refinement, which takes the end
  position;
• SubcomponentImpl.getAllClassifier: a refinement without a classifier
  keeps the original's;
• SubcomponentImpl.getPropertyValueInternal: the refinement's values,
  then each refined original's, the first found winning.

collect_impl_chain_recursive now does the first two. The builder records
each refinement's refined-away originals, and one helper,
push_subcomponent_associations, used by all three property-map builders,
applies the originals' associations before the refinement's own.

Oracles:

• OSATE instance models (test-data/osate2/instances, 11 *.aaxl2),
  comparing component counts: main matches 7 of 11 and DCS.dualtier2 has
  6 against OSATE's 2. The fix matches 8 of 11 and DCS.dualtier2 has 2.
  The other three rows are unchanged and out of scope.
• unit tests: replacement (one hw, classified Plat.single, with its
  cpu); inherited and overriding properties; a refinement without a
  classifier keeps the original's. Hand mutants (no replacement, a plain
  replace without the property chain, no classifier fallback) are all
  caught.
• corpus, main b883cac vs this build: the 61 refining files plus 15
  controls, 285 runs, 283 instantiating, 106 differing, every line
  classified. Removed: 264 subcomponent_rules, 112 completeness and 66
  binding_rules ERRORs plus their warnings, all from the phantoms. Added
  ERRORs are confined to two models. resourcebudgets: deadline misses and
  overloads that are TRUE findings from its own timing (47 ms / 20 ms =
  235% on hw.cpu; 34/20 = 170% on pc2), masked while the phantoms broke
  resolution. CasePositionControl: "not computed" becomes a below-minimum
  ERROR on an incomplete best case, filed as Latency below-minimum Error fires on an incomplete best case (components without timing) — '0.000 ms is below the declared minimum' asserted from missing data #516. weight_power on
  resourcebudgets SCS: 3.20 kg becomes 1.70 kg, because the phantoms were
  weighed twice.
• workspace: 3334 passed, 0 failed.

The resolver's latest-first same-name sibling allowance (#497) is
removed. With it still in place, every test passed under a first-match
walk, so nothing exercised it once phantoms stopped existing. Its test
now checks that a reference resolves THROUGH a refinement, and
REQ-AADL-REF-CONTEXT-001's text is corrected.

rivet: REQ-AADL-SUBCOMPONENT-REFINE-001 (v0.47.0, verified), traces-to
REQ-AADL-REF-CONTEXT-001.

NOT CLAIMED: refined connections, flows or features; in_modes and
array-dimension inheritance across a refinement; a diagnostic for a
non-refined same-name redeclaration (still appended, as before);
SCS.Phys's component count, which is lower than OSATE's.

Refs #496, #516

🤖 Generated with Claude Code

https://claude.ai/code/session_01BnBWxgVEUPdsBTLNKtrA6Z

[avrabe]

Count correction: 107 runs differ, not 106. One run, multi-core/arinc653-p4080.aadl arinc_example.error, contains a NUL from #507's wrpc message (fixed in #518), so diff -r reported it as Binary files … differ and my header count missed it. Read with diff -a, it is the same phantom-removal class as the rest: the duplicate cpu and its diagnostics are gone, 13 → 12 components. The added-ERROR classification is unchanged.

[github-actions]

Rivet verification gate

✅ 20/20 passed

	count
Passed	20
Failed	0
Skipped (no steps)	0

Filter: (and (= type "feature") (or (has-tag "v093") (has-tag "v0100")))

Failed artifacts

(none)

_{Updated automatically by tools/post_verification_comment.py. Source of truth: artifacts/verification.yaml.}

[codecov]

Codecov Report

❌ Patch coverage is 99.25373% with 1 line in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
crates/spar-hir-def/src/instance.rs	99.24%	1 Missing ⚠️

📢 Thoughts on this report? Let us know!