ARCADIA-Studio/harness/release-completeness-audit-20260906.md

Independent release-completeness audit — 2026-09-06

Read-only requirements/code review of the working release tree after source6, artifact review and Linux journal integration. Authority: the supplied 10b0f701-1743-4de7-9ee1-225920e19649/goal-objective.md, docs/language.rst, docs/scope.md, harness/PLAN.md, selected DECISIONS.md contracts and current evidence. No production changes, Cargo, VM launch or native input were performed. The older coverage audit is a requirements map; its statements that communication means, selective views and review controls are absent are historical, not current findings.

Conclusion: this review did not establish another missing OA/SA core concept in the selected R1–R6/source6 profile. It did find an unfinished required teaching/example deliverable and a narrower behavior-export gap. Neither the implemented catalog nor this inspection proves full release acceptance. Current platform/runtime/package gates remain real obligations; they must not be recast as optional limitations.

Requirement → realization → remaining gap

Requirement	Current implementation and bounded evidence	Finding
OA responsibility; SA system boundary; meaningful nesting	model.rs, validation.rs, projection.rs, ports.rs; separate work ownership, leaf allocation, roles and one SA System. Prior architecture/port native records.	No new absence found. Internal logical/physical components belong to the explicitly deferred layers.
Missions and reusable capabilities	Typed include/extend/generalization, mission involvement and exploitation in model.rs:847, validation cycle/condition rules and workflow.rs.	Earlier R2 absence is resolved. Finish the integrated engineering example rather than add duplicate concepts.
Ordinary OA participant communication	RelationBody::CommunicationMean, allocation validation, pure resolve_message_endpoints (validation.rs:1294), explicit SA port decisions in transitions/communication.rs; source6 core/HTTP tests.	Implemented. Current native CM/transition proof is root-owned work, not missing semantics.
Contextual operational processes / functional chains	Occurrence Work/Control/SequenceUse plus SequenceLink; behavior/process.rs, direct canvas events and shared validation.	Branch/reuse/control semantics are present. Derived layout is an explicit design choice; historical static-view criticism does not describe this implementation.
Scenarios with reusable, typed interactions	OccurrenceBody, explicit viewpoints, correlated Call/Reply, qualified values and bindings; behavior/scenario.rs and editable Scene.	No new absence found. Free lifeline-column positioning is deliberately outside the current persisted scenario layout (release-scenario-editing.md:12–20); do not mislabel it missing message/reorder authoring.
State/mode behavior and typed events/actions	state_validation.rs, behavior/state_canvas.rs, state_edit.rs, state_view.rs; distinct pseudostates, nested regions, internal/local/external transitions and maintained action links.	Core and editing exist. Faithful behavior-image export is incomplete; see F2.
General-system data/interface semantics	data.rs, ports.rs, forms/data_fields.rs, data_values.rs; bounds/defaults, physical nature vs mechanism, record/reference storage, qualified fields, interfaces and protocols.	No new required type-family absence found. Executable expressions, object-instance defaults and a units conversion engine were explicitly excluded in the selected data contract; their absence is not a newly discovered blocker.
Bidirectional, iterative semantic reconciliation	transitions.rs, closed semantic/*, desktop semantic_review/*; scalar/constructor/keyed/order, Child/Incident, artifact lifecycle and combined requests. Core absence oracle, GUI tests and recent exact source/history native records.	Former prose-only and missing attachment/artifact/multi-source claims are obsolete. Remaining native cases and corrected diagnostic wording are acceptance work, not absent adapters.
Saved readable views	view_options.rs, projection/reference lifecycle, forms/view_fields.rs, duplication command and diagram canvas; category/coverage/context/hide/label settings. Forty-exchange native fixture and pure/HTTP evidence.	Implemented; grouping was conditional on failure of the dense-view gate, which did not occur. Do not add artificial summary exchanges as an ungrounded requirement.
Actionable completeness checks	validation/coverage.rs, shared behavior evidence, source locations and REST diagnostics. release-rest-coverage-verification-20260906.md.	Non-leaf exchange and behavior/mission/capability warnings are implemented. Do not reopen the old task3 absence claim.
Equivalent graphical/textual models, modular Git-friendly source	Normative source6, parser/updater, references, frozen older admission, typed forms, source workspace/lint/highlighting and coupled Document history. Exact Save-as and source5 checkpoint/native oracles.	No newly proven missing persistent-field editor in this bounded inspection. A closed core catalog alone does not prove all GUI parity; final-tree cross-representation acceptance remains required.
Arbitrary artifact traceability and deployable automation	Artifact metadata and owner lifecycle; shared REST commands/queries, source-bundle envelopes, ETags and hosted desktop adapter. Actual HTTP48 and native hosted evidence.	Opaque references intentionally do not execute/fetch CAD/code/documents. Plugin ABI, replicated collaboration and remote deployment execution are not implied first-release requirements. Linux full application verification remains open.
Intuitive onboarding, guide and worked engineering scenarios	Home/recents, searchable Rust guide, recovery, source diagnostics; Home opens the rescue demo. Existing teaching examples contain useful OA/SA/ports/state/artifact work.	F1: the expanded first-release worked project is not shipped. Current examples do not teach the newly required source6 engineering chain.

Actionable implementation/deliverable gaps

F1 — Required release teaching project is missing (P1 deliverable, not a missing metamodel). PLAN milestone2 requires a worked example for each addition, and milestone5 requires guided OA goals/roles/process alternatives through SA contracts/chains/scenarios/states/evidence. The shipped examples/coastal-rescue.arcadia, routing-workshop.arcadia and source-project/project.arcadia still use arcadia 3; old version alone is not a defect, because these are valuable migration fixtures. The stronger finding is content: examples/ and the Home demo builder crates/oa-core/src/demo.rs contain no CommunicationMean, ExchangeCategory/view options, SequenceLink/reusable process contribution, or capability include/extend example. Their useful demonstrations currently live in tests and harness/fixtures, not an ordinary user-facing model. Home's Explore rescue example (home.rs:311) cannot expose that missing content.

Add a separate current-profile worked project and concise tutorial mapping. Preserve frozen legacy examples. It should connect one coherent OA mission/process and participant communication to explicit SA boundary-port/functional counterparts, an alternate/reused behavior path, a bounded data contract, state behavior and an artifact. Provide two saved views of the same identities and one deliberate nonblocking incompleteness warning. Reopening and one reviewed change should produce documented source/REST results. Root's active guide/CM work is the natural integration point; this audit does not assert that its in-progress changes are finished.

F2 — Shareable behavior diagrams are not implemented faithfully (P2 product/scope gap). diagram.rs:1547 exports the generic projected architecture view. Its node emitter always writes rounded rectangles with kind/name; it has no StateKind glyph or typed StateAction/internal-transition row handling. The state renderer instead provides these in behavior/state_canvas.rs:326–413 and its state drawing functions. behavior/state_view.rs:85–104 offers views/Fit/zoom but no export action; process/scenario Scene likewise has no vector export path. Thus the guide's general report-image wording (guide.rs, Files) is broader than the actual behavior output capability. This is an inspected path difference, not a native export failure reproduced here.

The original objective does not independently mandate SVG for every scene, so this is not automatically a release-blocking new requirement. Root should explicitly settle the boundary: preferably reuse the existing scene/geometry semantics for faithful state/scenario/process export; otherwise clearly scope the current command to architecture views. Do not present a generic state rectangle as an equivalent exported state machine, and do not invent new semantic records to solve an output-adapter problem.

Three highest-value next gates

1. Complete and exercise the current-profile teaching project. This closes F1 while testing several features together in an engineer's actual OA→SA→review workflow, rather than adding disconnected demos. Preserve existing migration inputs.
2. Resolve behavior-output scope, then finish the end-to-end workflow matrix. Close F2 with an explicit decision/implementation and validate the remaining root-owned CM, behavior, hosted and recovery cases on one traceable tree. Do not treat old executables or unrelated fixtures as current acceptance.
3. Verify the complete Linux application and final distributable tree. The recent Ubuntu/Rust1.90 core38 baseline, patched42 library tests and three executed syscall injections are valuable but do not compile or run the desktop/server. Follow the measured server-first plan in release-linux-validation-options.md, then actual Linux GUI; finish final locked suites, strict checks, rendered reST/OpenAPI, independent review and clean-environment packages/checksums.

Accepted non-goals remain LA/PA/deployment modeling, execution/simulation, native Capella file compatibility, full SysML conformance, semantic Git merging/LSP, plugin ABI, production multi-user synchronization and signed installers. These boundaries are supported by the objective's later-layer deferral and selected contracts; they do not excuse missing primary OA/SA workflows, user-facing examples or required Windows/Linux acceptance.

Follow-up decisions and implementation evidence

The findings above describe the audited tree before this follow-up. F1 now has a separate source-6 project at examples/release1/project.arcadia, with a guided engineering walkthrough and local evidence artifact in that folder. The shared Rust generator (crates/oa-core/examples/generate_release1.rs, examples/release1/model.rs) executes the real OA→SA planner with explicit port decisions, then completes the interface/port allocations and saved views. It emits 90 elements, 60 relations, 5 views, 65 traces and 28 source files. Existing legacy examples are untouched.

crates/oa-core/tests/release1_teaching.rs passes 4/4: deterministic shipped-source/interchange parity and the one deliberate located warning; Call/Reply direction, process/scenario reuse and stable filtered-view port geometry; missing port decisions and invalid performer atomic rejection; and a real reviewed Name change preserving unrelated modules. Generator, focused tests and strict Clippy passed under the current Windows gnullvm Rust1.98.1 toolchain, recorded in .local/release1-teaching-{generator,tests,clippy}.log. This closes the absent worked-project implementation finding. The Home/guide hook, packaging and native walkthrough are integrated/verified separately by their owners; these core checks do not claim their acceptance or a Linux/full-service run of this particular example.

Root explicitly selected the narrow F2 boundary: current export is Architecture SVG; faithful process/scenario/state-scene SVG is deferred beyond this first-release export scope. The new tutorial states this boundary. F2 is therefore a scope/wording qualification, not an invented blocker or a claim that generic state rectangles reproduce the interactive behavior scene. Final guide/package wording remains part of root's release review.

Independent teaching review then caught an overbroad assessment-capability context on the repair conversation. The corrected lesson scopes that conversation to restoration and gives assessment its own single-action procedure and ChainInvolvement, reused by the main process. This preserves assessment coverage without hiding extra warnings. Current generated counts are 94 elements, 62 relations, 5 views, 68 traces, 30 source files; the single audit-function warning remains. Expanded four tests passed using a bounded direct rustc -D warnings compilation against the already-built Windows core (no new Cargo build), including both-layer context/procedure assertions and exact decoded equality of all shipped replay/diagnostic JSON. Logs: .local/release1-teaching-corrected-generator.log and .local/release1-teaching-corrected-tests.log. Subsequent integrated Cargo/native/package checks must use these corrected files. The contemporaneous Linux server gate deliberately retains its separately hashed earlier snapshot and does not imply verification of this later example correction.