ARCADIA-Studio/harness/STATE.md

First release state — R4 authoring / native iteration / R5 integration

Objective and baseline

Active goal: feature-complete OA + SA/Architecture first release, beyond V2; objective attachment81e84f53-2b51-4151-9e34-59414bda8b2d. User demands closer original Capella study and practical usability. Later layers remain future work. V2 source ZIP and dist packages preserved; do not mark release complete. PLAN.md is current milestone/acceptance source.

Verified implementation

• R2 missions/capabilities, R1 process topology and R4 declared scenarios/message semantics implemented in shared core; exact handoffs release-capability-core.md, release-topology-core.md, release-scenarios-core.md.
• Fresh independent release-semantic-review.md findings CLOSED: branch-local duplicate replies/use-after-delete, Role compatibility, loop repetition diagnostics, reverse Reply picker and state-owner picker. Independent27 R1/R2/R4 tests + compiled external repro passed. Core worker71 focused regressions + strict Clippy passed before R5. Loop warnings distinguish uncertainty from execution claims.
• R1 contextual tools now work from empty and legacy Process/Chain; preserve legacy linear interpretation and perform explained atomic path conversion. Root added measured Fit/zoom and scaled hit testing after native excessive-scroll finding; empty process now has guidance. Latest source not native reverified yet.
• R4 dedicated scenario.rs authoring supports typed lifelines, all message styles, paired Call+Reply, explicit bound ScenarioUse, fragments/operand destinations/actions/state observations, carried values. Core helpers drive endpoint/viewpoint compatibility. Worker six scenario tests; root56 full desktop tests passed session50129 before source-find addition.
• Root R4 property forms include typed message/payload/reuse/binding fields and new constructors. Shared core reply/role helpers; distinct state-owner/effective-performer rule restored after review and tested. Modeling actions picklist near top of capability/mission properties; Connections inspector traverses real core references and diagram placements.
• Source lint runs bounded max1worker/latestversion, caches accepted model checks, rejects stale diagnostics across reset. Source navigation uses layout sensor; root further added focus, declaration context and literal Find/Go to line (CtrlF/CtrlG/F3/ShiftF3), UTF-8 byte cursor offsets, invalid-draft preservation. Latest59 desktop test run session22577 currently building after tests; inspect result before claiming complete.

Current native verification

Root exclusively owns Computer Use. Sky persistent bindings sky/studioWindow/studioState/studioDialog/dialogState. Refresh inventory before next action. CUA through node_repl @oai/sky skill26.901.41600, not disabledcua_repl. Native file dialogs sometimes return tiny16px screenshot first; use screenshot width>100 for coordinates. One observed action then fresh capture; typing after focus observation.

Verified checkpoint executable: D:/CodexBuilds/Capella2-release1-01a06e2e/ui-r1-r4-checkpoint/arcadia-studio.exe SHA256B24611C9AA7BF686BDA3FDD0B25BB0CE71715D5FED471287ECBAE2B0BF385F98; window395472. This predates scenario authoring, topology zoom and Find. AltF4 requested after SaveAs, fresh inventory needed to confirm closure. stderr empty and process responsive during run.

• Native1440x900 and1100x700: guide/search, persistent task/explorer navigation; selected capability correctly identifies placement; Modeling actions menu appears near top. SourceForSelection now reveals correct declaration, but it lands near viewport bottom without focus; root further focus/context patch pending native verification.
• Help→resize→Open saved capability model now automatically fits47% with all nodes visible: original cache/layout Fit bug fixed and native reverified.
• Connections outgoing extension→target capability→incoming capability/chain relationships works. Right panel retains connection tab during creation, hiding name fields; root now switches new element to Properties and added EditProperties(id) message for explicit property actions. Pending native check.
• Created empty OA Process Choose rescue response, involved Assess incident through +Work, used Branch/parallel to create Report distress / Respond to distress alternatives after it, condition Incident needs more information + else. Observed5newSequenceLinks,4branchnodes, rendered guard/alternative routes. At minimum width excessively horizontal graph; root now adds Fit/zoom (not yet native checked). Branch form scroll works and commit is atomic via same Command path.
• Saved full result through native SaveAs to ui-r1-r4-checkpoint/process-branch-ui.arcadia. On-disk source verified process name, exact guard and5SequenceLinks. No original fixture overwritten.
• Earlier R2 native extension fixture ui-r2-checkpoint/capability-extension-ui.arcadia preserved: Rescue in heavy weather extends Protect life at sea, condition Sea state is 6 or higher; exact source verified. Prior app closed.
• Original Capella six native benchmark workflows recorded in release-capella-workflow-benchmark.md. Original window524804 disappeared from fresh inventory this turn; do not assume open. Disposable workspace/sample under D:/CodexBuilds/Capella2-release1-01a06e2e; preserve it. Reopen if more comparisons needed.

Active ownership

• Root: desktop app/forms/views/workflow/navigation/guide/source_navigation, topology.rs (released by worker after wrapper edits), state docs, exclusive CUA. Next: finish current59test/build result, copy new checkpoint, native Fit/zoom/sourceFind/scenarioauthoring/save-reopen; strict desktop Clippy. Root basicR5forms enums/refs/Event constructors already added; ergonomic state workflow still needed.
• release_capella_benchmark: ONLY behavior.rs + behavior/scenario.rs and benchmark docs. R4 authoring code complete, final focused checks/report then release. Root owns topology now. Shared behaviorEvent::is_submit guards both editors; root integrated. Event routing explicitly tags scenario commit so viewport events cannot redirect it.
• release_capability_core: R5 core model/validation/private state_validation.rs/references/session/transitions/projection/demo/coretests(exceptlanguage). Exact adopted contract release-state-contract.md: Event Signal/Time/Change, After/At, External/Local/Internal, typed trigger/effect refs,9newpseudostates, is_stable_state. Events/refs and state module compile; hierarchy/cardinality/focused tests underway. Source factory defaults preserved; Project::relate/Command::Connect use Local from EntryPoint. No desktop ownership.
• release_text_service_fix: language/persistence/tests/server/API/OpenAPI/language.rst. R5 source/API landed; previously16language+17HTTP/config+1unit passed, reST warning-free. Await core finalfreeze for final focused/strict checks; may read-only draft R3contract while waiting. R6 read-only report done: release-data-contract.md; not implemented.
• release_semantics_review: idle, focused review closed. Reuse after R5 / current graphical changes for fresh review.

Required remaining work

R5 ergonomic state/event authoring/rendering plus finalcore/source verification; R6 data/contracts per design (lead still must freeze decisions); R3 fullsemantic drift and mapped existing-counterpart review; categories/persistentfilters/coverage; home/recent/recovery; optional hosteddesktop ETagopen/publish/save; explicit schema3 losslessV2migration; workedexamples and currentdocs; independent finalrequirements/correctness/UI/simplicity reviews; Windows/Linux fullchecks/nativeflows/packages. No V2 exclusion substitutes for release acceptance.

Environment

Windows Rust1.98.1, target D:/CodexBuilds/Capella2-01a06e2e. Build cargo build --locked -p oa-desktop; tests cargo test --locked -p oa-desktop. Copy checkpoint exe before native launch so further builds can link. WSLUbuntu24.04 config per verification-v2-linux.md. Git unborn/untracked preserve. No disk blocker. Prior target/debug deletion rejected: do not retry/bypass. FinalMSRV/platform/clean-workflow checks remain.

Usage-limit interruption recovery

All three workers stopped with explicit account usage-limit errors; no worker is still active. Root owns recovery/integration of their current files. R5 core+server full locked tests passed in root session11632 (exit0). Worker left behavior.rs declaring states module without writing states.rs; root is completing that module before the next build. Native currentcheckpoint disappeared from process/window inventory without stderr output; root relaunched preserved ui-scenario-checkpoint executable, new returned window1050736. No unsaved native work was lost (last work was only reopening saved fixture). Root just verified Find/Enter/F3/CtrlG+line1 with actual selection and minimum1100x700layout; Fit/scenarioauthoring remain next native checks.

User correction: diagram hierarchy and ports — current priority

User reports architecture diagrams have become generic graphs and port semantics/rendering differ dubiously from Capella. Lead confirmed projection includes almost every relation between placed objects, Overview flattens every parent, and renderer has no boundary-port anchors. Reopening original Capella to inspect SAB/SDFB/OAB ownership and port workflows before freezing corrective contract. Preserve completed state editor work; state native verification is pending, not release complete. Current desktop:59 prior full regressions +4 new state authoring tests pass, cargo check passes after recovery; strict Clippy/current full63 tests not yet run. New state_scene.rs gives actual nested geometry, internal transition rows and vector pseudostate symbols. No fresh agents available: three existing agents still errored on account usage limits; root continues.

Git adoption and hierarchy checkpoint

User explicitly authorized Git and optional worktrees. Existing .git was unborn; created codex/release1-engineering, added repository ignore/line-ending rules, and preparing the first source checkpoint. Future isolated implementation can use worktrees after this commit; read-only reviews do not need them. Nothing pushed or published.

Hierarchy correction implementation: dedicated FunctionalDataflow/Capabilities projections, architecture filters participants/work + exchanges, six focused demo views with architecture preferred on open/layer change, calculated boundary ports referenced by FE/BoundaryExchange, shared movement/hit/SVG anchors. Layout measures port room and 64px headers, orders root containers by actual exchange flow. Source/OpenAPI enum support updated. Functional/component port metamodel correction remains required; do not mistake improved rendering for semantic completion.

Verification: full locked core+desktop+server tests exit0 root session12111; desktop64 tests, new3 core projection regressions and1 desktop port hit/preview/SVG test. Earlier new group-movement test incorrectly used projected origin as stored MoveNode target; corrected to command's absolute persisted-coordinate contract, and shared layout/header measurement now avoids the native offset mismatch. Strict workspace Clippy passed before projection changes; rerun needed on this checkpoint. Native checkpoint ui-hierarchy-checkpoint SHA256D6055F659754E9159232483EAF9FA595C1D7F53683F8FB618D855F04B9C931A5 built/copied; native launch/verification next. Capella graphical window395654 remains a splash, console launcher14556 unresponsive; no new native port study claimed.