Files
ARCADIA-Studio/harness/release-engineering-guide-audit-20260906.md

7.8 KiB

Built-in engineering guide: current workflow audit

Audited the 16 existing offline topics against the user objective, current controls, normative language contract and committed checkpoint 2b2d2bc. Expanded the guide to 18 navigable/searchable topics by separating semantic review from counterpart creation and engineering checks from crash recovery. Only guide.rs was edited; the recent Fill constraints on search, article content, step rows and Clear control remain intact. No new dialogs, model mutations or application navigation paths were introduced.

Existing-topic assessment and changes

Existing topic Assessment and bounded correction
First model Creation tutorial was useful but stopped before a reviewable OA→SA baseline. Added mission/capability→operational behavior→reviewed system contribution→contracts/states/evidence→checks/source milestone.
OA and SA Retained explicit need versus system contribution and later-layer boundary. No claims of full Capella/SysML conformance.
Participants, roles and work Retained direct performer versus role responsibility and work decomposition.
Ownership/allocation/views Existing independent placements, collapse, routing, manual ports, SVG and hide/delete distinction remain accurate. Selective filters/categories remain a later integration section below.
Missions/capabilities Replaced generic scenario-context prose with the actual Scenario Context property and explicit Chain involvement workflow.
Exchanges Scoped typed component-port/Boundary drawing to SA. Explained current Legacy OA compatibility and that ordinary participant Communication means are not yet offered.
Processes/chains Replaced stale +Work/+Exchange/Connect instructions with actual Work, Exchange, Sequence link, Select / reorder, Two paths… and imported ordered-path conversion controls.
Scenarios Added Draw message source-position→receiver gesture and selected ↕ reorder handle; retained precise forms, fragments, reuse and correlated call/reply. Clarified occurrence-specific values versus shared exchange definitions.
States/modes Added selectable typed Entry/Do/Exit rows without fabricated containment and a timed transition editing recipe preserving existing triggers/guard/effects.
Data/interfaces Existing bounded primitives/enums/records, multiplicity/defaults, inheritance/reference recursion, parameter phase, identified interface allocations and port contracts retained.
OA↔SA counterparts Removed stale “ordered collections unfinished” statement. Kept creation/scope/reuse distinct from typed drift decisions and linked to the new review topic.
Artifacts Named the actual typed References/Satisfies/Verifies/Implements choices, with URI/revision/fragment, owner and non-fetching behavior retained.
Text/Git Added explicit include syntax and the risk of reviewing only an entry file when an edit adds a declaration in another file.
Checks/recovery Split engineering completeness into its own topic. Existing raw invalid property/source/behavior/workflow retention, private checkpoint delay, detached Save As and absent recovered Undo stack remain explicit.
Files/shortcuts Added actual Home Continue editing label and source-entry Save As history/Redo preservation, include-name collision and complete-file save guidance.
Hosted/REST Added source-project candidate validation versus conditional commit, warning-only validity and source-location parity; preserved 412/unknown-outcome and no automatic merge semantics.

The new Review semantic changes topic covers trace/direction selection, Previously reviewed versus Current, Keep/Take/final candidate decisions, exact reference pairs, qualified annotation paths, lifeline bindings, row presence versus identity, membership versus order, Reconcile with selected membership, exact history resets and one-transaction apply/Undo. It explicitly identifies the current one-trace dialog; it does not present independent sequential reviews as a multi-source batch.

The new Read model checks topic explains structural rejection versus advisory warnings, structural gating of completeness, responsible leaf work and typed ports, actual included occurrence/reuse evidence, mission-capability links, explicit capability behavior context, group exclusions and the limits of coverage. Its search keywords include the five actual diagnostic codes so pasted codes lead to guidance.

Evidence and verification boundary

Controls and meanings were checked in behavior/topology/view.rs, behavior/scenario.rs, behavior/states.rs, behavior/state_canvas.rs, semantic_review/view.rs, semantic_review/ordered.rs, semantic recovery, home.rs, app/checkpoints.rs, document Save As, forms, core coverage and server routes. The actual typed timed-save result and complete included-file preservation are established separately by release-timed-bundle-oracle-20260906.md; REST candidate/rejection/restart parity by release-rest-coverage-verification-20260906.md.

Ordered review was source/core tested when this guide was written; its current native acceptance was still in progress. This audit does not elevate that evidence to native verification. Prior minimum-window guide acceptance demonstrated the existing Fill/Clear layout fix, not these expanded articles.

Owned-file rustfmt and diff-check pass. No Cargo or native input was performed by this guide worker. The existing guide search/navigation tests should run with the next coordinated desktop check. Native acceptance should search “semantic order”, “work_without_behavior_context” and “time event”, follow related topics and workspace links, and read long walkthroughs at the normal and minimum window sizes. The result must retain the current model and raw drafts and expose all text without Clear-button or scrollbar clipping. These checks remain for the native operator.

Later integration hooks — not current feature claims

Slice that must land first Guide follow-up and acceptance needed
First-class OA CommunicationMean Replace the temporary boundary limitation with the actual participant→participant gesture/form, orientation and actual-performer Interaction allocation rules. Explain legacy records without automatic conversion; include call/reply semantics only after that contract is frozen. Verify the named controls and save/reopen example.
Persisted selective views/categories Expand the views topic with actual Work/Participant category selectors, All versus empty selections, hide/kind/category/context intersections, omitted-edge reasons and labels. Distinguish global behavior coverage from selected Process/Chain/Scenario scope. Explain summary member selection only if the dense-diagram acceptance gate requires and verifies it.
Attached records/artifact reconciliation Extend semantic review with exact attach/remove, owner/endpoint choices, artifact-field decisions and collateral changes once authorable; do not imply that a current mapped reference creates or reparents records.
Multi-source review Replace the explicit one-trace limitation only after a shared final-candidate workflow exists. Teach source selection, competing proposals to the same destination, shared reference mapping, exact per-trace observation decisions and one combined preview/commit. Sequential Take operations are not an adequate substitute.
Source/schema6 freeze Update examples/version language only with real migration/admission tests and source/GUI/API round trips. Keep the current source5 guide accurate until then.

The current docs/scope.md still contains stale unfinished statements about ordered collections and the now-implemented completeness diagnostics. The root was notified; that file is outside this worker's ownership. This is a remaining documentation synchronization item, not a reason to repeat those stale claims in built-in help.