ARCADIA-Studio/harness/release-capella-basic-editing-20260907.md

# Capella 7.1 basic editing benchmark and redesign

The user's rejection of the large horizontal creation buttons supersedes the
previous toolbar proposal. On 2026-09-07 the user explicitly requested consulting
the installed Capella through Computer Use. That renewed authorization follows
the earlier Escape stop; native consultation resumed on a disposable project.

## Environment and method

Actual executable: `C:\Users\User\capella-7.1.0.202606231255-win32-win32-x86_64\capella\capella.exe`.
An isolated workspace `.local/capella-basic-ux-20260907` contains the native-created
project `Basic Editing Benchmark`. The `[OAB] Operational Entities` diagram was
opened from the Operational Analysis workflow. Observations below were made in
the real application at approximately 1010 × 760, using Computer Use. Saved
`.capella` XML independently establishes the semantic result of creation and
allocation. This is separate from the older Capella 7.0 benchmark.

## Observed behavior

* A narrow right palette groups Entities, Roles, Activities, Accelerators and
  Common tools. It collapses completely. Its context menu offers List, Columns,
  Icons Only and Details, plus an independent large-icons option. Icons Only was
  exercised: small glyphs and tooltips expose the contextual tools without a
  horizontal creation row. Some labels in List mode clipped at this width.
* Selecting Operational Entity and clicking the canvas creates `OE 1`. F2 edits
  the label directly on the node; entering `Rescue team` and pressing Enter updates
  the model and Properties. Selection also changes a formatting toolbar and shifts
  Capella's canvas vertically; that jump is a weakness to avoid copying.
* Selecting Operational Activity and clicking inside Rescue team creates a nested
  activity and expands its graphical container. The saved activity remains under
  the root Operational Activity, with a separate ComponentFunctionalAllocation
  from the Entity. Graphical nesting therefore does not imply semantic ownership.
* Saved identities: Entity `3cb395ad-2932-402c-bd51-6a667a45940e`, Activity
  `29a873f8-eb2a-41bd-b48f-3ef6a973027b`, allocation
  `5693d9ff-9a45-4809-b090-edf4b0f01c0e`. Only this disposable model was changed.
* The Manage Activity Allocation tooltip explicitly describes connecting an
  existing activity to a participant and displaying it there. The endpoint action
  itself was not exercised in this pass.
* Interactions and Operational Processes Insert/Remove tooltips explicitly
  distinguish changing graphical presence from deleting semantic objects. These
  tooltips were inspected; no insertion/removal operation is claimed as tested.
* The Entity context menu exposes Transitions and Allocation Management. Create
  System Actor opens a large candidate/result merge preview. Clicking OK closed
  it, but the saved model did **not** contain the proposed SA actor. This establishes
  contextual initiation and preview, not a successful transition. No automatic
  merge behavior is inferred from the dialog closing.

## Alternatives and decision

The rejected horizontal buttons made common commands visible but consumed canvas
height and changed its origin during placement. Context-menu-only creation would
save space but hide the primary workflow. A permanently expanded catalogue would
repeat the global-list problem. Select a compact side palette with explicit label
expansion and contextual menus as a second entry point.

The implementation target is a neutral 36 px palette, explicitly expandable to
156 px. It contains projection-compatible creation and connection tools, compact
vector icons, tooltips and clear selected-tool feedback. Only toggling its width
may resize the canvas; arming a tool, typing, validation, committing and cancelling
must not move the clicked location on screen.

Creation is tool → canvas point → name in place → Enter. Escape cancels without
accepting a placeholder. F2 uses the same in-place editor for Rename. Add existing
offers compatible unplaced objects by default, with explicit broader filtering
and Locate for already placed objects. Contextual Add activity / Add function on
a participant names that performer and creates the work plus allocation in one
undoable transaction. Ordinary root creation remains unallocated. Connections
belong in the same palette, not a second permanent button row.

This adopts Capella's useful interaction and domain distinctions while avoiding
its clipped labels, immediate placeholder commits and large merge dialogue for
ordinary creation. The proposed editor is still an implementation hypothesis;
compactness alone does not establish usability.

## Acceptance still required

Inspect the actual new executable at wide and minimum sizes. Check discovery,
tool state, pointer placement, name focus, Enter/Escape, long names and edge
clamping; canvas origin must remain stable through every drafting phase. Exercise
F2, compatible existing-object insertion, explicit allocated work and connection
creation. Verify exact source/model Save, Undo, Redo and recovery, followed by
ordinary forward/reverse transition destination selection. Preserve the existing
independent oracle; update its interaction instructions, not its expected model
to match whatever the new UI happens to produce.

The prior v1 build passed 584 desktop tests and build/lint gates, but remains a
rejected UI with an observed 22 px placement jump. Its pass does not transfer to
this redesign. The release goal remains active and incomplete.