Dock projection: migrate remaining consumers to the identity reconciler

Context

The architecture challenge on #15136 exposed a shared ownership boundary: repeated dock projection is not a Workstation-private concern. PR #15170 therefore introduces Neo.dashboard.DockProjectionReconciler in src/dashboard/ and proves the same renderer-safe transaction in two consumers with different pane policies: Workstation and Dock Demo B.

A live all-state GitHub duplicate sweep, latest-open queue check, and source-consumer sweep were revalidated on 2026-07-14. No equivalent ticket or landed successor exists. Three direct DockLayoutAdapter consumers still rebuild their projection shell coarsely:

• DemoAWorkspace.refreshDockWorkspace() removes/inserts child 0 while preserving overlay siblings;
• FleetCockpit.refreshDockWorkspace() uses removeAll() around interactive drag/splitter scheduling;
• examples/dashboard/dock/MainContainer.refreshProjection() uses removeAll() as the public example pattern.

The Problem

After #15170, the shared identity-preserving transaction exists, but these three consumers still destroy and recreate surviving tab chrome on every committed dock-document projection. That leaves repository examples and demos teaching two contradictory ownership models:

• surviving logical tab nodes lose their tab.Container, toolbar, body, strip, Overflow, and button identities;
• app-local pane parking or recreation substitutes for shared native reparenting;
• coarse teardown remains entangled with overlay ordering and mid-interaction scheduling;
• new consumers can copy the stale pattern even though the dashboard layer now owns the correct one.

The migration cannot be a blind mechanical replacement. Each consumer has a distinct boundary that must remain intact.

Architectural Reality

• DockLayoutAdapter remains the pure, stateless document-to-config projector.
• DockProjectionReconciler.reconcileProjection(options) owns the staged target shell, separate descendant/ancestor ownership commits, retired-shell cleanup, and final Overflow reprojection.
• Existing live items are discovered from model-owned tab order before the app resolver is consulted; an app resolver creates or retrieves only genuinely absent items.
• Demo A's preview/indicator overlays are persistent siblings outside projection child 0.
• Fleet Cockpit deliberately defers refresh out of splitter/drag handlers because destroying an active interaction owner mid-handler is unsafe.
• The dashboard example is public architectural teaching substrate; its implementation must reflect the normative ownership path.

Intended Solution

1. Migrate Demo A to DockProjectionReconciler using projection child index 0 while preserving its preview and indicator siblings, drag-affordance invalidation, and FLIP bracket.
2. Migrate Fleet Cockpit without weakening its deferred-refresh interaction guard; retain controller/state ownership and prove active splitter/drag handlers are never destroyed mid-callback.
3. Migrate examples/dashboard/dock/MainContainer so the public example demonstrates the shared staged transaction rather than removeAll().
4. Give each consumer a narrow item-placeholder factory and resolver policy; do not duplicate the reconciler's four ownership commits in app code.
5. Remove or rewrite comments that call coarse rebuilding normative after the migration.
6. Keep DockLayoutAdapter stateless and keep app-specific animation, overlay, and controller policies in their consumers.

Contract Ledger

Acceptance Criteria

• Demo A, Fleet Cockpit, and the dashboard dock example call the shared projection reconciler for committed document refresh.
• None of the three uses removeAll() or projection-child remove/insert as its normal refresh path.
• Every surviving logical tabs node keeps its tab.Container, header toolbar, body, strip, Overflow plugin/control, and surviving tab-button component identities.
• Existing live pane identity is preserved; an app resolver is consulted only for an item absent from the current live projection.
• A genuinely removed logical tab node and its retired chrome are destroyed exactly once.
• Demo A's preview and indicator overlay component identities and ordering survive refresh.
• Fleet drag, resize, and deferred-refresh behavior remains green with no use-after-destroy interaction owner.
• The dashboard example's comments describe the reconciler as the normative repeated-projection path.
• Focused unit coverage pins each consumer's special boundary, and whitebox E2E covers Demo A plus Fleet interaction continuity.
• DockLayoutAdapter remains stateless; no second app-local copy of the ownership transaction appears.

Decision Record Impact

None expected. ADR 0029 and learn/agentos/HarnessDockZoneModel.md already place reusable dock projection in src/dashboard/. Escalate only if a consumer falsifies the current reconciler contract rather than extending it through app-local branching.

Out of Scope

Changing the dock-zone document schema · rewriting FLIP geometry · changing tab.Container public APIs · merging app-specific overlay/controller policy into the reconciler · generic keyed-VDOM diffing.

Avoided Traps

• Do not migrate by copying Workstation's animation/menu witnesses into every app.
• Do not remove Fleet's mid-handler scheduling guard merely because shell teardown disappears.
• Do not destroy Demo A's persistent overlay siblings while replacing projection child 0.
• Do not make the adapter stateful.
• Do not claim pane permanence from a cache while leaving tab chrome disposable.

Related

Blocked by #15136 · Related: #15170 · #15099 · #14780

Origin Session ID: adddb25d-fc36-4b08-b9a3-3a62a108cda1

Retrieval Hint: DockProjectionReconciler remaining coarse consumers DemoA FleetCockpit dashboard example removeAll identity handoff

Authored by Emmy (GPT-5.6 Sol Ultra, Codex). Session adddb25d-fc36-4b08-b9a3-3a62a108cda1.