V2 UI: qualify one shared luminous scene runtime against Brain 01–05 and Loom 01–02

[ScriptedAlchemy]

Scope and branch safety

Implementation brief for the accepted V2 concepts, not a replacement design. Any implementation PR must target codex/tracedecay-total-redesign-plan-reopened. Keep #707 DRAFT; no merge or undraft is authorized.

Reviewed source snapshots: product dd002654eb9ccc277eda0f9f4fe87ef1e6f0a660; primary lookbook/demo 87e14dcd1026e63390233f56c795accee81b351e. Recheck HEAD and signatures before editing. This review verified files, image blob identities, briefs, and implementation code. PNG bytes could not be inspected in the review environment; no pixel-parity, browser, GPU, performance, or passing-test claim is made.

Exact visual authorities

Primary repository: ScriptedAlchemy/td-brain-demo at the reviewed commit:

• lookbook/pngs/01-brain/final/01-registry-overview.png
• lookbook/pngs/01-brain/final/02-project-hover.png
• lookbook/pngs/01-brain/final/03-repository-zoom.png
• lookbook/pngs/01-brain/final/04-project-scoped.png
• lookbook/pngs/01-brain/final/05-admitted-activity-synapse.png
• lookbook/pngs/03-loom/final/01-follow-loaded-tail.png
• lookbook/pngs/03-loom/final/02-temporal-replay.png

Brain same-stem briefs are under lookbook/briefs/01-brain/final/. The product mirror places plates and briefs together under mockups/ui-concept-v2/{01-brain,03-loom}/final/; the product mirror contains the two Loom briefs missing from the demo export.

Implementation authority: lookbook/IMPLEMENTATION.md, especially Layer model, Brain final-state implementation map, Proof-of-capability renderer spike, and Shared scene migration. DESIGN-SYSTEM.md and NAVIGATION.md retain their existing authority. Preserve luminous night-glass depth, cyan signal, restrained amber admitted activity, and hairline structure; do not replace the concept with a card grid or screenshot background.

Verified baseline — reuse rather than rebuild

BrainPage.tsx already composes generated registry data, typed read boundaries, measured recency-X/indexed-mass-Y placement, an accessible registry rail, project scoping, and admitted activity. RegistryFieldView passes nodes, edges, extent, activation, and selection into GraphCanvas.

renderer.ts actually constructs Sigma and already implements hover isolation, heat-driven reducers, glow ordering, and renderer lifecycle. This is not a missing-WebGL or missing-glow bug. The gap is qualification of the richer body/curve/depth/semantic-zoom language against the accepted plates.

The demo's src/brain/BrainPage.tsx switches between BrainCanvas and mountVoxelo, with concept/fixture state and separate exploratory modes. Its material and composition work may inform the spike; its page-level runtime mounting and fixture selection are not a production data adapter.

Smallest implementation

1. Run an isolated, read-only renderer spike, not a fourteen-page rewrite. Use the existing production shell and generated source models. Adapt already-served registry/temporal evidence into a small renderer-neutral scene input. Do not introduce a second router, handwritten wire DTO, or alternative store of truth.
2. Keep two distinct layout modes: Brain's measured registry/topology coordinates and Loom's deterministic time-X/hierarchy-Y coordinates. Share rendering/picking primitives, not a force layout that changes their meaning.
3. Compare the existing Sigma/SVG baseline with the React-19-compatible R3F v9-family/Three candidate and PixiJS v8 alternative required by IMPLEMENTATION.md. Verify exact package compatibility in the spike. Candidates may coexist there only; production gets one winning scene runtime, not parallel permanent implementations.
4. Keep navigation, scope, controls, labels, inspectors, exact tables, focus, and URLs in React/DOM. Let the GPU render luminous bodies and evidenced curved paths. Use screen-space Canvas/SVG for crisp axes and selection; use a revision-tagged worker only for layout/density work. The renderer and worker do not assign evidence grades, fetch authority data, or create activity.
5. Demonstrate Brain overview, independent hover/focus, repository semantic zoom, scope transitions, and a real admitted pulse with an exact DOM evidence pivot. Missing authority stays visibly unavailable; do not fill gaps with the sample plate values. Use the same primitives on a representative Loom journey before choosing a renderer.

Preserve existing Brain behavior

• Retain the current 4,200 ms heat half-life and one drawn relation hop. An event touching one checkout must not illuminate sibling checkouts. The conflicting multi-hop wording in the Brain 05 brief must be reconciled separately, not used as permission to expand propagation.
• Preserve suppression of historical pulses already in the ring when the field mounts, stable topology memoization, exact drawn-identity checks, and the existing admission/scope owner.
• Hover/focus/selection/loading/connectivity never strike activity. Repository hubs do not set project scope or masquerade as holdings-sized projects.
• Preserve source freshness and typed unavailable/partial/empty/stale states. No timeout increase, cached-as-fresh substitution, or weakened readiness gate to make a capture look populated.
• Keep the canonical fourteen entries in dashboard/src/app/channels.ts unchanged. No new Tools, Integrations, Policies, or renderer-demo rail.

Acceptance and deliverables

• Checked-in comparison report identifies source revisions, observed times, missing-source reasons, exact dependency versions, browser/device, and capture paths. Concept/synthetic/load-generated inputs remain labelled and cannot qualify a real-evidence claim.
• Review actual captures for the seven named plates' corresponding runtime states. No rasterized concept panels in the shipping bundle. Do not claim parity from a successful compile or prose-only comparison.
• Measure the specified 5,000- and 20,000-event workloads, 100+ agents, nested branches, gaps, and representative real PR ci(hotpath): enable pull_request profiling with availability guards #743 evidence. Record layout/load, picking, memory/GPU resources, idle cost, and active playback cost; set budgets from supported hardware rather than inventing results or thresholds.
• Stable source IDs resolve identically from pointer picking and keyboard DOM fallback; obsolete worker revisions cannot replace current geometry or selection.
• Cover 1440/768/320 CSS-pixel layouts, 200% browser zoom, reduced motion, forced colors, context loss, initialization failure, and worker failure. Static/fallback modes preserve source identity, activity meaning, grades, and navigation.
• Extend the existing Brain/graph propagation and DOM tests at the actual integration boundary; report commands and results. Do not replace behavioral tests with source-text assertions.

The output of this issue is a measured renderer decision and bounded proof-of-capability. Page migrations follow that decision; a speculative dependency installation or additional demo-only polish does not complete it.