Make the Brain the Chat's living background; float components on legibility-floored Glass #56

Acceptance Criteria

• On the Landing and in-session, the Brain fills the entire Chat behind every component; the Rail, titlebar, and Panels stay solid. Switching the titlebar session tab swaps the Brain to that session's graph.
• Every chat component (bubbles, prose, composer, code, diffs, run-plots) floats on Glass; there is exactly one Brain render loop. The inline timeline strip is gone (its expand/collapse toggle obviated — no node interactivity existed); the timeline keeps its existing working-indicator shimmer (cycling gerund + live elapsed).
• Contrast, standard tint: body text clears ≥4.5:1 with a small safety margin over the reference frame — a synthetic Brain at peak bloom whose sampled backdrop is the brightest palette value the Brain actually paints in that theme (peak #fff676 on dark; peak #8f8000 on light — the light Brain never paints #fff676, so validating against it there would derive a near-transparent tint that silently fails over the real light frame) — in each shipped chat theme (light + dark ship day one). Measured against that fixture, not eyeballed.
• Contrast, dense tint: code/diff text clears ≥4.5:1 and graphical marks (syntax colors, diff fills, plot lines) clear ≥3:1 (WCAG 1.4.11) over the reference frame, each shipped chat theme.
• Theme switch: changing the chat theme mid-session re-keys the Glass to that theme's pinned tint set and re-satisfies both contrast criteria. The tint follows the chat theme (default oc-2), not the app-shell theme.
• The Brain is data-true (nodes light up on real file/tool/vault activity) with an adaptive heartbeat: lively when working, ~8fps breathing at rest, hard-pause when hidden or under prefers-reduced-motion. It densifies in place within the fixed viewport, bounded by a node cap / recency window so a long thread grows neither the graph nor per-frame cost without limit.
• At rest the Brain shows a sparse seed; a fresh session opens nearly empty and grows only from that session's activity (no cross-session persistence).
• Perf gate: on the reference laptop VS Code webview (battery-powered, pinned before the gate runs), p95 frame time ≤ 16.7 ms with zero frames > 33 ms, over a ≥10 s continuous scroll of a long thread (≥40 cards) at full Brain tempo, in both themes. Verified by manual sign-off on that machine; a headless-Chromium rAF trace is a pre-gate proxy only. When the gate can't hold, the motion-ease fallback engages (see Key Decisions) and blur fidelity is preserved.

Key Decisions

• Blur vs. tint split: blur is high + constant (calm, cheap); the tint carries contrast, opacity pinned to clear the AA floor with a small safety margin over the reference frame. Blur alone never guarantees legibility.
• Two tints, chat-theme-keyed: standard (prose, bubbles, composer) and dense (code, diffs, run-plots, more opaque). The set follows the chat's own theme (default oc-2), independent of the app-shell theme.
• Perf priority (pre-agreed, non-negotiable mid-build): protect the frosted look. A frame-time monitor eases Brain motion when p95 stays over budget for a sustained window (with hysteresis before restoring, to avoid oscillation); the terminal valve — if easing motion to a full stop still misses budget — is a hard-pause to a static blurred field. Blur fidelity is never degraded; full blur + full motion always at rest.
• Build by reuse on a fresh branch (kate/chat-brain-background off local/amicode): reuse the native engine + token generator + mount wiring, fresh design layer, no iframe.
• Fixed backdrop, bounded growth: the Brain is viewport-anchored (cards scroll over a stationary Brain); the session graph densifies in place, bounded by a node cap / recency window.

Constraints & Invariants

• The standard tint cannot be made ultra-transparent — its opacity is bounded below by the contrast floor (the exact failure the earlier build hit); muted/secondary text rides the dense tint or a locally-dimmed zone.
• Amicode design-system law still applies to the chrome (single yellow, black-on-yellow / yellow-on-dark, geometry tokens). The Brain graph itself is exempt (it may use its categorical palette; #fff676 reserved for live thought).
• Browser-first iteration (vite dev on packages/app), not VSIX rebuilds.
• npm run lint clean + app typechecks before the PR.

Risks

• Perf (primary): high-radius backdrop-blur recomputed per frame over an animating canvas, across many cards, while scrolling, is the expensive combination and exceeds what the earlier build measured (moderate blur). Mitigation: the pre-agreed motion-ease fallback + a real laptop-webview gate (headless Chromium is only a proxy).
• Legibility regression: derived tints must be re-validated against the reference frame in each shipped chat theme; a too-transparent tint silently fails AA. Mitigation: contrast is a tested acceptance criterion, not a visual judgment.

Source

• Durable record: ADR packages/app/docs/adr/0002-brain-as-chat-background.md (this issue implements it); shell vocabulary in packages/app/CONTEXT.md (Brain, Glass).
• Supersedes the design direction of the unpushed kate/chat-brain-atmosphere branch (per-bubble tiers + skeleton atlas).

Notes

Large enough to decompose into vertical slices (mount → glass tokens → float components → motion/at-rest → perf gate → lint/PR) via break-into-subissues once published.