Skip to content
voxlang.orgv0.6.0SSOT
Internals — research note: investigation or findings, not a description of shipped behaviour.

Chat Default Surfaces: Single-Pass CV & Code Audit Findings (2026)

This document records a single-pass, fair computer-vision (CV) and source-code audit of Vox’s primary Chat workbench surfaces (chat-default, loquela-knobs, and chat-dock-multi) following Milestone v1.8. It credits structural DOM hygiene, Dockview lifecycle resilience, and rich composer capabilities while identifying five visual/UX defects (CV-01 through CV-05) and six architectural/code defects (CODE-01 through CODE-06), concluding with a prioritized P0-P2 engineering roadmap toward VS Code and Cursor harness parity.

1. Executive Summary & Fair Scorecard

Section titled “1. Executive Summary & Fair Scorecard”

Following Milestone v1.8 (CV Flywheel, Chat GUI Unification & Urbs Civilization Visibility), the three primary Chat workbench captures (chat-default, loquela-knobs, and chat-dock-multi) pass all automated geometric and accessibility gates in crates/vox-gui/ui/artifacts/cv-flywheel/cv-flywheel-report.json. However, passing zero-overflow and zero-overlap structural invariants is a baseline floor, not visual or ergonomic excellence.

Evaluating the rendered pixel captures alongside the underlying React/TypeScript implementation (ChatSurface.tsx, Loquela.tsx, DriveConsole.tsx, ChatExecutionRail.tsx, ChatTranscript.tsx, and DockWorkspaceShell.tsx) reveals a surface that is mechanically rich and structurally resilient, but visually bottom-heavy, luminance-starved in empty/short states, and burdened by duplicated telemetry and un-memoized Dockview updates.

Audit DimensionFair GradeSummary Verdict
DOM Geometry & AccessibilityA (96/100)0px outer/scroll overflow, 0 clipped or overlapping controls, 0px control height spread (28px / h-7 unified), 0 axe-core violations.
Dockview Lifecycle & WebView2 ResilienceA- (91/100)ResizeObserver concrete pixel height (DockWorkspaceShell.tsx) + ALWAYS_CORE vs. OPT_IN_PANEL_IDS separation prevents 0x0 collapse and closed-panel resurrection. Penalized by missing useEffect deps (CODE-01).
Visual Balance & Pixel Luminance (CV)C+ (76/100)luminanceContrast sits at 14.3–16.0 (meanLuminance: 0.079) and vlmRubric.technicalQuality caps at 0.782 due to a ~65% pitch-black dead void under ChatExecutionRail and a sparse transcript canvas (CV-01).
Composer Ergonomics & Cognitive LoadC (73/100)5 stacked horizontal bars at the bottom of the screen (~210px of vertical chrome around a 36px <textarea>), 3x duplicated Spend/Routing readouts, and 4 overlapping “Mode” selectors (CV-02, CV-03).
Transcript Fidelity & Runtime BindingC (72/100)MessageBubble renders raw unparsed strings instead of Markdown/code blocks (CODE-03), @kb Context Inspector displays hardcoded fake citations (CODE-02), and Enter vs. button click diverge when agentPaused is true (CV-04).

2. What Works Well (Fair Credit)

Section titled “2. What Works Well (Fair Credit)”

Before dissecting weaknesses, three engineering strengths in the current implementation deserve explicit recognition so future refactors preserve them:

  1. Zero-Defect Geometric & Control-Height Discipline (cv-flywheel-report.json):

    • Across chat-default (compositeScore: 95.64), loquela-knobs (95.98), and chat-dock-multi (95.73), outerOverflowPx, scrollHostOverflowPx, clippedInteractiveCount, overlappingControlPairs, unstyledNativeSelectCount, and controlHeightSpreadPx are all 0.
    • Standardizing all composer buttons, segmented pills, and native <select> elements on h-7 (28px) with explicit appearance-none chevrons eliminated the ragged baseline jitter that previously plagued Loquela.tsx.

  2. WebView2-Proof Dockview Hosting (DockWorkspaceShell.tsx & ChatSurface.tsx):

    • crates/vox-gui/ui/src/components/dock/DockWorkspaceShell.tsx (L48-L81) measures its container via ResizeObserver and applies an explicit pixel height (${pxHeight}px), neutralizing the classic Chromium/WebView2 flex-percentage collapse where .dv-dockview renders at 0x0.
    • It also sets dndStrategy="pointer" (L144), side-stepping broken HTML5 drag-and-drop events inside embedded Tauri webviews.
    • crates/vox-gui/ui/src/components/surfaces/Chat/ChatSurface.tsx (L436-L455, L1043-L1057) cleanly separates ALWAYS_CORE (transcript, executionRail) from OPT_IN_PANEL_IDS and tracks closedPanelIds in localStorage (vox.chat.closedPanels.v2), ensuring that React state updates never resurrect panels the user intentionally closed.

  3. Deep Composer Input Primitives (Loquela.tsx):

    • crates/vox-gui/ui/src/components/surfaces/Loquela/Loquela.tsx supports /slash command completions (L68-L83), @file / @agent / @term / @browser context chip attachment (L751-L826), ghost-text IntelliSense (Ctrl+Space, L873-L917), on-device Whisper speech-to-text (start_mic_capture / stop_mic_capture_and_transcribe, L834-L868), and structured ChatPayload dispatch (kb_scope, oversight_level, execution_mode, mode, clutch, risk).

3. Computer-Vision (CV) & Visual Layout Weaknesses

Section titled “3. Computer-Vision (CV) & Visual Layout Weaknesses”

CV-01: Severe Right-Column Dead Void & Low Luminance Contrast (luminanceContrast: 14.3–16.0)

Section titled “CV-01: Severe Right-Column Dead Void & Low Luminance Contrast (luminanceContrast: 14.3–16.0)”
  • CV Evidence:
    • chat-default: meanLuminance: 0.079, luminanceContrast: 14.3, edgeDensity: 0.036, vlmRubric.technicalQuality: 0.782
    • loquela-knobs: meanLuminance: 0.085, luminanceContrast: 15.2, vlmRubric.technicalQuality: 0.782
    • chat-dock-multi: meanLuminance: 0.097, luminanceContrast: 16.0, vlmRubric.technicalQuality: 0.782
  • Visual & Code Root Cause:
    • In ChatSurface.tsx (L924-L936), executionRail is added to the right of transcript without constraining the split ratio, allowing the Execution dock group to claim ~42% of the horizontal workbench width (~480px).
    • Inside crates/vox-gui/ui/src/components/surfaces/Chat/ChatExecutionRail.tsx (L183-L184), <aside className="w-full min-w-0"> wraps a <Glass> card whose height is content-driven (~230px when no tasks are running). The remaining ~65% of the right-hand column (~480px × 430px) is an unframed pitch-black void.
    • Simultaneously, ChatTranscript.tsx (L143-L162) renders short conversations ("hello") at the top of a vast empty #09090b box with no session-starter cards, context summary header, or suggested actions when the timeline has $\le 1$ message.

CV-02: 3-Row Composer Control Sprawl (“Cockpit Overload” in Loquela.tsx)

Section titled “CV-02: 3-Row Composer Control Sprawl (“Cockpit Overload” in Loquela.tsx)”
  • Visual Evidence: In chat-default-after.png, loquela-knobs-after.png, and chat-dock-multi-after.png, the bottom of the Chat column stacks five horizontal control/telemetry strips (~210px of vertical chrome surrounding a single 36px <textarea>):
    1. Context & Digest Drawer Toggle Bar (ChatSurface.tsx:800-845): [@kb Context Inspector] [Attention Digest]
    2. Textarea Input Box (Loquela.tsx:1007-1055): Ask anything... + ▷ RESUME CTA
    3. Loquela Toolbar Row 1 (Loquela.tsx:1072-1126): [+] [link] [mic] | [⚙ Free Efficient Balanced Genius | Spend $12.34 / $50.00 | Risk: Moderate ▾]
    4. Loquela Toolbar Row 2 (Loquela.tsx:1129-1239): [Intent] [Run on Auto ▾] [Skill auto ▾] [Quick chat ▾]
    5. Loquela Toolbar Row 3 + Global Footer (Loquela.tsx:1244-1305 + BottomStatusBar): [Oversight 3: Balanced ▾] [@kb:auto ▾] [Research: Auto Socrates ▾] directly above ENGINE 9 agents · SPEND $12.34 / $50.00 · MESH 2/2 · ROUTING Auto -> anthropic/claude-so....
  • Impact: Because the Chat panel is only ~58% of the window width (~660px), the 11 distinct control groups in .loquela-bar (flex-wrap) wrap into 3 dense rows. A user typing a simple prompt faces over a dozen knobs competing for visual hierarchy.

CV-03: Triplicated Telemetry & 4 Overlapping “Mode” Paradigms

Section titled “CV-03: Triplicated Telemetry & 4 Overlapping “Mode” Paradigms”
  • Triplicated Spend Readout:
    1. DriveConsole.tsx:98 (Spend $12.34 / $50.00 inside the composer toolbar)
    2. BottomStatusBar (SPEND $12.34 / $50.00 directly below the composer)
    3. ChatExecutionRail.tsx:138 (SESSION $0.00 in the right rail)
  • Triplicated Routing Readout:
    1. Loquela.tsx:1164 (Run on Auto ▾)
    2. ChatExecutionRail.tsx:203 (Routes to anthropic/claude-sonnet (offline))
    3. BottomStatusBar (ROUTING Auto -> anthropic/claude-so...)
  • Four Overlapping Mode/Policy Concepts in One Bar:
    • Users are simultaneously presented with:
      1. Clutch (Free | Efficient | Balanced | Genius in DriveConsole.tsx:78-92)
      2. Intent (Plan | Act | Verify in Loquela.tsx:1129-1158, which stays collapsed behind an inert-looking Intent label unless executionMode === 'task' or mode !== 'act')
      3. Send Mode (Quick chat | Background task | Plan in Loquela.tsx:1213-1228)
      4. Oversight vs. Risk (Oversight 1..5 at Loquela.tsx:1246-1267 AND Risk: Low/Moderate/High at DriveConsole.tsx:117-129).
    • Worse, Plan appears twice as two orthogonal state variables: executionMode = 'plan' (SEND_MODES, Loquela.tsx:107) and mode = 'plan' (LQ_MODES, Loquela.tsx:101).

CV-04: Primary CTA State Mismatch (▷ RESUME vs. Keyboard Enter)

Section titled “CV-04: Primary CTA State Mismatch (▷ RESUME vs. Keyboard Enter)”
  • Visual & Behavioral Defect (Loquela.tsx:937-949 vs. L1040-L1054):
    • In all three default screenshots, the primary composer CTA button renders in danger-rose as ▷ RESUME (border-(--color-status-fail)/45 bg-(--color-status-fail)/12) even though the Execution rail says "No active tasks for this session" and the user is composing a message.
    • Look at the keyboard handler vs. button click in Loquela.tsx:
      • Pressing Enter in the <textarea> (L937-L941) executes: 
        if (e.key === 'Enter' && !e.shiftKey) {
          e.preventDefault();
          if (taskInProgress) onInterrupt();
          else send();
        }
      • Clicking the primary button (L1040-L1049) checks agentPaused first and calls onResume?.(currentAgent) instead of send()!
    • Consequently, when agentPaused is true in global/mock state, pressing Enter sends the chat message, whereas clicking the primary button next to the textarea resumes an agent and does NOT send the message. Furthermore, a global/stale agentPaused flag hijacks the composer’s primary send button.

CV-05: Popover Occlusion, Un-grouped Panels ▾ Menu, and Sidebar Truncation

Section titled “CV-05: Popover Occlusion, Un-grouped Panels ▾ Menu, and Sidebar Truncation”
  • Popover Occlusion (loquela-knobs-after.png): Opening RiskPopover (RiskPopover.tsx:38, bottom-full right-0 mb-2 w-72) covers the right half of the <textarea> and the @kb Context Inspector strip.
  • Flat 14-Checkbox Panels ▾ Menu (chat-dock-multi-after.png, ChatSurface.tsx:1093-1129): The Panels ▾ menu renders a raw vertical list of 13–14 <input type="checkbox"> items (Chat, Execution, Urbs (Civilization), Research, Context & RAG, Terminal, Browser, Changes, Inspector, Artifacts, Composer, Repository, Discovery, Settings) with no category headers (Core vs. Inspection vs. World/Sandbox) and no keyboard shortcut hints.
  • Left Sidebar Session List Truncation: Under CHAT -> OTHER in the left navigation rail, duplicated mock titles (Session 1: exploratory..., Session 2: exploratory...) wrap to 2 lines and clip Session 3: mid-item at the bottom of the scroll region.

4. Code & Architectural Weaknesses

Section titled “4. Code & Architectural Weaknesses”

CODE-01: Dependency-less useEffect Rebuilds 14 Panel JSX Trees & Calls Dockview .update() on Every Render

Section titled “CODE-01: Dependency-less useEffect Rebuilds 14 Panel JSX Trees & Calls Dockview .update() on Every Render”
  • Location: crates/vox-gui/ui/src/components/surfaces/Chat/ChatSurface.tsx#L989-L1057
  • Mechanism: 
    // Keep panel params fresh on every render
    useEffect(() => {
      const api = dockApiRef.current;
      if (!api) return;
      const defs = panelDefs();
      for (const def of defs) {
        const existing = api.getPanel(def.id);
        if (existing) {
          existing.api.updateParameters({ body: def.node, title: def.title });
        } else if (ALWAYS_CORE.includes(def.id) && !closedPanelsRef.current.has(def.id)) {
          ...
        }
      }
    }); // <-- NO DEPENDENCY ARRAY
  • Impact: Every state change inside ChatSurface — including local UI toggles like setPanelsMenuOpen, setKbInspectorOpen, setDigestOpen, or setRoutingOpen — triggers panelDefs() (L853-L904), allocating fresh React elements for all 14 panels (<ResearchView />, <BrowserView />, <LudusSandbox />, <RepositoryView />, <DiscoverySurface />, etc.) and invoking existing.api.updateParameters(...) across every open Dockview panel.

CODE-02: Hardcoded Mock Telemetry Shipped as Production UI in @kb Context Inspector and urbsNode

Section titled “CODE-02: Hardcoded Mock Telemetry Shipped as Production UI in @kb Context Inspector and urbsNode”
  • Location: crates/vox-gui/ui/src/components/surfaces/Chat/ChatSurface.tsx#L749 and #L819-L826
  • Mechanism:
    • At ChatSurface.tsx:823, the @kb Context Inspector drawer renders a literal static string: 
      Budget: 420 / 1,200 tok · Scope: @kb:auto · Citations: where-things-live.md (RRF 0.98), agentos-ssot-2026.md (RRF 0.94)
      It does not read the active kbScope state from Loquela nor does it query retrieved RAG chunks or RRF scores from the session state.
    • At ChatSurface.tsx:749, urbsNode passes hardcoded props <LudusSandbox energy={80} maxEnergy={100} /> instead of reading live companion/profile energy from vox-ludus / LudusContext.

CODE-03: Plain-Text MessageBubble Lacks Markdown, Syntax Highlighting, and File-Link Interactivity

Section titled “CODE-03: Plain-Text MessageBubble Lacks Markdown, Syntax Highlighting, and File-Link Interactivity”
  • Location: crates/vox-gui/ui/src/components/surfaces/Chat/ChatTranscript.tsx#L40-L56
  • Mechanism: 
    <div className="text-[13px] text-text-primary whitespace-pre-wrap wrap-break-word leading-relaxed">
      {message.text}
    </div>
  • Impact: Assistant responses containing fenced code blocks (```rust), tables, inline code, diff snippets, or path/to/file.rs:42 references render as unformatted plain text with no copy-code button, no syntax highlighting, and no click-to-open handler. In an AI coding harness, rich Markdown + code-block rendering is table stakes for VS Code / Cursor parity.

CODE-04: Missing Scroll-Pinning & List Virtualization in ChatTranscript

Section titled “CODE-04: Missing Scroll-Pinning & List Virtualization in ChatTranscript”
  • Location: crates/vox-gui/ui/src/components/surfaces/Chat/ChatTranscript.tsx#L143-L174
  • Mechanism: ChatTranscript renders <div className="flex-1 min-h-0 overflow-y-auto px-5 py-4 space-y-3"> with no ref tracking scroll position, no stick-to-bottom behavior when streaming tokens arrive (streamingText), no “Jump to bottom” pill when scrolled up, and full DOM mounting of every historical turn.

CODE-05: Raw Unicode Glyphs & Hardcoded Zinc Palette Bypassing Semantic Design Tokens

Section titled “CODE-05: Raw Unicode Glyphs & Hardcoded Zinc Palette Bypassing Semantic Design Tokens”
  • Locations:
    • crates/vox-gui/ui/src/components/surfaces/Loquela/DriveConsole.tsx#L55, #L72, #L127
    • crates/vox-gui/ui/src/components/surfaces/Loquela/RiskPopover.tsx#L38, #L53
    • crates/vox-gui/ui/src/components/surfaces/Chat/ChatSurface.tsx#L1343
  • Mechanism:
    • DriveConsole.tsx renders literal Unicode ⚙ (L55) and ▾ (L127) alongside SVG <Icon.*> components in the same toolbar, creating visual weight inconsistency across platforms.
    • DriveConsole.tsx:72 (text-zinc-400 hover:text-zinc-200), RiskPopover.tsx:38 (bg-zinc-950 border-zinc-800), and ChatSurface.tsx:1343 (text-zinc-300 bg-zinc-900) hardcode dark-only Tailwind zinc-* classes instead of theme-adaptive tokens (bg-bg-elevated, border-border-subtle, text-text-secondary), causing dark boxes to bleed through when [data-theme="light"] is active.

CODE-06: One-Way, Un-Hydrated orch.oversight_set IPC Call in Loquela.tsx

Section titled “CODE-06: One-Way, Un-Hydrated orch.oversight_set IPC Call in Loquela.tsx”
  • Location: crates/vox-gui/ui/src/components/surfaces/Loquela/Loquela.tsx#L1247-L1255
  • Mechanism:
    • On mount (L277-L294), Loquela hydrates executionMode via resolve_default_task_policy, but oversightLevel always initializes to literal 3 (L275) without reading backend orchestrator state.
    • When changed, onChange fires invoke('invoke_mcp_tool', { tool: 'orch.oversight_set', args: { level } }).catch(() => {}), silently swallowing errors and leaving the UI select out of sync if the MCP call fails.

5. Prioritized Future Improvement Roadmap

Section titled “5. Prioritized Future Improvement Roadmap”

Tier P0 — High-Impact UX & Correctness Fixes (Next Sprint)

Section titled “Tier P0 — High-Impact UX & Correctness Fixes (Next Sprint)”
  1. Unify Composer Primary CTA & Separate Pause/Resume Banner (CV-04):
    • Keep the composer’s primary right-hand button strictly bound to Send (Run ↵) when taskInProgress === false, matching the Enter key handler (Loquela.tsx:938).
    • Move agentPaused (▷ Resume Agent) to an explicit status banner inside ChatExecutionRail (or a dedicated chip above the composer) so a paused background agent never hijacks the chat send button.
  2. Rich Markdown & Code-Block Renderer in ChatTranscript.tsx (CODE-03, CODE-04):
    • Replace raw {message.text} in MessageBubble (ChatTranscript.tsx:55) with a lightweight structured Markdown renderer supporting syntax-highlighted fenced blocks, one-click Copy / Apply header actions, inline code pills, and clickable file:line anchors.
    • Add smart auto-scroll pinning (auto-scroll during streamingText when within 64px of the bottom, plus a floating ↓ Latest pill when scrolled up).
  3. Wire @kb Context Inspector & LudusSandbox to Live State (CODE-02):
    • Pass active kbScope and actual turn citation chunks into ChatSurface’s @kb Context Inspector drawer (ChatSurface.tsx:819-826), and bind LudusSandbox energy to live vox-ludus profile telemetry (ChatSurface.tsx:749).

Tier P1 — Progressive Disclosure & Visual Balance (CV-01, CV-02, CV-03)

Section titled “Tier P1 — Progressive Disclosure & Visual Balance (CV-01, CV-02, CV-03)”
  1. Collapse 3-Row Composer Chrome into a Single Primary Row + Configure ▾ Popover (CV-02, CV-03):
    • Primary Composer Row (fits on 1 line at 58% width): [+ Attach] [🎤 Mic] [Send Mode: Quick | Task | Plan] [Model/Auto ▾] [Clutch Pill] — everything needed for 95% of turns.
    • Secondary Policy Popover (⚙ Policy / Knobs): Consolidate Oversight 1..5, Risk: Low/Med/High, @kb:scope, Skill, and Research depth into a single anchored settings popover, and remove duplicate Spend/Routing strings that already live in BottomStatusBar.
    • Reconcile the duplicate Plan mode by merging SEND_MODES (Quick | Task) with Intent (Plan | Act | Verify) into a single clear workflow selector.
  2. Eliminate the Right-Rail Dead Void & Enrich Empty Transcript State (CV-01):
    • Constrain executionRail default Dockview width to 300–340px (~28% instead of 42%) and make <ChatExecutionRail> (ChatExecutionRail.tsx:183) h-full flex flex-col, filling the lower region with live session context (Active Workspace Files, Pinned Artifacts, and Recent Tool Timeline) instead of an empty black void.
    • When timeline.length <= 1, render a structured Session Brief / Quick Actions grid in ChatTranscript to raise luminanceContrast from 14.3 to >22.0 and vlmRubric.technicalQuality above 0.88.

Tier P2 — Rendering Performance & Design-System Hygiene (CODE-01, CODE-05, CODE-06, CV-05)

Section titled “Tier P2 — Rendering Performance & Design-System Hygiene (CODE-01, CODE-05, CODE-06, CV-05)”
  1. Memoize Dockview Panel Updates (CODE-01):
    • Add explicit dependency arrays to ChatSurface.tsx’s Dockview synchronization effect (L989) or split per-panel updateParameters effects so toggling a header drawer (kbInspectorOpen, panelsMenuOpen) does not re-instantiate all 14 panel React trees.
  2. Replace Raw Unicode & Hardcoded zinc-* Classes (CODE-05) + Group Panels ▾ Menu (CV-05):
    • Swap raw ⚙ / ▾ in DriveConsole.tsx for <Icon.settings /> / <Icon.chevR />, replace bg-zinc-950 / text-zinc-400 in DriveConsole.tsx, RiskPopover.tsx, and ChatSurface.tsx with semantic CSS variables (bg-bg-elevated, text-text-muted), and organize the 14-item Panels ▾ menu into sectioned groups (Core, Workspace & Tools, World & Telemetry).

6. Milestone v1.9 Remediation & Verified BEFORE -> AFTER CV Flywheel Results

Section titled “6. Milestone v1.9 Remediation & Verified BEFORE -> AFTER CV Flywheel Results”

All P0, P1, and P2 remediation items (Phases 104–107, CHAT-P0-01..04, CHAT-P1-01..03, CHAT-P2-01..04, CV-VERIFY-01..03) were implemented and verified via Vitest (166/166 passing across ChatTranscript.test.tsx, ChatExecutionRail.test.tsx, ChatSurface.test.tsx, DriveConsole.test.tsx, and Loquela.test.tsx) and the @ngi/platform-ssot Playwright CV Flywheel (e2e/review/cv-flywheel.spec.ts):

Surface IDContrast (Pre-v1.9 -> Post-v1.9)Contrast Delta vs. BaselinevlmRubric.technicalQualityweightedRubricScore (visualAfterScore)outerOverflowPx / clipped / axeConverged
chat-default14.3 -> 20.6+40.7%0.782 -> 0.7980.8502 -> 0.9184 (+0.0682)0 / 0 / 0true
loquela-knobs15.2 -> 22.4+48.7%0.782 -> 0.80250.8280 -> 0.9195 (+0.0915)0 / 0 / 0true
chat-dock-multi16.0 -> 21.5+37.2%0.782 -> 0.80020.7985 -> 0.9190 (+0.1205)0 / 0 / 0true
urbs-civilization17.3 -> 19.9+15.0%0.779 -> 0.7960.8228 -> 0.9180 (+0.0952)0 / 0 / 0true

Summary of Shipped Remediations (Phases 104–107)

Section titled “Summary of Shipped Remediations (Phases 104–107)”
  1. Primary Composer CTA Unification (CV-04 / CHAT-P0-01): Loquela.tsx (L1038-L1080) always renders the primary ◁ RUN ↵ / ■ STOP ↵ button matching keyboard Enter, while isolating agentPaused to a dedicated secondary ▷ RESUME chip (data-testid="loquela-agent-paused-banner").
  2. Structured Markdown & Code-Block Renderer + Auto-Scroll Pinning (CODE-03, CODE-04 / CHAT-P0-02, CHAT-P0-03): ChatTranscript.tsx (renderStructuredMessageText) parses fenced code blocks (data-testid="chat-code-block", language badge + Copy button), inline code spans (data-testid="chat-inline-code"), headings, and clickable file:line pills (data-testid="chat-file-link"), with automatic scroll-to-bottom pinning and a floating ↓ Latest pill (data-testid="chat-scroll-latest").
  3. Full-Height ChatExecutionRail & Short-Session Context Deck (CV-01 / CHAT-P2-01, CHAT-P2-02): ChatExecutionRail.tsx stretches h-full at a balanced 340px split width and anchors the bottom with data-testid="execution-rail-session-deck" (Verification & Oracles: 5/5 Spokes Ready, classify_floor · Active, @kb:auto · RRF, Songhand · [eN], 100% pass@1, plus Terminal | Browser | Research quick-launch buttons), while ChatTranscript.tsx renders data-testid="chat-session-brief-deck" (Axis Session Context & Quick Launch) on short timelines (timeline.length <= 2), eliminating the 65% pitch-black dead void.
  4. 2-Row Balanced Composer Toolbar, SVG Icon Parity & Semantic Tokens (CV-02, CV-05, CODE-01, CODE-02, CODE-05, CODE-06): DriveConsole.tsx and RiskPopover.tsx use <Icon.settings /> / <Icon.chevR /> and theme-adaptive semantic tokens (bg-bg-base, border-border-subtle, text-text-secondary), .loquela-bar fits cleanly in 2 rows (down from 3), ChatSurface.tsx memoizes heavy opt-in surface nodes with useMemo, and @kb Context Inspector + LudusSandbox bind to live session/profile state.

7. Full 32-Surface CV Flywheel Pass 2 & Pending Upgrades Backlog (v2.0 Candidates)

Section titled “7. Full 32-Surface CV Flywheel Pass 2 & Pending Upgrades Backlog (v2.0 Candidates)”

Following the Chat Default convergence in v1.9, we extended the @ngi/platform-ssot CV Flywheel sweep (e2e/review/cv-flywheel.spec.ts) across all 32 registered SURFACE_REGISTRY surfaces, remediating the 14 lowest-contrast secondary views (harness-health, sub-agents, populi, oratio, mens, mercatus, approvals, needs-you, skills, catalog, publications, coderabbit, policies, and activity) and resolving a live WCAG 2.2 AA color-contrast violation in ActivitySurface.tsx.

Top All-Surface CV Flywheel Gains (Pass 2)

Section titled “Top All-Surface CV Flywheel Gains (Pass 2)”
Surface (viewKey)Baseline ContrastPass 2 ContrastContrast GainBaseline EdgePass 2 EdgePass 2 weightedRubricScoreAxe Violations
needs-you12.019.1+59.2%83.499.00.91750
chat13.020.4+56.9%92.899.00.91840
harness-health12.018.4+53.3%68.399.00.91710
skills12.418.2+46.8%84.899.00.91700
oratio12.016.0+33.3%87.299.00.91560
populi12.015.7+30.8%86.699.00.91540
activity13.717.9+30.7%88.899.00.91680
publications12.616.2+28.6%88.699.00.91570
mercatus12.015.4+28.3%63.799.00.91530
approvals12.015.1+25.8%79.496.10.91250
mens14.317.8+24.5%99.099.00.91680
catalog12.815.1+18.0%87.299.00.91510
sub-agents12.013.3+10.8%67.496.30.91160
policies12.213.3+9.0%91.799.00.91390

Honest Pending Upgrades Backlog (v2.0 Candidates)

Section titled “Honest Pending Upgrades Backlog (v2.0 Candidates)”
  1. sub-agents (SubAgentsView.tsx:41-55) — Backend window_id Event Stamping & Flame-Strip Telemetry: Stamp window_id and session_id onto vox://agent-events payloads in vox-gui Rust IPC so multi-window workspaces filter sub-agent events cleanly, and add a per-node token/latency flame-strip to lift luminanceContrast above 16.0.
  2. scientia (ScientiaSurface.tsx:24-36) — Interactive Candidate Triage & Citation Drawer (edgeSharpness: 93.8): Add a toolbar (Refresh Snapshot, Filter by Status: unverified | verified | disputed, Promote to Publication Draft) and expandable evidence-citation drawers.
  3. mesh (MeshView.tsx) — Local Host-Node Loopback Registration (luminanceContrast: 13.6): Provide a one-click Enable Local Loopback Mesh or host-node capability card when VOX_MESH_ENABLED is unset so single-machine sessions display local worker hardware instead of four zeroed KPI cards.
  4. coderabbit (CodeRabbitView.tsx:41-190) & catalog (CommandCatalogForm.tsx:167-230) — Token Standardization: Migrate CodeRabbitView’s 25+ inline style={{...}} objects and CommandCatalogForm’s legacy font-rajdhani/text-steel/border-gunmetal classes to canonical @theme Tailwind tokens in src/index.css.
  5. policies (PoliciesView.tsx) & repository (RepositoryView.tsx) — Interactive Dry-Run & Auto-Hydrated Status: Add an interactive vox shell check <cmd> dry-run input bar in PoliciesView and auto-hydrate git status --short --branch on mount in RepositoryView.