Align judge bundle provenance with optional deploy/publish evidence

apps/demo-frontend/public/app.js

@@ -10030,6 +10030,9 @@ function bindEvents() {
 }
 
 async function bootstrap() {
+  const initialTabId = readTabIdFromHash() ?? readStoredTabId();
+  setActiveTab(initialTabId, { syncHash: false });
+
   const runtimeConfig = await loadRuntimeConfig();
   if (runtimeConfig?.wsUrl) {
     el.wsUrl.value = runtimeConfig.wsUrl;
@@ -10065,7 +10068,7 @@ async function bootstrap() {
   resetOperatorBoardView({ mode: readStoredOperatorBoardMode(), persistMode: false });
   renderTaskList();
   evaluateConstraints();
-  setActiveTab(readStoredTabId());
+  setActiveTab(readStoredTabId(), { syncHash: false });
   setUiTaskFieldsVisibility();
   initBackgroundVideoLoopBlend();
   enhanceSelectControls();

apps/demo-frontend/public/styles.css

@@ -738,9 +738,9 @@ textarea {
 }
 
 select {
-  appearance: none;
-  -webkit-appearance: none;
-  -moz-appearance: none;
+  appearance: none !important;
+  -webkit-appearance: none !important;
+  -moz-appearance: none !important;
   cursor: pointer;
   padding-right: 42px;
   --select-surface-start: var(--surface-control);
@@ -751,6 +751,10 @@ select {
     linear-gradient(180deg, var(--select-surface-start), var(--select-surface-end));
 }
 
+select::-ms-expand {
+  display: none;
+}
+
 select:hover {
   --select-surface-start: var(--surface-control-hover);
 }
@@ -1750,6 +1754,7 @@ textarea {
   min-height: 74px;
   padding-top: 11px;
   padding-bottom: 11px;
+  overflow: visible;
 }
 
 .panel-live-connection .action-group-primary > .export-menu {
@@ -1759,11 +1764,11 @@ textarea {
 }
 
 .panel-live-connection .export-menu[open] {
-  z-index: 240;
+  z-index: 520;
 }
 
 .panel-live-connection .export-menu-list {
-  z-index: 260;
+  z-index: 540;
 }
 
 .action-group {
@@ -1788,6 +1793,12 @@ textarea {
 .export-menu {
   position: relative;
   min-width: 0;
+  z-index: 140;
+  isolation: isolate;
+}
+
+.export-menu[open] {
+  z-index: 420;
 }
 
 .export-menu > summary {
@@ -1890,7 +1901,7 @@ textarea {
   position: absolute;
   right: 0;
   top: calc(100% + 8px);
-  z-index: 120;
+  z-index: 460;
   min-width: min(320px, calc(100vw - 56px));
   padding: 8px;
   display: grid;
@@ -2305,7 +2316,7 @@ button:focus-visible {
 
 .meta-row-status-live {
   margin-top: 14px;
-  padding: 8px;
+  padding: 10px;
   border: 1px solid color-mix(in oklch, var(--primary) 20%, var(--border-soft));
   border-radius: calc(var(--radius) - 4px);
   background:
@@ -2318,14 +2329,14 @@ button:focus-visible {
   display: grid;
   grid-template-columns: repeat(auto-fit, minmax(162px, 1fr));
   grid-template-columns: repeat(4, minmax(0, 1fr));
-  gap: 8px;
+  gap: 10px;
   overflow: visible;
 }
 
 .meta-row-status-live > div {
   min-width: 0;
-  min-height: 46px;
-  padding: 7px 10px;
+  min-height: 48px;
+  padding: 8px 11px;
   border-radius: calc(var(--radius) - 10px);
   display: inline-flex;
   align-items: center;
@@ -2401,18 +2412,18 @@ button:focus-visible {
 }
 
 .meta-row-status-live strong {
-  color: color-mix(in oklch, white 98%, var(--foreground));
-  font-size: 0.74rem;
+  color: color-mix(in oklch, white 99.6%, var(--foreground));
+  font-size: 0.76rem;
   text-transform: none;
-  letter-spacing: 0.015em;
+  letter-spacing: 0.012em;
   white-space: nowrap;
   text-shadow: 0 1px 0 color-mix(in oklch, black 30%, transparent);
 }
 
 .meta-row-status-live > div > span:not(.status-pill) {
-  color: color-mix(in oklch, white 99.5%, var(--foreground));
-  font-size: 0.82rem;
-  font-weight: 650;
+  color: color-mix(in oklch, white 99.8%, var(--foreground));
+  font-size: 0.84rem;
+  font-weight: 670;
   line-height: 1.28;
 }
 
@@ -2435,7 +2446,7 @@ button:focus-visible {
   max-width: 100%;
   padding: 4px 9px;
   border-radius: 999px;
-  border: 1px solid color-mix(in oklch, var(--primary) 32%, var(--border-soft));
+  border: 1px solid color-mix(in oklch, var(--primary) 38%, var(--border-soft));
   background:
     radial-gradient(160px 64px at 12% -34%, color-mix(in oklch, var(--primary) 10%, transparent), transparent 72%),
     linear-gradient(
@@ -2453,7 +2464,7 @@ button:focus-visible {
   word-break: break-word;
   box-shadow:
     inset 0 1px 0 color-mix(in oklch, var(--foreground) 10%, transparent),
-    0 0 0 1px color-mix(in oklch, var(--primary) 16%, transparent);
+    0 0 0 1px color-mix(in oklch, var(--primary) 20%, transparent);
 }
 
 .meta-row > div {

docs/judge-quickstart.md

@@ -6,7 +6,7 @@ Fast, judge-facing entry point for a 5-10 minute evaluation run.
 
 This project covers all three challenge categories in one platform:
 
-1. Live Agent (realtime speech, interruption, translation, negotiation)
+1. Live Agent (realtime speech, interruption, translation, negotiation, grounded research)
 2. Creative Storyteller (text + audio + image + video narrative flow)
 3. UI Navigator (computer-use style UI planning/execution with approval guardrails)
 
@@ -55,6 +55,8 @@ Artifacts:
 8. `artifacts/judge-visual-evidence/presentation.md`
 9. `artifacts/demo-e2e/epic-summary.json`
 
+If deploy/publish artifacts are present, `manifest.md` and `presentation.md` also surface compact deploy/publish provenance from `artifacts/deploy/railway-deploy-summary.json` and `artifacts/deploy/repo-publish-summary.json`. Ordinary local judge flows omit that section, and raw deploy/publish JSON is not embedded into the judge-facing markdown.
+
 ## 3) Validate Release Readiness
 
 ```bash
@@ -64,21 +66,30 @@ npm run verify:release
 ## 4) What Judges Should See in UI
 
 1. Connection + assistant lifecycle (`idle/streaming/speaking`).
-2. Live interruption, truncate/delete evidence, and gateway error correlation.
+2. Live interruption, truncate/delete evidence, gateway error correlation, and optional `research` citations/source URLs.
 3. Operator Console panels:
    - Live Bridge Status
    - Approvals Queue
+   - Workflow Runtime / Runtime Guardrails
    - Device Nodes Health / Updates
+   - Bootstrap Doctor / Browser Workers
    - Governance Policy Lifecycle
    - Skills Registry Lifecycle
    - Plugin Marketplace Lifecycle
    - Agent Usage Evidence
    - Cost & Tokens Evidence
-4. Session export controls:
+4. Operator support panels:
+   - `Runtime Drill Runner` for repo-owned dry-run/live recovery drills and `followUpContext` handoff.
+   - `Workflow Control Panel` for redacted assistive-router/runtime override posture.
+   - `Operator Session Ops` for saved `operatorPurpose`, session replay, and cross-agent discovery.
+   - `Bootstrap Doctor & Auth Profiles` for provider/auth-profile/device/fallback posture.
+   - `Browser Worker Control` for queue/checkpoint posture on long-running UI jobs.
+5. Session export controls:
    - `Export Session -> Export Markdown`
    - `Export Session -> Export JSON`
    - `Export Session -> Export Audio (WAV)`
-5. Story Timeline panel:
+   - Confirm exported Markdown/JSON include `runtimeGuardrailsSignalPaths`, `operatorPurpose`, `operatorSessionReplay`, and `operatorDiscovery`.
+6. Story Timeline panel:
    - Confirm `Timeline State` KPI transitions (`0%` idle -> ready/pending) as story output arrives.
    - Segment scrubber/selector reflects `output.story.timeline`
    - Preview card shows segment text + `image/video/audio` refs
@@ -100,15 +111,18 @@ npm run verify:release
    - Start mic, send live request, then trigger interruption.
    - Show truncate/delete/gateway-correlation evidence in Operator Console.
    - Mention roundtrip and interrupt KPI lanes in `artifacts/demo-e2e/badge-details.json`.
+   - If judges ask for grounded-research proof, switch to `intent=research` once and show citation-bearing `answer`, `citations`, and `sourceUrls`.
 3. `02:15-03:30` Creative Storyteller category:
    - Send storyteller prompt.
    - Open `Story Timeline` panel and scrub segments.
    - Show image/video/audio refs and async media behavior.
 4. `03:30-04:45` UI Navigator category:
    - Send `ui_task` intent with grounding fields.
    - Show approval flow and damage-control verdict in Operator Console.
+   - Save a short purpose in `Operator Session Ops`, then open `Bootstrap Doctor & Auth Profiles` and `Browser Worker Control` once to show runtime posture before execution.
    - Confirm safety gates before execution.
 5. `04:45-05:30` Evidence close:
    - Run `npm run demo:epic` (or fallback `npm run demo:e2e:visual:judge` if e2e/policy/badge were already executed).
    - Open `artifacts/judge-visual-evidence/presentation.md`.
    - Confirm all evidence lanes are `pass` in `artifacts/demo-e2e/badge-details.json`.
+   - Export session `JSON` or `Markdown` and confirm `runtimeGuardrailsSignalPaths`, `operatorPurpose`, `operatorSessionReplay`, and `operatorDiscovery`.

docs/judge-visual-evidence.md

@@ -5,8 +5,9 @@
 Create one reproducible visual bundle for judges:
 
 1. Screenshot checklist status (present/missing).
-2. Critical badge-evidence lane status (`pass/fail/unavailable`).
+2. Critical badge-evidence lane status (`pass/fail/unavailable`), including runtime guardrails and provider provenance.
 3. Single manifest for quick go/no-go before submission.
+4. One-page presentation bundle with runtime guardrails snapshot, provider adapter snapshot, and compact deploy/publish provenance when optional Railway/repo-publish artifacts are available.
 
 ## Commands
 
@@ -76,7 +77,9 @@ Defaults used by `scripts/judge-visual-evidence-pack.mjs`:
 
 1. Badge details: `artifacts/demo-e2e/badge-details.json`
 2. Demo summary: `artifacts/demo-e2e/summary.json`
-3. Screenshot directory: `artifacts/judge-visual-evidence/screenshots`
+3. Optional Railway deploy summary: `artifacts/deploy/railway-deploy-summary.json`
+4. Optional repo publish summary: `artifacts/deploy/repo-publish-summary.json`
+5. Screenshot directory: `artifacts/judge-visual-evidence/screenshots`
 
 Defaults used by `scripts/judge-visual-capture.mjs`:
 
@@ -94,6 +97,8 @@ Defaults used by `scripts/judge-visual-capture.mjs`:
 5. `artifacts/judge-visual-evidence/presentation.md`
 6. `artifacts/demo-e2e/epic-summary.json`
 
+`manifest.md` and `presentation.md` surface compact deploy/publish provenance from `railway-deploy-summary.json` / `repo-publish-summary.json` when those optional files are present. Ordinary local judge flows omit that section instead of filling the page with `unavailable` placeholders, and raw deploy/publish JSON is not embedded into the judge-facing markdown.
+
 ## Required Screenshot Filenames
 
 Put files into `artifacts/judge-visual-evidence/screenshots`:
@@ -120,4 +125,6 @@ Pack marks these as critical:
 6. `pluginMarketplace`
 7. `deviceNodes`
 8. `agentUsage`
-9. `deviceNodeUpdates` (derived from `deviceNodes` updates fields)
+9. `runtimeGuardrailsSignalPaths`
+10. `providerUsage`
+11. `deviceNodeUpdates` (derived from `deviceNodes` updates fields)