Skip to content
TruthVouch Docs

Dashboard & Health Score

The Shield dashboard is your command center for hallucination detection. It shows at a glance what AI engines are saying about you and where inaccuracies exist.

Shield Dashboard showing Health Score and AI accuracy metrics

Health Score

The Health Score is a 0-100 aggregate measure of AI accuracy about your organization.

What It Measures

Health Score = weighted average of all cross-checks against your Truth Nuggets across all monitored AI engines.

Calculation:

  • Each cross-check gets a 0-100 truth score (how well AI response matches your truth)
  • Scores are weighted by fact importance (Financial facts weighted higher than other categories)
  • Average across all engines and facts = your Health Score

Examples:

  • 95+ = Excellent. AI engines are highly accurate about you.
  • 85-94 = Good. Minor discrepancies; your main narrative is correct.
  • 70-84 = Fair. Some hallucinations; brand risk emerging.
  • 60-69 = Poor. Significant inaccuracies; urgent action needed.
  • <60 = Critical. Major hallucinations; deploy corrections immediately.

Trend Tracking

Health Score trends over time:

  • Green arrow (↑): Score improving. Corrections are working.
  • Gray line (→): Score stable. Hallucinations are consistent.
  • Red arrow (↓): Score declining. New hallucinations emerging; investigate.

Example: “Health Score was 82 on March 1, now 87 on March 14 (↑ 5 points). Corrections deployed 2 weeks ago are taking effect.”

Dashboard Widgets

1. Health Score Card (Top Left)

  • Large, prominent display
  • Current score (0-100)
  • Trend arrow + points gained/lost since last week
  • Click to see score history (last 30 days)

2. Recent Alerts (Top Center/Right)

  • Last 5-10 alerts
  • Each shows: severity, AI engine, fact name, time
  • Color coded: Red (critical), Orange (high), Yellow (medium), Gray (low)
  • Click any alert to review full details

3. Top Issues (Center Left)

  • Hallucinations cited most frequently across all AI engines
  • Ranked by impact (critical → low)
  • Shows: issue name, count, affected engines
  • Example: “Product pricing incorrect (8 citations, ChatGPT/Claude/Perplexity)“

4. Engine Comparison (Center Right)

  • How accurate each monitored engine is about you
  • Health score for ChatGPT, Claude, Gemini, etc., side-by-side
  • Identify which engines are most problematic
  • Example: ChatGPT (88), Claude (92), Gemini (76), Perplexity (85)

5. Correction Status (Bottom Left)

  • How many corrections are pending, deployed, verified
  • Breakdown: “8 deployed, 5 verified, 2 in progress”
  • Timeline: average time from detection to correction
  • Example: “Average correction time: 14 hours”

6. Trends & History (Bottom Right)

  • Health Score trend line (30-day view)
  • Annotation markers for major events:
    • “Correction deployed” (point on line)
    • “New product launched” (new fact added)
    • “Truth Nugget updated” (baseline shift)

Filtering & Views

Filter by Category

Click category tabs to see scores only for specific Truth Nugget categories:

  • Company
  • Products
  • Leadership
  • Financial
  • Technical
  • Legal
  • Competitive

Example: “Zoom into Product facts to see Health Score just for your product information (89).”

Filter by Engine

Click engine badges to focus on specific AI engines:

  • Just ChatGPT: 91
  • Just Claude: 93
  • Just Gemini: 82

See which engines are your weakest performers.

Filter by Severity

  • Critical (>80)
  • High (60-80)
  • Medium (40-60)
  • Low (<40)
  • Resolved

Focus on issues that matter most.

Time Range

  • Last 24 hours
  • Last 7 days
  • Last 30 days
  • Custom range

See how things have evolved over time.

Detailed Views

Alert Details

Click any alert to see:

  • Full AI Response: Exact text the engine generated
  • Your Truth: What you said vs what AI said (side-by-side)
  • Entity Extraction: Which specific entities (dates, names, numbers) conflict
  • Confidence: How confident NLI is about the mismatch (85%+)
  • Suggested Correction: Auto-generated fix
  • Source: URL or document being referenced

Engine Deep Dive

Click any engine in the comparison to see:

  • Overall accuracy score for that engine
  • Category breakdown (which categories is it accurate/inaccurate on)
  • All alerts specific to that engine
  • Correction history (which corrections improved accuracy for this engine)

Example: “Gemini: Overall 82. Accurate on Products (92), poor on Financial (71).”

Trend Analysis

Click the trend chart to see:

  • 90-day Health Score history (line chart)
  • Annotation overlay: when corrections deployed, facts added
  • Correlation: “Health Score improved 8 points after deploying corrections to products category”
  • Seasonality: “Every Monday AI responses are 3 points lower (fresh training data?)”

Actions from Dashboard

Approve Corrections

When Shield suggests a correction:

  1. Click the alert
  2. Review suggested correction
  3. Click Approve Correction
  4. Choose method: Neural Fact Sheet (default) or Direct Feedback
  5. System confirms: “Deploying correction…”

Correction deploys within seconds; verification happens 24-72 hours later.

Edit Truth Nuggets

When you spot a false positive (your nugget is wrong):

  1. Click the alert
  2. Review your Truth Nugget text
  3. Click Edit Nugget
  4. Update the fact text, confidence, or source
  5. Save
  6. Shield re-scores immediately; alert may resolve

Dismiss Alerts

For false positives you don’t want to fix (AI interpretation is valid even if not exactly matching your truth):

  1. Click the alert
  2. Click Dismiss
  3. Choose reason: “Paraphrase OK”, “Outdated nugget”, “Marketing hyperbole”
  4. System learns from dismissals

Schedule Follow-Up

For complex issues needing team discussion:

  1. Click the alert
  2. Click Assign to Team Member
  3. Choose team member and due date
  4. They get notified

Mobile View

On mobile (or narrow desktop), the dashboard adapts:

  • Health Score stays prominent at top
  • Other widgets stack vertically
  • Filters move to collapsible menu
  • Swipe between tabs for different views

Full functionality preserved — you can manage alerts from your phone.

Customization

Reorder Widgets

Drag and drop widgets to arrange dashboard your way:

  1. Hover over widget header
  2. Drag to new position
  3. Saved automatically

Hide/Show Widgets

Click the eye icon on any widget to hide it:

  • Hide low-relevance widgets
  • Focus on what matters to you
  • Your settings saved per account

Set Refresh Rate

Auto-refresh every: 1 min, 5 min, 15 min, 1 hour, never

  • Real-time monitoring: 1 min (updates every 60 sec)
  • Standard: 15 min
  • Low-priority: 1 hour

Save Custom Views

Save filtered dashboards for different purposes:

  • “CEO Morning Check” (Health Score + Top Issues only)
  • “Marketing Deep Dive” (Products category, top issues, engine comparison)
  • “Compliance Review” (Legal + Financial, last 90 days)

Click Save as View, give it a name, share URL with team.

Exporting Data

Export Dashboard Snapshot

Click Export → PDF:

  • Clean, printable dashboard view
  • Timestamp and data accuracy note
  • Share with executives or auditors

Export Raw Data

Click Export → CSV to get:

  • All alerts (raw data)
  • All cross-checks (queries, responses, scores)
  • Correction history (all deployed fixes)

For analysis in Excel, Tableau, or other tools.

Real-Time Monitoring (Professional+)

Enterprise teams can:

  • Pin dashboard to TV screen for real-time monitoring
  • Set up Slack notifications for new alerts
  • Stream Health Score via API to custom dashboards

Setup in Settings → Integrations.

Performance Tips

For Slow Networks

  • Increase refresh rate (1 hour instead of 1 min)
  • Hide trend chart (computationally heavy)
  • Filter to 1-2 engines instead of all

For Large Knowledge Bases (100+ Nuggets)

  • Filter by category (don’t view all at once)
  • Focus on top issues (sort by impact)
  • Use time range filters (last 7 days vs 90 days)

Next Steps